Download data/Basic2/MSDOS/basic.c from introvoyz041/tinybasic: direct link, hf CLI and curl.
- Browser
- Download file 224 kB
-
https://huggingface.co/datasets/introvoyz041/tinybasic/resolve/main/data/Basic2/MSDOS/basic.c
- Command line
-
hf download hf://datasets/introvoyz041/tinybasic/data/Basic2/MSDOS/basic.c
-
curl -L -o basic.c https://huggingface.co/datasets/introvoyz041/tinybasic/resolve/main/data/Basic2/MSDOS/basic.c
224 kB
| /*---------------------------------------------------------------- | |
| * Please read this before compiling: | |
| * - Review hardware.h for settings specific hardware settings. | |
| * Super important on Arduino and Raspberry PI. | |
| * - language.h controls the language features. | |
| * For Arduino Integer BASIC is default and a limited language | |
| * set. For the larger boards this can be extended a lot. | |
| *----------------------------------------------------------------- | |
| * | |
| * $Id: basic.c,v 1.5 2024/03/02 15:38:20 stefan Exp stefan $ | |
| * | |
| * Stefan's IoT BASIC interpreter - BASIC for everywhere. | |
| * | |
| * See the licence file on | |
| * https: | |
| * (GNU GENERAL PUBLIC LICENSE, Version 3, 29 June 2007) | |
| * | |
| * Author: Stefan Lenz, sl001@serverfabrik.de | |
| * | |
| * Currently there are two versions of the runtime environment. | |
| * One contains all platforms compiled in the Arduino IDE | |
| * (ESP8266, ESP32, AVR, MEGAAVR, SAM*, RP2040) | |
| * | |
| * Anothers contains all platforms compiled in gcc with a POSIX OS | |
| * (Mac, Raspberry, Windows/MINGW) plus rudimentary MSDOS with tc2.0. The | |
| * latter will be removed soon. | |
| * | |
| * The interface to BASIC is identical. | |
| */ | |
| /* the runtime environment */ | |
| /* | |
| * the core basic language headers | |
| */ | |
| /* use long jump for error handling */ | |
| /* Global BASIC definitions */ | |
| /* | |
| * All BASIC keywords for the tokens, PROGMEM on Arduino | |
| * Normal memory elsewhere. | |
| */ | |
| const char sge[] PROGMEM = "=>"; | |
| const char sle[] PROGMEM = "<="; | |
| const char sne[] PROGMEM = "<>"; | |
| /* Palo Alto language set */ | |
| const char sprint[] PROGMEM = "PRINT"; | |
| const char slet[] PROGMEM = "LET"; | |
| const char sinput[] PROGMEM = "INPUT"; | |
| const char sgoto[] PROGMEM = "GOTO"; | |
| const char sgosub[] PROGMEM = "GOSUB"; | |
| const char sreturn[] PROGMEM = "RETURN"; | |
| const char sif[] PROGMEM = "IF"; | |
| const char sfor[] PROGMEM = "FOR"; | |
| const char sto[] PROGMEM = "TO"; | |
| const char sstep[] PROGMEM = "STEP"; | |
| const char snext[] PROGMEM = "NEXT"; | |
| const char sstop[] PROGMEM = "STOP"; | |
| const char slist[] PROGMEM = "LIST"; | |
| const char snew[] PROGMEM = "NEW"; | |
| const char srun[] PROGMEM = "RUN"; | |
| const char sabs[] PROGMEM = "ABS"; | |
| const char srnd[] PROGMEM = "RND"; | |
| const char ssize[] PROGMEM = "SIZE"; | |
| const char srem[] PROGMEM = "REM"; | |
| /* Apple 1 language set */ | |
| const char snot[] PROGMEM = "NOT"; | |
| const char sand[] PROGMEM = "AND"; | |
| const char sor[] PROGMEM = "OR"; | |
| const char slen[] PROGMEM = "LEN"; | |
| const char ssgn[] PROGMEM = "SGN"; | |
| const char speek[] PROGMEM = "PEEK"; | |
| const char sdim[] PROGMEM = "DIM"; | |
| const char sclr[] PROGMEM = "CLR"; | |
| const char shimem[] PROGMEM = "HIMEM"; | |
| const char stab[] PROGMEM = "TAB"; | |
| const char sthen[] PROGMEM = "THEN"; | |
| const char sbend[] PROGMEM = "END"; | |
| const char spoke[] PROGMEM = "POKE"; | |
| /* Stefan's basic additions */ | |
| const char scont[] PROGMEM = "CONT"; | |
| const char ssqr[] PROGMEM = "SQR"; | |
| const char spow[] PROGMEM = "POW"; | |
| const char smap[] PROGMEM = "MAP"; | |
| const char sdump[] PROGMEM = "DUMP"; | |
| const char sbreak[] PROGMEM = "BREAK"; | |
| /* LOAD and SAVE is always there */ | |
| const char ssave[] PROGMEM = "SAVE"; | |
| const char sload[] PROGMEM = "LOAD"; | |
| const char sget[] PROGMEM = "GET"; | |
| const char sput[] PROGMEM = "PUT"; | |
| const char sset[] PROGMEM = "SET"; | |
| const char scls[] PROGMEM = "CLS"; | |
| const char slocate[] PROGMEM = "LOCATE"; | |
| const char selse[] PROGMEM = "ELSE"; | |
| /* Arduino functions */ | |
| const char spinm[] PROGMEM = "PINM"; | |
| const char sdwrite[] PROGMEM = "DWRITE"; | |
| const char sdread[] PROGMEM = "DREAD"; | |
| const char sawrite[] PROGMEM = "AWRITE"; | |
| const char saread[] PROGMEM = "AREAD"; | |
| const char sdelay[] PROGMEM = "DELAY"; | |
| const char smillis[] PROGMEM = "MILLIS"; | |
| const char sazero[] PROGMEM = "AZERO"; | |
| const char sled[] PROGMEM = "LED"; | |
| const char stone[] PROGMEM = "PLAY"; | |
| const char spulse[] PROGMEM = "PULSE"; | |
| /* DOS functions */ | |
| const char scatalog[] PROGMEM = "CATALOG"; | |
| const char sdelete[] PROGMEM = "DELETE"; | |
| const char sfopen[] PROGMEM = "OPEN"; | |
| const char sfclose[] PROGMEM = "CLOSE"; | |
| const char sfdisk[] PROGMEM = "FDISK"; | |
| /* low level access functions */ | |
| const char susr[] PROGMEM = "USR"; | |
| const char scall[] PROGMEM = "CALL"; | |
| /* mathematics */ | |
| const char ssin[] PROGMEM = "SIN"; | |
| const char scos[] PROGMEM = "COS"; | |
| const char stan[] PROGMEM = "TAN"; | |
| const char satan[] PROGMEM = "ATAN"; | |
| const char slog[] PROGMEM = "LOG"; | |
| const char sexp[] PROGMEM = "EXP"; | |
| /* INT is always needed to make float/int programs compatible */ | |
| const char sint[] PROGMEM = "INT"; | |
| /* elemetars graphics */ | |
| const char scolor[] PROGMEM = "COLOR"; | |
| const char splot[] PROGMEM = "PLOT"; | |
| const char sline[] PROGMEM = "LINE"; | |
| const char scircle[] PROGMEM = "CIRCLE"; | |
| const char srect[] PROGMEM = "RECT"; | |
| const char sfcircle[] PROGMEM = "FCIRCLE"; | |
| const char sfrect[] PROGMEM = "FRECT"; | |
| /* Dartmouth BASIC extensions */ | |
| const char sdata[] PROGMEM = "DATA"; | |
| const char sread[] PROGMEM = "READ"; | |
| const char srestore[] PROGMEM = "RESTORE"; | |
| const char sdef[] PROGMEM = "DEF"; | |
| const char sfn[] PROGMEM = "FN"; | |
| const char son[] PROGMEM = "ON"; | |
| /* The Darkarts commands unthinkable in Dartmouth */ | |
| const char smalloc[] PROGMEM = "MALLOC"; | |
| const char sfind[] PROGMEM = "FIND"; | |
| const char seval[] PROGMEM = "EVAL"; | |
| /* complex error handling */ | |
| const char serror[] PROGMEM = "ERROR"; | |
| /* iot extensions */ | |
| const char savail[] PROGMEM = "AVAIL"; | |
| const char sstr[] PROGMEM = "STR"; | |
| const char sinstr[] PROGMEM = "INSTR"; | |
| const char sval[] PROGMEM = "VAL"; | |
| const char snetstat[] PROGMEM = "NETSTAT"; | |
| const char ssensor[] PROGMEM = "SENSOR"; | |
| const char swire[] PROGMEM = "WIRE"; | |
| const char ssleep[] PROGMEM = "SLEEP"; | |
| /* events and interrupts */ | |
| const char safter[] PROGMEM = "AFTER"; | |
| const char severy[] PROGMEM = "EVERY"; | |
| const char sevent[] PROGMEM = "EVENT"; | |
| const char swhile[] PROGMEM = "WHILE"; | |
| const char swend[] PROGMEM = "WEND"; | |
| const char srepeat[] PROGMEM = "REPEAT"; | |
| const char suntil[] PROGMEM = "UNTIL"; | |
| const char sswitch[] PROGMEM = "SWITCH"; | |
| const char scase[] PROGMEM = "CASE"; | |
| const char sswend[] PROGMEM = "SWEND"; | |
| const char sdo[] PROGMEM = "DO"; | |
| const char sdend[] PROGMEM = "DEND"; | |
| const char sfend[] PROGMEM = "FEND"; | |
| const char sasc[] PROGMEM = "ASC"; | |
| const char schr[] PROGMEM = "CHR"; | |
| const char sright[] PROGMEM = "RIGHT"; | |
| const char sleft[] PROGMEM = "LEFT"; | |
| const char smid[] PROGMEM = "MID"; | |
| const char sspc[] PROGMEM = "SPC"; | |
| const char sedit[] PROGMEM = "EDIT"; | |
| /* zero terminated keyword storage */ | |
| const char* const keyword[] PROGMEM = { | |
| sge, sle, sne, sprint, slet, sinput, | |
| sgoto, sgosub, sreturn, sif, sfor, sto, | |
| sstep, snext, sstop, slist, snew, srun, | |
| sabs, srnd, ssize, srem, | |
| snot, sand, sor, slen, ssgn, speek, sdim, | |
| sclr, shimem, stab, sthen, | |
| sbend, spoke, | |
| scont, ssqr, spow, smap, sdump, sbreak, | |
| ssave, sload, | |
| sget, sput, sset, scls, slocate, selse, | |
| spinm, sdwrite, sdread, sawrite, saread, | |
| sdelay, smillis, sazero, sled, | |
| stone, | |
| spulse, | |
| scatalog, sdelete, sfopen, sfclose, sfdisk, | |
| susr, scall, | |
| ssin, scos, stan, satan, slog, sexp, | |
| sint, | |
| scolor, splot, sline, scircle, srect, | |
| sfcircle, sfrect, | |
| sdata, sread, srestore, sdef, sfn, son, | |
| smalloc, sfind, seval, | |
| /* complex error handling */ | |
| serror, | |
| savail, sstr, sinstr, sval, | |
| snetstat, ssensor, swire, ssleep, | |
| safter, severy, | |
| sevent, | |
| swhile, swend, srepeat, suntil, sswitch, scase, sswend, | |
| sdo, sdend, | |
| sfend, | |
| sasc, schr, sright, sleft, smid, sspc, | |
| sedit, | |
| 0 | |
| }; | |
| /* the zero terminated token dictonary needed for scalability */ | |
| const token_t tokens[] PROGMEM = { | |
| GREATEREQUAL, LESSEREQUAL, NOTEQUAL, TPRINT, TLET, | |
| TINPUT, TGOTO, TGOSUB, TRETURN, TIF, TFOR, TTO, TSTEP, | |
| TNEXT, TSTOP, TLIST, TNEW, TRUN, TABS, TRND, TSIZE, TREM, | |
| TNOT, TAND, TOR, TLEN, TSGN, TPEEK, TDIM, TCLR, | |
| THIMEM, TTAB, TTHEN, TEND, TPOKE, | |
| TCONT, TSQR, TPOW, TMAP, TDUMP, TBREAK, | |
| TSAVE, TLOAD, | |
| TGET, TPUT, TSET, TCLS, TLOCATE, TELSE, | |
| TPINM, TDWRITE, TDREAD, TAWRITE, TAREAD, TDELAY, TMILLIS, | |
| TAZERO, TLED, | |
| TTONE, | |
| TPULSE, | |
| TCATALOG, TDELETE, TOPEN, TCLOSE, TFDISK, | |
| TUSR, TCALL, | |
| TSIN, TCOS, TTAN, TATAN, TLOG, TEXP, | |
| TINT, | |
| TCOLOR, TPLOT, TLINE, TCIRCLE, TRECT, | |
| TFCIRCLE, TFRECT, | |
| TDATA, TREAD, TRESTORE, TDEF, TFN, TON, | |
| TMALLOC, TFIND, TEVAL, | |
| TERROR, | |
| TAVAIL, TSTR, TINSTR, TVAL, TNETSTAT, | |
| TSENSOR, TWIRE, TSLEEP, | |
| TAFTER, TEVERY, | |
| TEVENT, | |
| TWHILE, TWEND, TREPEAT, TUNTIL, TSWITCH, TCASE, TSWEND, | |
| TDO, TDEND, | |
| TFEND, | |
| TASC, TCHR, TRIGHT, TLEFT, TMID, TSPC, | |
| TEDIT, | |
| 0 | |
| }; | |
| /* experimental, do not use right now */ | |
| const bworkfunction_t workfunctions[] PROGMEM = { | |
| 0, 0, 0, xprint, 0 | |
| }; | |
| /* errors and messages */ | |
| const char mfile[] PROGMEM = "file.bas"; | |
| const char mprompt[] PROGMEM = "> "; | |
| const char mgreet[] PROGMEM = "Stefan's Basic 2.0alpha"; | |
| const char mline[] PROGMEM = "LINE"; | |
| const char mnumber[] PROGMEM = "NUMBER"; | |
| const char mvariable[] PROGMEM = "VARIABLE"; | |
| const char marray[] PROGMEM = "ARRAY"; | |
| const char mstring[] PROGMEM = "STRING"; | |
| const char mstringv[] PROGMEM = "STRINGVAR"; | |
| const char egeneral[] PROGMEM = "Error"; | |
| const char eunknown[] PROGMEM = "Syntax"; | |
| const char enumber[] PROGMEM = "Number"; | |
| const char edivide[] PROGMEM = "Div by 0"; | |
| const char eline[] PROGMEM = "Unknown Line"; | |
| const char emem[] PROGMEM = "Memory"; | |
| const char estack[] PROGMEM = "Stack"; | |
| const char erange[] PROGMEM = "Range"; | |
| const char estring[] PROGMEM = "String"; | |
| const char evariable[] PROGMEM = "Variable"; | |
| const char eloop[] PROGMEM = "Loop"; | |
| const char efile[] PROGMEM = "File"; | |
| const char efun[] PROGMEM = "Function"; | |
| const char eargs[] PROGMEM = "Args"; | |
| const char eeeprom[] PROGMEM = "EEPROM"; | |
| const char esdcard[] PROGMEM = "SD card"; | |
| const char* const message[] PROGMEM = { | |
| mfile, mprompt, mgreet, | |
| mline, mnumber, mvariable, marray, | |
| mstring, mstringv, | |
| egeneral | |
| , eunknown, enumber, edivide, eline, | |
| emem, estack, erange, | |
| estring, evariable, eloop, efile, efun, eargs, | |
| eeeprom, esdcard | |
| }; | |
| /* | |
| * maxnum: the maximum accurate(!) integer of a | |
| * 32 bit float | |
| * strindexsize: in the new code this is simply the size of the | |
| * stringlength type. Currently only 1 byte and 2 bytes are tested. | |
| */ | |
| const number_t maxnum=16777216; | |
| const number_t maxnum=(number_t)~((number_t)1<<(sizeof(number_t)*8-1)); | |
| const int numsize=sizeof(number_t); | |
| const int addrsize=sizeof(address_t); | |
| const int eheadersize=sizeof(address_t)+1; | |
| const int strindexsize=sizeof(stringlength_t); /* default in the meantime, strings up to unsigned 16 bit length */ | |
| const address_t maxaddr=(address_t)(~0); | |
| /* | |
| * The basic interpreter is implemented as a stack machine | |
| * with global variable for the interpreter state, the memory | |
| * and the arithmetic during run time. | |
| */ | |
| /* the stack, all BASIC arithmetic is done here */ | |
| accu_t stack[STACKSIZE]; | |
| address_t sp=0; | |
| /* a small buffer to process string arguments, mostly used for Arduino PROGMEM and string functions */ | |
| /* use with care as it is used in some string functions */ | |
| char sbuffer[SBUFSIZE]; | |
| /* the input buffer, the lexer can tokenize this and run from it, bi is an index to this. | |
| bi must be global as it is the program cursor in interactive mode */ | |
| char ibuffer[BUFSIZE] = "\0"; | |
| char *bi; | |
| /* a static array of variables A-Z for the small systems that have no heap */ | |
| number_t vars[VARSIZE]; | |
| /* the BASIC working memory, either malloced or allocated as a global array */ | |
| mem_t mem[MEMSIZE]; | |
| mem_t* mem; | |
| address_t himem, memsize; | |
| /* reimplementation of the loops, will replace the forstack */ | |
| bloop_t loopstack[FORDEPTH]; | |
| index_t loopsp = 0; | |
| /* the GOSUB stack remembers an address to jump to */ | |
| address_t gosubstack[GOSUBDEPTH]; | |
| index_t gosubsp = 0; | |
| /* arithmetic accumulators */ | |
| number_t x, y; | |
| /* the name of on object, replaced xc and xy in BASIC 1 */ | |
| name_t name; | |
| /* an address accumulator, used a lot in string operations */ | |
| address_t ax; | |
| /* a string index registers, new style identifying a string either in C memory or BASIC memory */ | |
| string_t sr; | |
| /* the active token */ | |
| token_t token; | |
| /* the curent error, can be a token, hence token type */ | |
| token_t er; | |
| /* the jmp buffer for the error handling */ | |
| jmp_buf sthook; | |
| /* a trapable error */ | |
| mem_t ert; | |
| /* the interpreter state, interactive, run or run from EEPROM */ | |
| mem_t st; | |
| /* the current program location */ | |
| address_t here; | |
| /* the topmost byte of a program in memory, beginning of free BASIC RAM */ | |
| address_t top; | |
| /* used to format output with # */ | |
| mem_t form = 0; | |
| /* do we use the Microsoft convention of an array starting at 0 or 1 like Apple 1 | |
| two seperate variables because arraylimit can be changed at runtime for existing arrays | |
| msarraylimit says if an array should be created with n or n+1 elements */ | |
| mem_t msarraylimits = 1; | |
| address_t arraylimit = 0; | |
| mem_t msarraylimits = 0; | |
| address_t arraylimit = 1; | |
| /* behaviour around boolean, needed to change the interpreters personality at runtime */ | |
| /* -1 is microsoft true while 1 is Apple 1 and C style true. */ | |
| mem_t booleanmode = BOOLEANMODE; | |
| /* setting the interpreter to integer at runtime */ | |
| mem_t forceint = 0; | |
| /* the default size of a string now as a variable */ | |
| stringlength_t defaultstrdim = STRSIZEDEF; | |
| /* the base of the random number generator | |
| * 0 is Apple 1 style RND from 0 to n-epsilon | |
| * 1 is Palo Alto style from 1 to n | |
| */ | |
| mem_t randombase = 0; | |
| /* is substring logic used or not */ | |
| mem_t substringmode = 0; | |
| mem_t substringmode = 1; | |
| /* the flag for true MS tabs */ | |
| mem_t reltab = 0; | |
| /* the flag for lower case names */ | |
| mem_t lowercasenames = 0; | |
| /* the number of arguments parsed from a command */ | |
| mem_t args; | |
| /* the random number seed, this is unsigned */ | |
| address_t rd; | |
| unsigned long rd; | |
| /* the RUN debuglevel */ | |
| mem_t debuglevel = 0; | |
| /* DATA pointer, where is the current READ statement */ | |
| address_t data = 0; | |
| address_t datarc = 1; | |
| /* | |
| * process command line arguments in the POSIX world | |
| * bnointafterrun is a flag to remember if called as command | |
| * line argument, in this case we don't return to interactive | |
| */ | |
| int bargc; | |
| char** bargv; | |
| mem_t bnointafterrun = 0; | |
| /* formaters lastouttoken and spaceafterkeyword to make a nice LIST */ | |
| mem_t lastouttoken; | |
| mem_t spaceafterkeyword; | |
| mem_t outliteral = 0; | |
| mem_t lexliteral = 0; | |
| /* | |
| * The cache for the heap search - helps the string code. | |
| * The last found object on the heap is remembered. This is needed | |
| * because the string code sometime searches the heap twice during the | |
| * same operation. Also, bfind is used to remember the length of the | |
| * last found object. | |
| */ | |
| heap_t bfind_object; | |
| /* | |
| * a variable for string to numerical conversion, | |
| * telling you were the number ended. | |
| */ | |
| address_t vlength; | |
| /* the timer code - very simple needs to be converted to to a struct */ | |
| /* timer type */ | |
| btimer_t after_timer = {0, 0, 0, 0, 0}; | |
| btimer_t every_timer = {0, 0, 0, 0, 0}; | |
| /* the event code */ | |
| /* the event list */ | |
| int nevents = 0; | |
| int ievent = 0; | |
| mem_t events_enabled = 1; | |
| volatile bevent_t eventlist[EVENTLISTSIZE]; | |
| /* the extension of the GOSUB stack */ | |
| mem_t gosubarg[GOSUBDEPTH]; | |
| /* the error handler type, very simple for now */ | |
| typedef struct { | |
| mem_t type; | |
| address_t linenumber; | |
| } berrorh_t; | |
| berrorh_t berrorh = {0 , 0}; | |
| mem_t erh = 0; | |
| /* the string for real time clocks */ | |
| char rtcstring[20] = { 0 }; | |
| /* the units pulse operates on, in microseconds*/ | |
| address_t bpulseunit = 10; | |
| /* only needed for POSIXNONBLOCKING */ | |
| mem_t breakcondition = 0; | |
| /* the FN context, how deep are we in a nested function call, negative values reserved */ | |
| int fncontext = 0; | |
| /* | |
| * BASIC timer stuff, this is a core interpreter function now | |
| */ | |
| /* the millis function for BASIC */ | |
| void bmillis() { | |
| number_t m; | |
| /* millis is processed as integer and is cyclic mod maxnumber and not cast to float!! */ | |
| m=(number_t) (millis()/(unsigned long)pop() % (unsigned long)maxnum); | |
| push(m); | |
| } | |
| /* | |
| * Determine the possible basic memory size. | |
| * using malloc causes some overhead which can be relevant on the smaller | |
| * boards. | |
| * | |
| * Set MEMSIZE instead to a static value. In this case ballocmem | |
| * just returns the static MEMSIZE. | |
| * | |
| * If SPIRAMINTERFACE is defined, we use the memory from a serial RAM and dont | |
| * allocate it here at all. | |
| * | |
| */ | |
| address_t ballocmem() { | |
| /* on most platforms we know the free memory for BASIC, this comes from runtime */ | |
| long m=freememorysize(); | |
| /* we subtract some language feature depended things, this is only needed on | |
| small Arduino boards with memories below 16kb */ | |
| if (m < 16000) { | |
| m-=64; /* strings cost memory */ | |
| m-=160; /* odd but true on Ardunio UNO and the like */ | |
| m-=96; | |
| m-=256; | |
| } | |
| /* and keep fingers crossed here */ | |
| if (m<0) m=128; | |
| /* we allocate as much as address_t can handle */ | |
| if (m>maxaddr) m=maxaddr; | |
| /* try to allocate the memory */ | |
| mem=(mem_t*)malloc(m); | |
| if (mem != 0) return m-1; | |
| /* fallback if allocation failed, 128 bytes */ | |
| mem=(mem_t*)malloc(128); | |
| if (mem != 0) return 128; else return 0; | |
| } | |
| address_t ballocmem(){ return MEMSIZE-1; }; | |
| /* | |
| * Layer 0 function - variable handling. | |
| * | |
| * These function access variables and data | |
| */ | |
| /* | |
| * Eeprom load / save / autorun functions | |
| * needed for SAVE and LOAD to an EEPROM | |
| * autorun is generic. | |
| * | |
| * The eeprom is accessed through the runtime functions | |
| * elength(), eupdate(), ewrite() and eflush(); | |
| * | |
| */ | |
| /* save a file to EEPROM, disabled if we use the EEPROM directly */ | |
| void esave() { | |
| address_t a=0; | |
| /* does the program fit into the eeprom */ | |
| if (top+eheadersize < elength()) { | |
| /* EEPROM per default is 255, 0 indicates that there is a program */ | |
| eupdate(a++, 0); | |
| /* store the size of the program in byte 1,2,... of the EEPROM*/ | |
| setaddress(a, eupdate, top); | |
| a+=addrsize; | |
| /* store the program */ | |
| while (a < top+eheadersize){ | |
| eupdate(a, memread2(a-eheadersize)); | |
| a++; | |
| } | |
| eupdate(a++,0); | |
| /* needed on I2C EEPROM and other platforms where we buffer */ | |
| eflush(); | |
| } else { | |
| error(EOUTOFMEMORY); | |
| er=0; | |
| } | |
| } | |
| /* load a file from EEPROM, disabled if the use the EEPROM directly */ | |
| void eload() { | |
| address_t a=0; | |
| /* have we stored a program? */ | |
| if (elength()>0 && (eread(a) == 0 || eread(a) == 1)) { | |
| /* how long is it? */ | |
| a++; | |
| top=getaddress(a, eread); | |
| a+=addrsize; | |
| /* load it to memory, memwrite2 is direct mem access */ | |
| while (a < top+eheadersize){ | |
| memwrite2(a-eheadersize, eread(a)); | |
| a++; | |
| } | |
| } else { | |
| /* no valid program data is stored */ | |
| error(EEEPROM); | |
| } | |
| } | |
| /* autorun something from EEPROM or a filesystem */ | |
| char autorun() { | |
| /* autorun from EEPROM if there is an EEPROM flagged for autorun */ | |
| if (elength()>0 && eread(0) == 1) { /* autorun from the EEPROM */ | |
| top=getaddress(1, eread); | |
| st=SERUN; | |
| return 1; /* EEPROM autorun overrules filesystem autorun */ | |
| } | |
| /* autorun from a given command line argument, if we have one */ | |
| if (bargc > 0 && ifileopen(bargv[1])) { | |
| xload(bargv[1]); | |
| st=SRUN; | |
| ifileclose(); | |
| bnointafterrun=TERMINATEAFTERRUN; | |
| return 1; | |
| } | |
| /* on a platform with a file system, autoexec from a file */ | |
| if (ifileopen("autoexec.bas")) { | |
| xload("autoexec.bas"); | |
| st=SRUN; | |
| ifileclose(); | |
| return 1; | |
| } | |
| /* nothing to autorun */ | |
| return 0; | |
| } | |
| /* | |
| * The new malloc code. Heap structure is now | |
| * payload | |
| * (payload size) | |
| * name | |
| * type | |
| * | |
| * In the new heap implementation himem points to the first free byte on the heap. | |
| * The payload is stored first and then the header. | |
| * | |
| * The new heap code uses name_t for the name of the object. | |
| * | |
| */ | |
| address_t bmalloc(name_t* name, address_t l) { | |
| address_t payloadsize; /* the payload size */ | |
| address_t heapheadersize = sizeof(name_t) + addrsize; /* this is only used to estimate the free space, it is the maximum */ | |
| address_t b=himem; /* the current position on the heap, we store it in case of errors */ | |
| /* Initial DEBUG message. */ | |
| if (DEBUG) { | |
| outsc("** bmalloc with token "); | |
| outnumber(name->token); outspc(); | |
| outname(name); outspc(); | |
| outnumber(l); outcr(); | |
| } | |
| /* | |
| * How much space does the payload of the object need? | |
| */ | |
| switch(name->token) { | |
| case VARIABLE: /* a variable needs numsize bytes*/ | |
| payloadsize=numsize; | |
| break; | |
| case ARRAYVAR: /* a one dimensional array needs numsize*l bytes */ | |
| payloadsize=numsize*l; | |
| break; | |
| case ARRAYVAR: /* a two dimensional array needs numsize*l bytes plus one word for the additional dimension*/ | |
| payloadsize=numsize*l+addrsize; | |
| break; | |
| case TFN: /* the jump address, the type of function/type of return value, the number of vars | |
| and all variables are stored*/ | |
| payloadsize=addrsize+2+sizeof(name_t)*l; | |
| break; | |
| /* these are plain buffers allocated by the MALLOC call in BASIC */ | |
| default: | |
| payloadsize=l; | |
| } | |
| /* enough memory ?, on an EEPROM system we limit the heap to the RAM */ | |
| if ((himem-top) < payloadsize+heapheadersize) { error(EOUTOFMEMORY); return 0;} | |
| if (himem-(elength()-eheadersize) < payloadsize+heapheadersize) { error(EOUTOFMEMORY); return 0;} | |
| /* first we reserve space for the payload, address points to the first byte of the payload */ | |
| /* b points to the first free byte after the payload*/ | |
| b-=payloadsize; | |
| bfind_object.address=b+1; | |
| /* for ARRAYS, STRINGS and BUFFERS, store the object length now - these are variable size objects*/ | |
| if (name->token != VARIABLE) { | |
| b-=(addrsize-1); | |
| setaddress(b, memwrite2, payloadsize); | |
| if (DEBUG) { | |
| outsc("** bmalloc writes payloadsize "); outnumber(payloadsize); | |
| outsc(" at "); outnumber(b); outcr(); | |
| } | |
| b--; | |
| } | |
| /* store the name of the objects including the type as identifier */ | |
| b=setname_heap(b, name); | |
| /* store the type of the object */ | |
| memwrite2(b--, name->token); | |
| /* if anything went wrong we exit here without changing himem */ | |
| if (b < top || er) { error(EOUTOFMEMORY); return 0; } | |
| /* we fill the cache here as well, both right now for compatibility */ | |
| bfind_object.name=*name; | |
| bfind_object.size=payloadsize; | |
| /* himem is the next free byte now again */ | |
| himem=b; | |
| if (DEBUG) { | |
| outsc("** bmalloc returns "); outnumber(bfind_object.address); | |
| outsc(" himem is "); outnumber(himem); outcr(); | |
| } | |
| /* return the address of the payload */ | |
| return bfind_object.address; | |
| } | |
| address_t bfind(name_t* name) { | |
| address_t b, b0; | |
| address_t i=0; | |
| /* Initial DEBUG message. */ | |
| if (DEBUG) { | |
| outsc("*** bfind called for "); outname(name); | |
| outsc(" on heap with token "); outnumber(name->token); | |
| outsc(" himem is "); outnumber(himem); outcr(); | |
| } | |
| /* do we have anything on the heap? */ | |
| if (himem == memsize) return 0; else b=himem+1; | |
| /* we have the object already in cache and return */ | |
| if (name->token == bfind_object.name.token && cmpname(name, &bfind_object.name)) { | |
| if (DEBUG) { outsc("*** bfind found in cache "); outname(name); outsc(" at "); outnumber(bfind_object.address); outcr(); } | |
| return bfind_object.address; | |
| } | |
| /* walk through the heap from the last object added to the first */ | |
| while (b <= memsize) { | |
| /* get the name and the type */ | |
| bfind_object.name.token=memread2(b++); | |
| b=getname(b, &bfind_object.name, memread2); | |
| /* determine the size of the object and advance */ | |
| if (bfind_object.name.token != VARIABLE) { | |
| bfind_object.size=getaddress(b, memread2); | |
| b+=addrsize; | |
| } else { | |
| bfind_object.size=numsize; | |
| } | |
| /* this is the location of the payload */ | |
| bfind_object.address=b; | |
| /* have we found the object */ | |
| if (name->token == bfind_object.name.token && cmpname(name, &bfind_object.name)) { | |
| if (DEBUG) { outsc("*** bfind found "); outname(name); outsc(" at "); outnumber(bfind_object.address); outcr(); } | |
| return bfind_object.address; | |
| } | |
| /* advance on the heap */ | |
| b0=b; | |
| b+=bfind_object.size; | |
| /* safety net */ | |
| if (b0 > b) { | |
| error(EVARIABLE); | |
| return 0; | |
| } | |
| } | |
| /* nothing found return 0 and clear the cache */ | |
| if (DEBUG) { outsc("bfind returns 0"); outcr(); } | |
| zeroheap(&bfind_object); | |
| return 0; | |
| } | |
| /* reimplementation bfree with name interface */ | |
| address_t bfree(name_t* name) { | |
| address_t b; | |
| address_t i; | |
| if (DEBUG) { outsc("*** bfree called for "); outname(name); outsc(" on heap with token "); outnumber(name->token); outcr(); } | |
| /* use bfind to find the place */ | |
| b=bfind(name); | |
| /* nothing found, return 0 */ | |
| if (b == 0) return 0; | |
| if (DEBUG) { outsc("** bfree found "); outnumber(b); outcr(); } | |
| /* clear the entire memory area */ | |
| for (i=himem; i<=b+bfind_object.size-1; i++) memwrite2(i, 0); | |
| /* set the number of variables to the new value */ | |
| himem=b+bfind_object.size-1; | |
| if (DEBUG) { outsc("** bfree returns "); outnumber(himem); outcr(); } | |
| /* forget the chache, because heap structure has changed !! */ | |
| zeroheap(&bfind_object); | |
| return himem; | |
| } | |
| /* the length of an object, we directly return from the cache */ | |
| address_t blength(name_t* name) { | |
| if (bfind(name)) return bfind_object.size; else return 0; | |
| } | |
| /* reimplementation of getvar and setvar with name_t */ | |
| number_t getvar(name_t *name){ | |
| address_t a; | |
| if (DEBUG) { outsc("* getvar "); outname(name); outspc(); outcr(); } | |
| /* the special variables */ | |
| if (name->c[0] == '@') { | |
| switch (name->c[1]) { | |
| case 'A': | |
| return availch(); | |
| case 'S': | |
| return ert|ioer; | |
| case 'I': | |
| return id; | |
| case 'O': | |
| return od; | |
| case 'C': | |
| if (availch()) return inch(); else return 0; | |
| case 'E': | |
| return elength()/numsize; | |
| case 0: | |
| return (himem-top)/numsize; | |
| case 'R': | |
| return rd; | |
| case 'U': | |
| return getusrvar(); | |
| case 'V': | |
| return vlength; | |
| case 'X': | |
| return dspgetcursorx(); | |
| case 'Y': | |
| return dspgetcursory(); | |
| } | |
| } | |
| /* search the heap first */ | |
| a=bfind(name); | |
| if (!USELONGJUMP && er) return 0; | |
| /* if we don't find on the heap and it is not a static variable, we autocreate */ | |
| if (a == 0) { | |
| a=bmalloc(name, 0); | |
| if (!USELONGJUMP && er) return 0; | |
| } | |
| /* something went wrong */ | |
| if (a == 0) { | |
| error(EVARIABLE); | |
| return 0; | |
| } | |
| /* retrieve the value */ | |
| return getnumber(a, memread2); | |
| /* we only have the static variable array */ | |
| if (name->c[0] >= 65 && name->c[0] <= 91 && name->c[1] == 0) return vars[name->c[0]-65]; | |
| /* systems without Apple1 extension i.e. HEAP throw an error */ | |
| error(EVARIABLE); | |
| return 0; | |
| } | |
| /* set and create a variable */ | |
| void setvar(name_t *name, number_t v){ | |
| address_t a; | |
| if (DEBUG) { outsc("* setvar "); outname(name); outspc(); outnumber(v); outcr(); } | |
| /* the special variables */ | |
| if (name->c[0] == '@') | |
| switch (name->c[1]) { | |
| case 'S': | |
| ert=v; | |
| ioer=v; | |
| return; | |
| case 'I': | |
| id=v; | |
| return; | |
| case 'O': | |
| od=v; | |
| return; | |
| case 'C': | |
| outch(v); | |
| return; | |
| case 'R': | |
| rd=v; | |
| return; | |
| case 'U': | |
| setusrvar(v); | |
| return; | |
| case 'V': | |
| return; | |
| case 'X': | |
| dspsetcursorx((int)v); | |
| /* set the charcount, this is half broken but works */ | |
| if (od > 0 && od <= OPRT) charcount[od-1]=v; | |
| return; | |
| case 'Y': | |
| dspsetcursory((int)v); | |
| return; | |
| } | |
| /* dynamically allocated vars */ | |
| a=bfind(name); | |
| /* autocreate if not found */ | |
| if (a == 0) { | |
| a=bmalloc(name, 0); | |
| if (!USELONGJUMP && er) return; | |
| } | |
| /* something went wrong */ | |
| if (a == 0) { | |
| error(EVARIABLE); | |
| return; | |
| } | |
| /* set the value */ | |
| setnumber(a, memwrite2, v); | |
| /* the static variable array */ | |
| if (name->c[1] == 0 && name->c[0] >= 65 && name->c[0] <= 91) { | |
| vars[name->c[0]-65]=v; | |
| return; | |
| } | |
| error(EVARIABLE); | |
| } | |
| /* clr all variables */ | |
| void clrvars() { | |
| /* delete all static variables */ | |
| address_t i; | |
| /* clear static variable (only on no heap systems) */ | |
| for (i=0; i<VARSIZE; i++) vars[i]=0; | |
| /* then set the entire mem area to zero */ | |
| for (i=himem; i<memsize; i++) memwrite2(i, 0); | |
| /* reset the heap start*/ | |
| himem=memsize; | |
| /* and clear the cache */ | |
| zeroheap(&bfind_object); | |
| } | |
| /* | |
| * The BASIC memory access function. | |
| * | |
| * getnumber2, getaddress and getstrlength are purely local, they | |
| * can be used with any memory reader function unlike the old BASIC 1 | |
| * code. | |
| */ | |
| /* a generic memory reader for numbers */ | |
| number_t getnumber(address_t m, memreader_t f) { | |
| mem_t i; | |
| accu_t z; | |
| for (i=0; i<numsize; i++) z.c[i]=f(m++); | |
| return z.n; | |
| } | |
| /* same for addresses */ | |
| address_t getaddress(address_t m, memreader_t f) { | |
| mem_t i; | |
| accu_t z; | |
| for (i=0; i<addrsize; i++) z.c[i]=f(m++); | |
| return z.a; | |
| } | |
| /* same for strings */ | |
| stringlength_t getstrlength(address_t m, memreader_t f) { | |
| mem_t i; | |
| accu_t z; | |
| z.a=0; | |
| for (i=0; i<strindexsize; i++) z.c[i]=f(m++); | |
| return z.a; | |
| } | |
| /* set a number at a memory location, new version */ | |
| void setnumber(address_t m, memwriter_t f, number_t v){ | |
| mem_t i; | |
| accu_t z; | |
| z.n=v; | |
| for (i=0; i<numsize; i++) f(m++, z.c[i]); | |
| } | |
| /* set an address at a memory location */ | |
| void setaddress(address_t m, memwriter_t f, address_t a){ | |
| mem_t i; | |
| accu_t z; | |
| z.a=a; | |
| for (i=0; i<addrsize; i++) f(m++, z.c[i]); | |
| } | |
| /* set a stringlength at a memory location */ | |
| void setstrlength(address_t m, memwriter_t f, stringlength_t s){ | |
| mem_t i; | |
| accu_t z; | |
| z.s=s; | |
| for (i=0; i<strindexsize; i++) f(m++, z.c[i]); | |
| } | |
| /* | |
| * Code to handle names. These function mostly deal with the true | |
| * name part of name_t and not the token. The token has to be | |
| * processed by the caller. Name byte order conventon is | |
| * to have the first character in the lower byte and the second | |
| * character in the higher byte. Optionally, for HASLONGNAMES | |
| * the length of the name is stored before the name. | |
| * | |
| * setname_* sets a name and advance the number of bytes the name uses. | |
| * Two versions are needed because the heap is counted down | |
| * while the pgm is counted up. The length of the name | |
| * is always 2 bytes now but will be variable in the future. | |
| * | |
| * getname needs to go through a memreader because names are | |
| * read from eeproms as well! | |
| * | |
| */ | |
| /* this one is for the heap were we count down writing*/ | |
| address_t setname_heap(address_t m, name_t* name) { | |
| memwrite2(m--, name->c[1]); | |
| memwrite2(m--, name->c[0]); | |
| return m; | |
| } | |
| /* this one is for the pgm were we count up writing */ | |
| address_t setname_pgm(address_t m, name_t* name) { | |
| memwrite2(m++, name->c[0]); | |
| memwrite2(m++, name->c[1]); | |
| return m; | |
| } | |
| /* get a name from a memory location */ | |
| address_t getname(address_t m, name_t* name, memreader_t f) { | |
| name->c[0]=f(m++); | |
| name->c[1]=f(m++); | |
| return m; | |
| } | |
| /* compare two names */ | |
| mem_t cmpname(name_t* a, name_t* b) { | |
| if (a->c[0] == b->c[0] && a->c[1] == b->c[1]) return 1; else return 0; | |
| } | |
| /* zero a name and a heap object */ | |
| void zeroname(name_t* name) { | |
| name->c[0]=0; | |
| name->c[1]=0; | |
| name->token=0; | |
| } | |
| void zeroheap(heap_t* heap) { | |
| heap->address=0; | |
| heap->size=0; | |
| zeroname(&heap->name); | |
| } | |
| /* output a name */ | |
| void outname(name_t* name) { | |
| outch(name->c[0]); | |
| if (name->c[1]) outch(name->c[1]); | |
| } | |
| /* this one is for the heap were we count down writing*/ | |
| address_t setname_heap(address_t m, name_t* name) { | |
| mem_t l; | |
| for(l=name->l; l>0; l--) memwrite2(m--, name->c[l-1]); | |
| memwrite2(m--, name->l); | |
| return m; | |
| } | |
| /* this one is for the pgm were we count up writing */ | |
| address_t setname_pgm(address_t m, name_t* name) { | |
| mem_t l; | |
| memwrite2(m++, name->l); | |
| for(l=0; l<name->l; l++) memwrite2(m++, name->c[l]); | |
| return m; | |
| } | |
| /* get a name from a memory location */ | |
| address_t getname(address_t m, name_t* name, memreader_t f) { | |
| mem_t l; | |
| name->l=f(m++); | |
| for(l=0; l<name->l; l++) name->c[l]=f(m++); | |
| for(; l<MAXNAME; l++) name->c[l]=0; /* should not be there, is needed for | |
| now because the lexer is not implemented correctly*/ | |
| return m; | |
| } | |
| /* compare two names */ | |
| mem_t cmpname(name_t* a, name_t* b) { | |
| mem_t l; | |
| if (a->l != b->l) return 0; | |
| for(l=0; l<a->l; l++) if (a->c[l] != b->c[l]) return 0; | |
| return 1; | |
| } | |
| /* zero a name and a heap object */ | |
| void zeroname(name_t* name) { | |
| mem_t l; | |
| name->l=0; | |
| for(l=0; l<MAXNAME; l++) name->c[l]=0; | |
| name->token=0; | |
| } | |
| void zeroheap(heap_t* heap) { | |
| heap->address=0; | |
| heap->size=0; | |
| zeroname(&heap->name); | |
| } | |
| /* output a name */ | |
| void outname(name_t* name) { | |
| mem_t l; | |
| for(l=0; l<name->l; l++) outch(name->c[l]); | |
| } | |
| /* create an array */ | |
| /* reimplementation with name_t */ | |
| address_t createarray(name_t* variable, address_t i, address_t j) { | |
| address_t a; | |
| /* if we want to me MS compatible, the array ranges from 0-n */ | |
| if (msarraylimits) { | |
| i+=1; | |
| j+=1; | |
| } | |
| /* this code allows redimension now for local variables */ | |
| if (DEBUG) { | |
| outsc("* create array "); outname(variable); outspc(); | |
| outsc("* with name length "); outnumber(variable->l); outspc(); | |
| outnumber(i); outspc(); outnumber(j); outcr(); | |
| } | |
| return bmalloc(variable, i); | |
| /* allocate the array space */ | |
| a=bmalloc(variable, i*j); | |
| /* store the dimension of the array at the beginning of the array area */ | |
| setaddress(a+i*j*numsize, memwrite2, j); | |
| /* return value is the address of the payload area */ | |
| return a; | |
| return 0; | |
| } | |
| /* | |
| * Reimplementation of the array function to avoid the various problem with the old one. | |
| * we use the lefthandside object here with the convention that i is the first index | |
| * and j the second index. This is inconsistent with the use in strings. Will be fixed | |
| * when a true indexing type is introduced. | |
| */ | |
| void array(lhsobject_t* object, mem_t getset, number_t* value) { | |
| address_t a; /* the address of the array element */ | |
| address_t h; /* the number of elements in the array */ | |
| address_t l=arraylimit; /* the lower limit, defaults to the arraylimit, here for further use */ | |
| address_t dim=1; /* the array dimension */ | |
| if (DEBUG) { | |
| outsc("* array2: accessing "); | |
| outname(&name); outspc(); outspc(); | |
| outnumber(object->i); outspc(); | |
| outnumber(object->j); outspc(); | |
| outsc(" getset "); outch(getset); outcr(); | |
| } | |
| /* handling the special array, range check and access is done here */ | |
| if (object->name.c[0] == '@') { | |
| switch(object->name.c[1]) { | |
| case 'E': | |
| h=elength()/numsize; | |
| a=elength()-numsize*object->i; | |
| if (a < eheadersize) { error(EORANGE); return; } | |
| if (getset == 'g') *value=getnumber(a, eread); | |
| else if (getset == 's') setnumber(a, eupdate, *value); | |
| return; | |
| case 'D': | |
| if (getset == 'g') *value=dspget(object->i-1); | |
| else if (getset == 's') dspset(object->i-1, *value); | |
| return; | |
| case 'T': | |
| if (getset == 'g') *value=rtcget(object->i); | |
| else if (getset == 's') rtcset(object->i, *value); | |
| return; | |
| case 'S': | |
| if (getset == 'g') *value=sensorread(object->i, 0); | |
| return; | |
| case 'U': | |
| if (getset == 'g') *value=getusrarray(object->i); | |
| else if (getset == 's') setusrarray(object->i, *value); | |
| return; | |
| case 0: | |
| h=(himem-top)/numsize; | |
| a=himem-numsize*(object->i+1)+1; | |
| if (object->i < 0 || a < top) { error(EORANGE); return; } | |
| if (getset == 'g') *value=getnumber(a, memread2); | |
| else if (getset == 's') setnumber(a, memwrite2, *value); | |
| return; | |
| case 'M': | |
| h=himem-top; | |
| a=himem-object->i; | |
| if (object->i < 0 || a < top) { error(EORANGE); return; } | |
| if (getset == 'g') *value=memread2(a); | |
| else if (getset == 's') memwrite2(a, *value); | |
| return; | |
| default: | |
| error(EVARIABLE); | |
| return; | |
| } | |
| } else { | |
| /* dynamically allocated arrays */ | |
| object->name.token=ARRAYVAR; | |
| if (!(a=bfind(&object->name))) a=createarray(&object->name, ARRAYSIZEDEF, 1); | |
| if (!USELONGJUMP && er) return; | |
| /* multidim reserves one address word for the dimension, hence we have less bytes */ | |
| h=bfind_object.size/numsize; | |
| h=(bfind_object.size-addrsize)/numsize; | |
| if (DEBUG) { | |
| outsc("** in array dynamical base address "); outnumber(a); | |
| outsc(" and array element number"); outnumber(h); | |
| outcr(); | |
| } | |
| dim=getaddress(a+bfind_object.size-2, memread2); | |
| if (DEBUG) { | |
| outsc("** in array, second dimension is "); outnumber(dim); | |
| outspc(); outnumber(a+bfind_object.size); | |
| outcr(); | |
| } | |
| a=a+((object->i-l)*dim+(object->j-l))*numsize; | |
| a=a+(object->i-l)*numsize; | |
| error(EVARIABLE); | |
| return; | |
| } | |
| /* range check */ | |
| if ( (object->j < l) || (object->j >= dim+l) || (object->i < l) || (object->i >= h/dim+l)) { error(EORANGE); return; } | |
| if ( (object->i < l) || (object->i >= h+l) ) { error(EORANGE); return; } | |
| /* set or get the array */ | |
| if (getset == 'g') *value=getnumber(a, memread2); | |
| else if (getset == 's') setnumber(a, memwrite2, *value); | |
| } | |
| /* Create a string on the heap, i is the length of the string, j the dimension of the array */ | |
| address_t createstring(name_t* variable, address_t i, address_t j) { | |
| address_t a; | |
| if (DEBUG) { outsc("Create string "); outname(variable); outcr(); } | |
| /* the MS string compatibility, DIM 10 creates 11 elements */ | |
| if (msarraylimits) j+=1; | |
| /* if no string arrays are in the code, we reserve the number of bytes i and space for the index */ | |
| /* allow redimension without check right now, for local variables */ | |
| a=bmalloc(variable, i+strindexsize); | |
| if (er != 0) return 0; | |
| return a; | |
| /* string arrays need the number of array elements which address_ hence addresize bytes and then | |
| the space for j strings */ | |
| /* allow redimension without check right now, for local variables */ | |
| a=bmalloc(variable, addrsize+j*(i+strindexsize)); | |
| if (er != 0) return 0; | |
| /* set the array length */ | |
| setaddress(a+j*(i+strindexsize), memwrite2, j); | |
| /* return the address of the first string */ | |
| return a; | |
| if (er != 0) return 0; | |
| return a; | |
| return 0; | |
| } | |
| /* | |
| * Get a string at position b. The -1+stringdexsize is needed because a string index starts with 1 | |
| * in addition to the memory pointer, return the address in memory. | |
| * We use a pointer to memory here instead of going through the mem interface with an integer variable. | |
| * This makes string code lean to compile but is awkward for systems with serial memory. | |
| * | |
| * Getstring returns a pointer to the first string element in question. | |
| * | |
| * All side effects are removed now. (Hopefully). | |
| */ | |
| /* helpers to handle strings */ | |
| /* stores a C string to a BASIC string variable */ | |
| void storecstring(address_t ax, address_t s, char* b) { | |
| address_t k; | |
| for (k=0; k<s-strindexsize && b[k] != 0; k++) memwrite2(ax+k+strindexsize, b[k]); | |
| setstrlength(ax, memwrite2, k); | |
| } | |
| /* length of a c string up to a limit l */ | |
| address_t cstringlength(char* c, address_t l) { | |
| address_t a; | |
| while(a < l && c[a] != 0) a++; | |
| return a; | |
| } | |
| /* get a memory pointer to a string, new version */ | |
| void getstring(string_t* strp, name_t* name, address_t b, address_t j) { | |
| address_t k, zt; | |
| address_t ax; /* ax is local now - no side effects any more - can be done in strp->address */ | |
| /* we know nothing about the string */ | |
| ax=0; | |
| strp->address=0; | |
| strp->ir=0; | |
| strp->length=0; | |
| strp->arraydim=1; | |
| strp->strdim=0; | |
| if (DEBUG) { | |
| outsc("* getstring from var "); outname(name); outspc(); | |
| outnumber(b); outspc(); | |
| outnumber(j); outcr(); | |
| } | |
| /* special string variables */ | |
| if (name->c[0] == '@') | |
| switch(name->c[1]) { | |
| case 0: | |
| strp->ir=ibuffer+b; | |
| strp->length=ibuffer[0]; | |
| strp->strdim=BUFSIZ-2; | |
| return; | |
| default: | |
| error(EVARIABLE); | |
| return; | |
| case 'U': | |
| makeusrstring(); /* a user definable special string in sbuffer */ | |
| strp->ir=sbuffer+b; | |
| strp->length=sbuffer[0]; | |
| return; | |
| case 'T': | |
| rtcmkstr(); /* the time string */ | |
| strp->ir=rtcstring+b; | |
| strp->length=rtcstring[0]; | |
| return; | |
| /* the arguments string on POSIX systems */ | |
| case 'A': | |
| if (bargc > 2) { | |
| strp->ir=bargv[2]; | |
| strp->length=cstringlength(bargv[2], BUFSIZE); | |
| return; | |
| } | |
| return; | |
| } | |
| /* dynamically allocated strings, create on the fly */ | |
| if (!(ax=bfind(name))) ax=createstring(name, defaultstrdim, arraylimit); | |
| if (DEBUG) { | |
| outsc("** heap address "); outnumber(ax); outcr(); | |
| if (ax) { outsc("** byte length of string memory segment "); outnumber(bfind_object.size); outcr(); } | |
| } | |
| /* string creating has caused an error, typically no memoryy */ | |
| if (!USELONGJUMP && er) return; | |
| /* the maximum length of the string */ | |
| strp->strdim=bfind_object.size-strindexsize; | |
| /* are we in range */ | |
| if ((b < 1) || (b > strp->strdim )) { error(EORANGE); return; } | |
| if (DEBUG) { outsc("** maximum string length "); outnumber(strp->strdim); outcr(); } | |
| /* get the actual full length, this is redundant to lenstring but lenstring does | |
| not autocreate */ | |
| strp->length=getstrlength(ax, memread2); | |
| /* now find the payload address */ | |
| ax=ax+strindexsize+(b-1); | |
| /* the dimension of the string array */ | |
| /* it is at the top of the string, this uses a side effect of bfind */ | |
| strp->arraydim=getaddress(ax + bfind_object.size - addrsize, memread2); | |
| /* is the array index in range */ | |
| if ((j < arraylimit) || (j >= strp->arraydim + arraylimit )) { error(EORANGE); return; } | |
| if (DEBUG) { outsc("** string dimension "); outnumber(strp->arraydim); outcr(); } | |
| /* the max length of a string */ | |
| strp->strdim=(bfind_object.size - addrsize)/strp->arraydim-strindexsize; | |
| /* are we in range */ | |
| if ((b < 1) || (b > strp->strdim )) { error(EORANGE); return; } | |
| if (DEBUG) { outsc("** maximum string length "); outnumber(strp->strdim); outcr(); } | |
| /* the base address of a string */ | |
| ax=ax+(j-arraylimit)*(strp->strdim + strindexsize); | |
| if (DEBUG) { outsc("** string base address "); outnumber(ax); outcr(); } | |
| /* from this base address we can get the actual length of the string */ | |
| strp->length=getstrlength(ax, memread2); | |
| /* the address of the payload */ | |
| ax=ax+b-1+strindexsize; | |
| if (DEBUG) { outsc("** payload address "); outnumber(ax); outcr(); } | |
| /* store the payload address to the string object, length to be done if needed!! */ | |
| strp->address=ax; | |
| /* return value is 0 if we have no direct memory access, the caller needs to handle the string | |
| through the mem address */ | |
| strp->ir=0; | |
| strp->ir=(char *)&mem[ax]; | |
| } | |
| /* reimplementation with name_t */ | |
| /* set the length of a string */ | |
| void setstringlength(name_t* name, address_t l, address_t j) { | |
| address_t a; | |
| stringlength_t stringdim; | |
| if (DEBUG) { | |
| outsc("** setstringlength "); | |
| outname(name); | |
| outspc(); outnumber(l); outspc(); outnumber(j); | |
| outcr(); | |
| } | |
| /* the special strings */ | |
| if (name->c[0] == '@') | |
| switch(name->c[1]) { | |
| case 0: | |
| *ibuffer=l; | |
| return; | |
| case 'U': | |
| /* do nothing here for the moment */ | |
| return; | |
| } | |
| /* find the variable address */ | |
| a=bfind(name); | |
| if (!USELONGJUMP && er) return; | |
| if (a == 0) { error(EVARIABLE); return; } | |
| /* stringdim calculation moved here */ | |
| stringdim=bfind_object.size-strindexsize; | |
| /* getaddress seeks the dimension of the string array directly after the payload */ | |
| stringdim=(bfind_object.size - addrsize)/(getaddress(a + bfind_object.size - addrsize, memread2)) - strindexsize; | |
| /* where do we write it to */ | |
| a=a+(stringdim+strindexsize)*(j-arraylimit); | |
| if (DEBUG) { outsc("** setstringlength writing to "); outnumber(a); outsc(" value "); outnumber(l); outcr(); } | |
| setstrlength(a, memwrite2, l); | |
| } | |
| /* the BASIC string mechanism for real time clocks, create a string with the clock data */ | |
| void rtcmkstr() { | |
| int cc = 1; | |
| int t; | |
| /* hours */ | |
| t=rtcget(2); | |
| rtcstring[cc++]=t/10+'0'; | |
| rtcstring[cc++]=t%10+'0'; | |
| rtcstring[cc++]=':'; | |
| /* minutes */ | |
| t=rtcget(1); | |
| rtcstring[cc++]=t/10+'0'; | |
| rtcstring[cc++]=t%10+'0'; | |
| rtcstring[cc++]=':'; | |
| /* seconds */ | |
| t=rtcget(0); | |
| rtcstring[cc++]=t/10+'0'; | |
| rtcstring[cc++]=t%10+'0'; | |
| rtcstring[cc++]='-'; | |
| /* days */ | |
| t=rtcget(4); | |
| if (t/10 > 0) rtcstring[cc++]=t/10+'0'; | |
| rtcstring[cc++]=t%10+'0'; | |
| rtcstring[cc++]='/'; | |
| /* months */ | |
| t=rtcget(5); | |
| if (t/10 > 0) rtcstring[cc++]=t/10+'0'; | |
| rtcstring[cc++]=t%10+'0'; | |
| rtcstring[cc++]='/'; | |
| /* years */ | |
| t=rtcget(6)%100; /* only 100 years no 19xx epochs */ | |
| if (t/10 > 0) rtcstring[cc++]=t/10+'0'; | |
| rtcstring[cc++]=t%10+'0'; | |
| rtcstring[cc]=0; | |
| /* needed for BASIC strings */ | |
| rtcstring[0]=cc-1; | |
| } | |
| /* | |
| * Layer 0 Extension: the user defined extension functions, use this function for a | |
| * quick way to extend BASIC. | |
| * | |
| * @U is a single variable that can be set and get. Every access to this variable | |
| * calls getusrvar() or setusrvar(). | |
| * @U() is a 1d array every access calls get or setusrarray(). | |
| * @U$ is a read only string created by makeusrstring(). | |
| * USR(32, V) is a user defined function created by usrfunction(). | |
| * CALL 32 is a user defined call | |
| * | |
| */ | |
| /* read or write the variable @U, you can do anything you want here */ | |
| number_t getusrvar() { return 0; } | |
| void setusrvar(number_t v) {return; } | |
| /* read or write the array @U(), you can do anything you want here */ | |
| number_t getusrarray(address_t i) {return 0;} | |
| void setusrarray(address_t i, number_t v) {return; } | |
| /* make the usr string from @U$, this function can be called multiple times for one operation */ | |
| void makeusrstring() { | |
| /* | |
| * sample code could be: | |
| * | |
| * mem_t i; | |
| * const char text[] = "hello world"; | |
| * for(i=0; i<SBUFSIZE-1 && text[i]!=0 ; i++) sbuffer[i]=text[i]; | |
| * | |
| * Always set sbuffer[0] to the string length, keep in mind that sbuffer | |
| * is 32 bytes long by default | |
| */ | |
| sbuffer[0]=0; | |
| } | |
| /* USR with arguments > 31 calls this */ | |
| number_t usrfunction(address_t i, number_t v) { return 0; } | |
| /* CALL with arguments > 31 calls this */ | |
| void usrcall(address_t i) { return; } | |
| /* | |
| * Layer 0 - keyword handling - PROGMEM logic goes here | |
| * getkeyword(), getmessage(), and getokenvalue() are | |
| * the only access to the keyword array in the code. | |
| * | |
| * Same for messages and errors. | |
| * | |
| * All keywords are stored in PROGMEM, the Arduino way to store | |
| * constant data in flash memory. sbuffer is used to recall the | |
| * keywords and messages from PROGMEM on Arduino. Other systems | |
| * use the keyword array directly. | |
| */ | |
| char* getkeyword(address_t i) { | |
| if (DEBUG) { outsc("** getkeyword from index "); outnumber(i); outcr(); } | |
| return (char *) keyword[i]; | |
| strcpy_P(sbuffer, (char*) pgm_read_ptr(&(keyword[i]))); | |
| return sbuffer; | |
| } | |
| /* messages are read from the message array */ | |
| char* getmessage(char i) { | |
| if (i >= sizeof(message) || i < 0) return 0; | |
| return (char *) message[i]; | |
| strcpy_P(sbuffer, (char*) pgm_read_ptr(&(message[i]))); | |
| return sbuffer; | |
| } | |
| /* tokens read here are token_t constructed from multi byte sequences */ | |
| token_t gettokenvalue(address_t i) { | |
| if (i >= sizeof(tokens)) return 0; | |
| return tokens[i]; | |
| return (token_t) pgm_read_byte(&tokens[i]); | |
| return (token_t) pgm_read_word(&tokens[i]); | |
| } | |
| /* print a message directly to the default outpur stream */ | |
| void printmessage(char i){ | |
| if (i > EGENERAL) return; | |
| outsc((char *)getmessage(i)); | |
| } | |
| /* | |
| * Layer 0 - error handling | |
| * | |
| * The general error handler. The static variable er | |
| * contains the error state. | |
| * | |
| * debugtoken() writes a token for debug | |
| * bdebug() is the general debug message function | |
| * | |
| * Strategy: the error() function writes the message and then | |
| * clears the stack. All calling functions must check er and | |
| * return after funtion calls with no further messages etc. | |
| * reseterror() sets the error state to normal and end the | |
| * run loop. | |
| */ | |
| void error(token_t e){ | |
| address_t i; | |
| /* store the error number */ | |
| er=e; | |
| /* clear the stacks */ | |
| clearst(); | |
| clrforstack(); | |
| clrgosubstack(); | |
| /* switch off all timers and interrupts */ | |
| resettimer(&after_timer); | |
| resettimer(&every_timer); | |
| /* is the error handler active? then silently go if we do GOTO or CONT actions in it */ | |
| if (st != SINT && (berrorh.type == TGOTO || berrorh.type == TCONT)) return; | |
| if (st != SINT && (berrorh.type == TGOTO || berrorh.type == TCONT)) longjmp(sthook, er); | |
| /* set input and output device back to default, and delete the form */ | |
| iodefaults(); | |
| form=0; | |
| /* find the line number if in RUN modes */ | |
| if (st != SINT) { | |
| outnumber(myline(here)); | |
| outch(':'); | |
| outspc(); | |
| } | |
| /* if we have error messages, display them */ | |
| if (e > 0) | |
| printmessage(e); | |
| else { | |
| for(i=0; gettokenvalue(i)!=0 && gettokenvalue(i)!=e; i++); | |
| outsc(getkeyword(i)); | |
| } | |
| outspc(); | |
| printmessage(EGENERAL); | |
| printmessage(EGENERAL); | |
| outspc(); | |
| outnumber(er); | |
| if (DEBUG) { outsc("** at "); outnumber(here); } | |
| outcr(); | |
| /* reset fncontext - this is odd */ | |
| fncontext=0; | |
| /* we return to the statement loop, bringing the error with us */ | |
| longjmp(sthook, er); | |
| } | |
| void reseterror() { | |
| er=0; | |
| here=0; | |
| st=SINT; | |
| } | |
| void debugtoken(){ | |
| outsc("* "); | |
| if (debuglevel>2) { outnumber(here); outsc(" * "); } | |
| if (token == EOL) { | |
| outsc("EOL"); | |
| return; | |
| } | |
| switch(token) { | |
| case LINENUMBER: | |
| printmessage(MLINE); | |
| break; | |
| case NUMBER: | |
| printmessage(MNUMBER); | |
| break; | |
| case VARIABLE: | |
| printmessage(MVARIABLE); | |
| break; | |
| case ARRAYVAR: | |
| printmessage(MARRAY); | |
| break; | |
| case STRING: | |
| printmessage(MSTRING); | |
| break; | |
| case STRINGVAR: | |
| printmessage(MSTRINGVAR); | |
| break; | |
| } | |
| outspc(); | |
| outputtoken(); | |
| } | |
| void bdebug(const char *c){ | |
| outch('*'); | |
| outspc(); | |
| outsc(c); | |
| debugtoken(); | |
| outcr(); | |
| } | |
| /* | |
| * Arithmetic and runtime operations are mostly done | |
| * on a stack of number_t. | |
| * | |
| * push(), pop(), clearst() handle the stack | |
| */ | |
| void push(number_t t){ | |
| if (DEBUG) { | |
| outsc("** push sp= "); outnumber(sp); outcr(); | |
| outsc("** push value= "); outnumber(t); outcr(); | |
| } | |
| /* in forced integer mode every operation is truncated */ | |
| if (forceint) t=trunc(t); | |
| if (sp == STACKSIZE) | |
| error(ESTACK); | |
| else | |
| stack[sp++].n=t; | |
| } | |
| number_t pop(){ | |
| if (DEBUG) { | |
| outsc("** pop sp= "); outnumber(sp); outcr(); | |
| outsc("** pop value= "); outnumber(stack[sp-1].n); outcr(); | |
| } | |
| if (sp == 0) { error(ESTACK); return 0; } | |
| return stack[--sp].n; | |
| } | |
| void pushaddress2(address_t a) { | |
| if (DEBUG) { | |
| outsc("** push sp= "); outnumber(sp); outcr(); | |
| outsc("** push value= "); outnumber(a); outcr(); | |
| } | |
| if (sp == STACKSIZE) | |
| error(ESTACK); | |
| else | |
| stack[sp++].a=a; | |
| } | |
| address_t popaddress2() { | |
| if (DEBUG) { | |
| outsc("** pop sp= "); outnumber(sp); outcr(); | |
| outsc("** pop value= "); outnumber(stack[sp-1].a); outcr(); | |
| } | |
| if (sp == 0) { error(ESTACK); return 0; } | |
| return stack[--sp].a; | |
| } | |
| void pushinteger(index_t i) { | |
| if (DEBUG) { | |
| outsc("** push sp= "); outnumber(sp); outcr(); | |
| outsc("** push value= "); outnumber(i); outcr(); | |
| } | |
| if (sp == STACKSIZE) | |
| error(ESTACK); | |
| else | |
| stack[sp++].i=i; | |
| } | |
| index_t popinteger() { | |
| if (DEBUG) { | |
| outsc("** pop sp= "); outnumber(sp); outcr(); | |
| outsc("** pop value= "); outnumber(stack[sp-1].i); outcr(); | |
| } | |
| if (sp == 0) { error(ESTACK); return 0; } | |
| return stack[--sp].i; | |
| } | |
| /* this one gets a positive integer from the stack and traps the error*/ | |
| address_t popaddress(){ | |
| number_t tmp = 0; | |
| tmp=pop(); | |
| if (tmp < 0) { error(EORANGE); return 0;} | |
| return (address_t) tmp; | |
| } | |
| void clearst(){ | |
| sp=0; | |
| } | |
| /* these are not really stack operations but a way to handle temp char data */ | |
| address_t charsp; | |
| void pushchar(char ch) {} | |
| char popchar() { return 0; } | |
| /* | |
| * clear the cursor for the READ/DATA mechanism | |
| */ | |
| void clrdata() { | |
| data=0; | |
| } | |
| /* | |
| * Stack handling for FOR | |
| * Reimplementation of the for stack with names and with all loops | |
| * cleaned up. | |
| */ | |
| /* the new stack type for loops */ | |
| void pushloop(name_t* name, token_t t, address_t here, number_t to, number_t step) { | |
| address_t i; | |
| if (DEBUG) { | |
| outsc("** loopsp and here in pushloopstack "); | |
| outnumber(loopsp); outspc(); outnumber(here); outcr(); | |
| if (name != 0) { outsc("** loop name "); outname(name); outcr(); } | |
| else { outsc("** loop name is 0"); outcr(); } | |
| outsc("** loop token "); outnumber(t); outcr(); | |
| outsc("** loop to "); outnumber(to); outcr(); | |
| outsc("** loop step "); outnumber(step); outcr(); | |
| } | |
| /* | |
| * Before pushing into the loop stack we check is an | |
| * old loop exists. | |
| * | |
| * There are two situations to handle: | |
| * 1. A loop is reentered because a GOTO went back to or even before | |
| * the loop start. This is identified by the here location. | |
| * 2. A new FOR loop is created after the existing loop with the same | |
| * variable name. This happens if a jump or break went outside the | |
| * loop. This is identified by the variable name. | |
| */ | |
| /* Situation 1, scan for here */ | |
| for(i=0; i<loopsp; i++) { | |
| if (loopstack[i].here == here) { | |
| loopsp=i; | |
| break; | |
| } | |
| } | |
| /* Situation 2, scan for the name */ | |
| if (name != 0) { | |
| for(i=0; i<loopsp; i++) { | |
| if (cmpname(&loopstack[i].var, name)) { | |
| loopsp=i; | |
| break; | |
| } | |
| } | |
| } | |
| /* Add the loop to the stack */ | |
| if (loopsp < FORDEPTH) { | |
| if (t == TWHILE || t == TREPEAT) { | |
| loopstack[loopsp].var.token=t; | |
| } else { | |
| if (name != 0) { | |
| loopstack[loopsp].var=*name; | |
| } else { | |
| loopstack[loopsp].var.c[0]=0; | |
| loopstack[loopsp].var.l=0; | |
| loopstack[loopsp].var.token=0; | |
| } | |
| } | |
| loopstack[loopsp].here=here; | |
| loopstack[loopsp].to=to; | |
| loopstack[loopsp].step=step; | |
| loopsp++; | |
| return; | |
| } else | |
| error(ELOOP); | |
| } | |
| /* what is the active loop */ | |
| bloop_t* activeloop() { | |
| if (loopsp>0) { | |
| return &loopstack[loopsp-1]; | |
| } else { | |
| error(ELOOP); | |
| return 0; | |
| } | |
| } | |
| void droploop() { | |
| if (loopsp>0) { | |
| loopsp--; | |
| } else { | |
| error(ELOOP); | |
| return; | |
| } | |
| } | |
| void clrforstack() { | |
| loopsp=0; | |
| } | |
| /* GOSUB stack handling */ | |
| void pushgosubstack(mem_t a){ | |
| if (gosubsp < GOSUBDEPTH) { | |
| gosubstack[gosubsp]=here; | |
| gosubarg[gosubsp]=a; | |
| gosubsp++; | |
| } else | |
| error(TGOSUB); | |
| } | |
| void popgosubstack(){ | |
| if (gosubsp>0) { | |
| gosubsp--; | |
| } else { | |
| error(TRETURN); | |
| return; | |
| } | |
| here=gosubstack[gosubsp]; | |
| } | |
| void dropgosubstack(){ | |
| if (gosubsp>0) { | |
| gosubsp--; | |
| } else { | |
| error(TGOSUB); | |
| } | |
| } | |
| void clrgosubstack() { | |
| gosubsp=0; | |
| } | |
| /* two helper commands for structured BASIC, without the GOSUB stack */ | |
| void pushlocation(blocation_t* l) { | |
| if (st == SINT) l->location=bi-ibuffer; | |
| else l->location=here; | |
| l->token=token; | |
| } | |
| void poplocation(blocation_t* l) { | |
| if (st == SINT) bi=ibuffer+l->location; | |
| else here=l->location; | |
| token=l->token; | |
| } | |
| /* little helpers of the io functions */ | |
| /* send a newline */ | |
| void outcr() { | |
| if (sendcr) outch('\r'); | |
| outch('\n'); | |
| } | |
| /* send a space */ | |
| void outspc() { | |
| outch(' '); | |
| } | |
| /* output a zero terminated string - c style */ | |
| void outsc(const char *c){ while (*c != 0) outch(*c++); } | |
| /* output a zero terminated string in a formated box padding spaces | |
| needed for catalog output */ | |
| void outscf(const char *c, index_t f){ | |
| int i = 0; | |
| while (*c != 0) { outch(*c++); i++; } | |
| if (f > i) { | |
| f=f-i; | |
| while (f--) outspc(); | |
| } | |
| } | |
| /* | |
| * reading a positive number from a char buffer | |
| * maximum number of digits is adjusted to 16 | |
| * ugly here, testcode when introducting | |
| * number_t was only 16 bit before | |
| */ | |
| address_t parsenumber(char *c, number_t *r) { | |
| address_t nd = 0; | |
| *r=0; | |
| while (*c >= '0' && *c <= '9' && *c != 0) { | |
| *r=*r*10+*c++-'0'; | |
| nd++; | |
| if (nd == SBUFSIZE) break; | |
| } | |
| return nd; | |
| } | |
| /* a poor man's atof implementation with character count */ | |
| address_t parsenumber2(char *c, number_t *r) { | |
| address_t nd = 0; | |
| index_t i; | |
| number_t fraction = 0; | |
| number_t exponent = 0; | |
| mem_t nexp = 0; | |
| *r=0; | |
| /* integer part */ | |
| i=parsenumber(c, r); | |
| c+=i; | |
| nd+=i; | |
| /* the fractional part */ | |
| if (*c == '.') { | |
| c++; | |
| nd++; | |
| i=parsenumber(c, &fraction); | |
| c+=i; | |
| nd+=i; | |
| if (i > 0) { | |
| while ((--i)>=0) fraction=fraction/10; | |
| *r+=fraction; | |
| } | |
| } | |
| /* the exponent */ | |
| if (*c == 'E' || *c == 'e') { | |
| c++; | |
| nd++; | |
| if (*c == '-') {c++; nd++; nexp=1;}; | |
| i=parsenumber(c, &exponent); | |
| nd+=i; | |
| while ((--exponent)>=0) if (nexp) *r=*r/10; else *r=*r*10; | |
| } | |
| return nd; | |
| } | |
| /* | |
| * convert a number to a string | |
| * the argument type controls the largest displayable integer | |
| * default is int but it can be increased without any side effects | |
| */ | |
| address_t writenumber(char *c, wnumber_t v){ | |
| address_t nd = 0; | |
| index_t i,j; | |
| mem_t s = 1; | |
| char c1; | |
| /* the sign */ | |
| if (v<0) s=-1; | |
| /* the digits */ | |
| do { | |
| c[nd++]=(v%10)*s+'0'; | |
| v=v/10; | |
| } while (v != 0); | |
| /* print the minus */ | |
| if (s < 0 ) c[nd]='-'; else nd--; | |
| /* reverse the order of digits */ | |
| i=0; | |
| j=nd; | |
| while (j>i) { | |
| c1=c[i]; | |
| c[i]=c[j]; | |
| c[j]=c1; | |
| i++; | |
| j--; | |
| } | |
| nd++; | |
| c[nd]=0; | |
| return nd; | |
| } | |
| /* | |
| * this is for floats, handling output without library | |
| * functions as well. | |
| */ | |
| address_t tinydtostrf(number_t v, index_t p, char* c) { | |
| index_t i; | |
| address_t nd = 0; | |
| number_t f; | |
| /* if we are in forced integer mode */ | |
| if (forceint) { | |
| v=trunc(v); | |
| return writenumber(c, (int)v); | |
| } | |
| /* we do the sign here and don't rely on writenumbers sign handling, guess why */ | |
| if (v<0) { | |
| v=fabs(v); | |
| c[nd++]='-'; | |
| } | |
| /* write the integer part */ | |
| nd+=writenumber(c+nd, (int)v); | |
| c[nd++]='.'; | |
| /* only the fraction to precision p */ | |
| f=fabs(v); | |
| /* get p digits of the fraction */ | |
| for (i=p; i>0; i--) { | |
| f=f-floor(f); | |
| f=f*10; | |
| c[nd++]=(int)floor(f)+'0'; | |
| } | |
| /* and a terminating 0 */ | |
| c[nd]=0; | |
| return nd; | |
| } | |
| address_t writenumber2(char *c, number_t vi) { | |
| index_t i; | |
| index_t nd; | |
| number_t f; | |
| index_t exponent = 0; | |
| mem_t eflag=0; | |
| const int p = 5; | |
| /* pseudo integers are displayed as integer | |
| zero trapped here */ | |
| f=floor(vi); | |
| if (f == vi && fabs(vi) < maxnum) return writenumber(c, vi); | |
| /* earlier, floats where displayed in POSIx using the libraties | |
| * return sprintf(c, "%g", vi); | |
| * we dont do this any more | |
| */ | |
| /* we check if we have anything to write */ | |
| if (!isfinite(vi)) { | |
| c[0]='*'; | |
| c[1]=0; | |
| return 1; | |
| } | |
| /* normalize the number and see which exponent we have to deal with */ | |
| f=vi; | |
| while (fabs(f)<1.0) { f=f*10; exponent--; } | |
| while (fabs(f)>=10.0-0.00001) { f=f/10; exponent++; } | |
| /* there are platforms where dtostrf is broken, we do things by hand in a simple way */ | |
| if (exponent > -2 && exponent < 7) { | |
| tinydtostrf(vi, 5, c); | |
| } else { | |
| tinydtostrf(f, 5, c); | |
| eflag=1; | |
| } | |
| /* remove trailing zeros */ | |
| for (i=0; (i < SBUFSIZE && c[i] !=0 ); i++); | |
| i--; | |
| /* */ | |
| while (c[i] == '0' && i>1) { i--; } | |
| i++; | |
| /* add the exponent */ | |
| if (eflag && exponent != 0) { | |
| c[i++]='E'; | |
| i+=writenumber(c+i, exponent); | |
| } | |
| c[i]=0; | |
| return i; | |
| } | |
| /* | |
| * innumber is used as a helper only by xinput(). It reads a number from an input | |
| * buffer and returns it in the number_t r. It returns 0 if the number is invalid, 1 if | |
| * the number is valid, and -1 if the user has pressed the BREAKCHAR key. | |
| * | |
| * Unlike the old innumber() implementation it is meant to read comma separated numbers | |
| * through input. Handling of the buffer is done by the calling function. | |
| * | |
| * The TINYBASICINPUT is an alternative implementation following the Palo Alto BASIC | |
| * way of doing it. The expression parser is reused. This allows variables names and | |
| * expressions to be used as input as well. | |
| * Due to the reuse of sbuffer in the calling function, this will probably only work | |
| * in RUN mode. | |
| */ | |
| int innumber(number_t *r, char* buffer, address_t k) { | |
| address_t i = k; | |
| mem_t s = 1; | |
| /* result is zero*/ | |
| *r=0; | |
| /* remove all leading whitespaces first */ | |
| while ((buffer[i] == ' ' || buffer[i] == '\t') && i <= (address_t) buffer[0]) i++; | |
| /* is there anything left */ | |
| if (i > (address_t) buffer[0]) return 0; | |
| /* now the sign */ | |
| if (buffer[i] == '-') { s=-1; i++; } | |
| /* check for the break character */ | |
| if (buffer[i] == BREAKCHAR) return -1; | |
| /* the number */ | |
| if (buffer[i] < '0' || buffer[i] > '9') return 0; | |
| i+=parsenumber(&buffer[i], r); | |
| if ((buffer[i] < '0' || buffer[i] > '9') && buffer[i] != '.') return 0; | |
| i+=parsenumber2(&buffer[i], r); | |
| /* the sign */ | |
| *r*=s; | |
| return i; | |
| char *b; | |
| token_t t; | |
| /* result is zero */ | |
| *r=0; | |
| /* save the interpreter state */ | |
| b=bi; | |
| s=st; | |
| t=token; | |
| /* switch to fake interactive with the buffer as input */ | |
| st=SINT; | |
| bi=buffer+k; | |
| /* BREAK handling */ | |
| if (*bi == BREAKCHAR) { | |
| return -1; | |
| } | |
| /* start to interpret the buffer as an expression */ | |
| nexttoken(); | |
| expression(); | |
| /* restore the interpreter state */ | |
| i=bi-buffer-1; | |
| bi=b; | |
| st=s; | |
| token=t; | |
| /* error handling, we trap the error and return zero */ | |
| if (er) { | |
| er=0; | |
| return 0; | |
| } | |
| /* the result is on the stack */ | |
| *r=pop(); | |
| return i; | |
| } | |
| /* prints a number */ | |
| void outnumber(number_t n){ | |
| address_t nd, i; | |
| /* number write to sbuffer, remember the number of digits in nd */ | |
| nd=writenumber(sbuffer, n); | |
| nd=writenumber2(sbuffer, n); | |
| /* negative number format aligns right */ | |
| if (form < 0) { i=nd; while(i < -form) {outspc(); i++;} } | |
| /* the number */ | |
| outs(sbuffer, nd); | |
| /* number formats in Palo Alto style, positive numbers align left */ | |
| if (form > 0) { while (nd < form) {outspc(); nd++;} } | |
| } | |
| /* | |
| * Layer 1 functions - providing data into the global variable and | |
| * changing the interpreter state | |
| */ | |
| /* | |
| * Lexical analyser - tokenizes the input line. | |
| * | |
| * nexttoken() increments the input buffer index bi and delivers values in the global | |
| * variable token, with arguments in the accumulators ax, x and the index register ir | |
| * name is used in the routine. | |
| * | |
| * name, sr, ax and x change values in nexttoken and deliver the result to the calling | |
| * function. | |
| * | |
| * bi and ibuffer should not be changed or used for any other function in | |
| * interactive node as they contain the state of nexttoken(). In run mode | |
| * bi and ibuffer are not used as the program is fully tokenized in mem. | |
| * | |
| * all this is pretty much stateless. | |
| */ | |
| /* skip whitespaces */ | |
| void whitespaces(){ | |
| while (*bi == ' ' || *bi == '\t') bi++; | |
| } | |
| /* upper case, don't trust the buildins on microcontrollers */ | |
| char btoupper(char c) { | |
| if (c >= 'a' && c <= 'z') return c-32; else return c; | |
| } | |
| /* the token stream */ | |
| void nexttoken() { | |
| address_t k, l, i; | |
| char* ir; | |
| char quotechar; | |
| /* RUN mode vs. INT mode, in RUN mode we read from mem via gettoken() */ | |
| if (st == SRUN || st == SERUN) { | |
| /* in the token stream we call fastticker - all fast timing functions are in stream */ | |
| fastticker(); | |
| /* read the token from memory */ | |
| gettoken(); | |
| /* show what we are doing */ | |
| if (debuglevel>1) { debugtoken(); outcr(); } | |
| return; | |
| } | |
| /* after change in buffer logic the first byte is reserved for the length */ | |
| if (bi == ibuffer) bi++; | |
| /* literal mode only(!) EOL ends literal mode, used to have REM without quotes */ | |
| if (lexliteral) { | |
| token=*bi; | |
| if (*bi != '\0') bi++; else lexliteral=0; | |
| return; | |
| } | |
| /* remove whitespaces outside strings */ | |
| whitespaces(); | |
| /* end of line token */ | |
| if (*bi == '\0') { | |
| token=EOL; | |
| if (DEBUG) debugtoken(); | |
| return; | |
| } | |
| /* unsigned numbers, value returned in x */ | |
| if (*bi <='9' && *bi >= '0') { | |
| bi+=parsenumber(bi, &x); | |
| if ((*bi <='9' && *bi >= '0') || *bi == '.') { | |
| bi+=parsenumber2(bi, &x); | |
| token=NUMBER; | |
| if (DEBUG) debugtoken(); | |
| return; | |
| } | |
| /* strings between " " or " EOL, value returned in ir and in sr for now */ | |
| if (*bi == '"' || *bi == '\'') { | |
| quotechar=*bi; | |
| k=0; | |
| bi++; | |
| ir=bi; | |
| sr.ir=bi; | |
| while(*bi != quotechar && *bi !='\0') { | |
| k++; | |
| bi++; | |
| } | |
| bi++; | |
| token=STRING; | |
| sr.length=k; | |
| sr.address=0; /* we don't find the string in BASIC memory, as we lex from bi */ | |
| if (DEBUG) debugtoken(); | |
| return; | |
| } | |
| /* | |
| * relations | |
| * single character relations are their own token | |
| * >=, =<, =<, =>, <> are tokenized | |
| */ | |
| if (*bi == '=') { | |
| bi++; | |
| whitespaces(); | |
| if (*bi == '>') { | |
| token=GREATEREQUAL; | |
| bi++; | |
| } else if (*bi == '<') { | |
| token=LESSEREQUAL; | |
| bi++; | |
| } else { | |
| token='='; | |
| } | |
| if (DEBUG) debugtoken(); | |
| return; | |
| } | |
| if (*bi == '>'){ | |
| bi++; | |
| whitespaces(); | |
| if (*bi == '=') { | |
| token=GREATEREQUAL; | |
| bi++; | |
| } else { | |
| token='>'; | |
| } | |
| if (DEBUG) debugtoken(); | |
| return; | |
| } | |
| if (*bi == '<'){ | |
| bi++; | |
| whitespaces(); | |
| if (*bi == '=') { | |
| token=LESSEREQUAL; | |
| bi++; | |
| } else if (*bi == '>') { | |
| token=NOTEQUAL; | |
| bi++; | |
| } else { | |
| token='<'; | |
| } | |
| if (DEBUG) debugtoken(); | |
| return; | |
| } | |
| /* | |
| * Keyworks and variables | |
| * | |
| * Isolate a word, bi points to the beginning, l is the length of the word. | |
| * ir points to the end of the word after isolating. | |
| * @ is a letter here to make the special @ arrays possible. | |
| */ | |
| l=0; | |
| ir=bi; | |
| while (-1) { | |
| if (*ir >= 'a' && *ir <= 'z') { | |
| if (!lowercasenames) *ir-=32; /* toupper code, changing the input buffer directly */ | |
| ir++; | |
| l++; | |
| } else if ((*ir >= '@' && *ir <= 'Z') || *ir == '_') { | |
| ir++; | |
| l++; | |
| } else { | |
| break; | |
| } | |
| } | |
| /* | |
| * Ir is reused here to implement string compares | |
| * scanning the keyword array. | |
| * Once a keyword is detected the input buffer is advanced | |
| * by its length, and the token value is returned. | |
| * | |
| * Keywords are an array of null terminated strings. | |
| * They are always matched uppercase. | |
| */ | |
| k=0; | |
| while (gettokenvalue(k) != 0) { | |
| ir=getkeyword(k); | |
| i=0; | |
| while (*(ir+i) != 0) { | |
| if (*(ir+i) != btoupper(*(bi+i))) { | |
| k++; | |
| i=0; | |
| break; | |
| } else | |
| i++; | |
| } | |
| if (i == 0) continue; | |
| bi+=i; | |
| token=gettokenvalue(k); | |
| if (token == TREM) lexliteral=1; | |
| if (DEBUG) debugtoken(); | |
| return; | |
| } | |
| /* | |
| * A variable has length 2 with either two letters or a letter | |
| * and a number. @ can be the first letter of a variable. | |
| * Here, no tokens can appear any more as they have been processed | |
| * further up. | |
| * | |
| * The longname code supports MAXNAME characters and _ as additional character. | |
| */ | |
| if (l>0 && l<=MAXNAME) { | |
| token=VARIABLE; | |
| zeroname(&name); | |
| while (((*bi >= '0' && *bi <= '9') || | |
| (*bi >= '@' && *bi <= 'Z') || | |
| (*bi >= 'a' && *bi <= 'z') || | |
| (*bi == '_') ) && name.l < MAXNAME && *bi != 0) { | |
| name.c[name.l]=*bi; | |
| bi++; | |
| name.l++; | |
| } | |
| if (*bi == '$') { | |
| token=STRINGVAR; | |
| bi++; | |
| } | |
| whitespaces(); | |
| if (token == VARIABLE && *bi == '(' ) { | |
| token=ARRAYVAR; | |
| } | |
| /* the new code filling the name variable directly, will be used in the entire code soon */ | |
| name.token=token; | |
| if (DEBUG) debugtoken(); | |
| return; | |
| } | |
| if (l == 1 || l == 2) { | |
| token=VARIABLE; | |
| name.l=0; | |
| name.c[0]=*bi; | |
| name.c[1]=0; | |
| bi++; | |
| if ((*bi >= '0' && *bi <= '9') || (*bi >= 'A' && *bi <= 'Z') || *bi == '_' ) { | |
| name.c[1]=*bi; | |
| bi++; | |
| } | |
| if (*bi == '$') { | |
| token=STRINGVAR; | |
| bi++; | |
| } | |
| whitespaces(); | |
| if (token == VARIABLE && *bi == '(' ) { | |
| token=ARRAYVAR; | |
| } | |
| /* the new code filling the name variable directly, will be used in the entire code soon */ | |
| name.token=token; | |
| if (DEBUG) debugtoken(); | |
| return; | |
| } | |
| /* other single characters are parsed and stored */ | |
| token=*bi; | |
| bi++; | |
| if (DEBUG) debugtoken(); | |
| return; | |
| } | |
| /* | |
| * Layer 1 - program editor | |
| * | |
| * Editing the program, the structure of a line is | |
| * LINENUMBER linenumber(2 or more bytes) token(n bytes) | |
| * | |
| * store* stores something to memory | |
| * get* retrieves information | |
| * | |
| * No matter how long number_t is in the C implementation | |
| * we pack into bytes, this is clumsy but portable | |
| * the store and get commands all operate on here | |
| * and advance it | |
| * | |
| * storetoken() operates on the variable top. | |
| * We always append at the end and then sort. | |
| * | |
| * gettoken() operate on the variable here | |
| * which will also be the key variable in run mode. | |
| * | |
| * tokens are stored including their payload. | |
| * | |
| * This group of functions changes global states and | |
| * cannot be called at program runtime with saving | |
| * the relevant global variable to the stack. | |
| */ | |
| /* | |
| * check if we still have memory left, on RAM only systems we make | |
| * sure that we don't hit himem. On systems with EEPROM program storage we make | |
| * sure that we stay within the EEPROM range. | |
| */ | |
| char nomemory(number_t b){ | |
| if (top >= himem-b) return 1; else return 0; | |
| if (top >= elength()-eheadersize-b) return 1; else return 0; | |
| } | |
| /* store a token - check free memory before changing anything */ | |
| void storetoken() { | |
| int i; | |
| switch (token) { | |
| case LINENUMBER: | |
| if (nomemory(addrsize+1)) break; | |
| memwrite2(top++, token); | |
| setaddress(top, memwrite2, ax); | |
| top+=addrsize; | |
| return; | |
| case NUMBER: | |
| if (nomemory(numsize+1)) break; | |
| memwrite2(top++, token); | |
| setnumber(top, memwrite2, x); | |
| top+=numsize; | |
| return; | |
| case ARRAYVAR: | |
| case VARIABLE: | |
| case STRINGVAR: | |
| if (nomemory(sizeof(name_t))) break; | |
| memwrite2(top++, token); | |
| top=setname_pgm(top, &name); | |
| return; | |
| case STRING: | |
| i=sr.length; | |
| if (nomemory(i+2)) break; | |
| memwrite2(top++, token); | |
| memwrite2(top++, i); | |
| while (i > 0) { memwrite2(top++, *sr.ir++); i--; } | |
| return; | |
| default: | |
| if (token >= -127) { /* the good old code with just one byte token */ | |
| if (nomemory(1)) break; | |
| memwrite2(top++, token); | |
| } else { | |
| if (nomemory(2)) break; /* this is the two byte token extension */ | |
| memwrite2(top++, TEXT1); | |
| memwrite2(top++, token+255); | |
| } | |
| return; | |
| } | |
| } | |
| /* | |
| * wrappers around mem access in genereal | |
| * | |
| * memread is used only in the token stream, it reads from a stream | |
| * read only. If run is done from eeprom then bytes are taken from this | |
| * stream, this would also be the place to implement direct run from | |
| * another device like a file system or embedded programs on flash. | |
| * Only the token stream uses memread. | |
| * | |
| * memread2 and memwrite2 always go to ram. They are read/write. Variables and | |
| * the program editor uses these functions. | |
| * | |
| * currently only the SPIRAM and the EEPROM direct edit interface is implemented. | |
| * This is handled in the runtime library. | |
| * | |
| * USEMEMINTERFACE is the macro to control this. | |
| * | |
| * The POSIX code has a test interface for SPIRAM as a dummy. | |
| * | |
| */ | |
| mem_t memread(address_t a) { | |
| if (st != SERUN) { | |
| return mem[a]; | |
| } else { | |
| return eread(a+eheadersize); | |
| } | |
| } | |
| mem_t memread2(address_t a) { return mem[a]; } | |
| void memwrite2(address_t a, mem_t c) { mem[a]=c; } | |
| mem_t memread(address_t a) { | |
| if (st != SERUN) { | |
| return spiram_robufferread(a); | |
| } else { | |
| return eread(a+eheadersize); | |
| } | |
| } | |
| mem_t memread2(address_t a) { return spiram_rwbufferread(a); } | |
| void memwrite2(address_t a, mem_t c) { spiram_rwbufferwrite(a, c); } | |
| mem_t memread(address_t a) { | |
| if (a < elength()-eheadersize) return eread(a+eheadersize); else return mem[a-(elength()-eheadersize)]; | |
| } | |
| mem_t memread2(address_t a) { return memread(a); } | |
| void memwrite2(address_t a, mem_t c) { | |
| if (a < elength()-eheadersize) eupdate(a+eheadersize, c); else mem[a-(elength()-eheadersize)]=c; | |
| } | |
| /* get a token from memory */ | |
| void gettoken() { | |
| stringlength_t i; | |
| /* if we have reached the end of the program, EOL is always returned | |
| we don't rely on mem having a trailing EOL */ | |
| if (here >= top) { | |
| token=EOL; | |
| return; | |
| } | |
| /* if we have no data type we are done reading just one byte */ | |
| token=memread(here++); | |
| name.token=token; | |
| /* if there are multibyte tokens, get the next byte and construct a token value <-127 */ | |
| if (token == TEXT1) { | |
| token=memread(here++)-255; | |
| } | |
| /* otherwise we check for the argument */ | |
| switch (token) { | |
| case LINENUMBER: | |
| ax=getaddress(here, memread); | |
| here+=addrsize; | |
| break; | |
| case NUMBER: | |
| x=getnumber(here, memread); | |
| here+=numsize; | |
| break; | |
| case ARRAYVAR: | |
| case VARIABLE: | |
| case STRINGVAR: | |
| here=getname(here, &name, memread); | |
| name.token=token; | |
| break; | |
| case STRING: | |
| sr.length=(unsigned char)memread(here++); | |
| /* | |
| * if we run from EEPROM, the input buffer is used to get string constants. | |
| * if we run on a system with real memory, we produce a mem pointer | |
| * otherwise the caller has to handle strings through the address (SPIRAM systems) | |
| */ | |
| if (st == SERUN) { | |
| for(i=0; i<sr.length; i++) ibuffer[i]=memread(here+i); | |
| sr.ir=ibuffer; | |
| } else { | |
| sr.ir=(char*)&mem[here]; | |
| sr.ir=0; | |
| } | |
| sr.address=here; | |
| here+=sr.length; | |
| } | |
| } | |
| /* goto the first line of a program */ | |
| void firstline() { | |
| if (top == 0) { | |
| ax=0; | |
| return; | |
| } | |
| here=0; | |
| gettoken(); | |
| } | |
| /* goto the next line, search forward */ | |
| void nextline() { | |
| while (here < top) { | |
| gettoken(); | |
| if (token == LINENUMBER) return; | |
| if (here >= top) { | |
| here=top; | |
| ax=0; | |
| return; | |
| } | |
| } | |
| } | |
| /* | |
| * the line cache mechanism, useful for large codes. | |
| * addlinecache does not test if the line already exist because it | |
| * assumes that findline calls it only if a new line is to be stored | |
| * the LINECACHE size depends on the architecture. | |
| */ | |
| const unsigned char linecachedepth = LINECACHESIZE; | |
| typedef struct {address_t l; address_t h;} linecacheentry; | |
| linecacheentry linecache[LINECACHESIZE]; | |
| unsigned char linecachehere = 0; | |
| void clrlinecache() { | |
| unsigned char i; | |
| for(i=0; i<linecachedepth; i++) linecache[i].l=linecache[i].h=0; | |
| linecachehere=0; | |
| } | |
| void addlinecache(address_t l, address_t h) { | |
| linecache[linecachehere].l=l; | |
| linecache[linecachehere].h=h; | |
| linecachehere=(linecachehere+1)%linecachedepth; | |
| } | |
| address_t findinlinecache(address_t l){ | |
| unsigned char i; | |
| for(i=0; i<linecachedepth && linecache[i].l != 0; i++) { | |
| if (linecache[i].l == l) return linecache[i].h; | |
| } | |
| return 0; | |
| } | |
| void clrlinecache() {} | |
| void addlinecache(address_t l, address_t h) {} | |
| address_t findinlinecache(address_t l){ return 0; } | |
| /* find a line, look in cache then search from the beginning | |
| * x is used as the valid line number once a line is found | |
| * hence x must be global | |
| * (this is the logic of the gettoken mechanism) | |
| */ | |
| void findline(address_t l) { | |
| address_t a; | |
| /* we know it already, here to advance */ | |
| if ((a=findinlinecache(l))) { | |
| here=a; | |
| token=LINENUMBER; | |
| ax=l; | |
| return; | |
| } | |
| /* we need to search */ | |
| here=0; | |
| while (here < top) { | |
| gettoken(); | |
| if (token == LINENUMBER && ax == l ) { | |
| /* now that we know we cache */ | |
| addlinecache(l, here); | |
| return; | |
| } | |
| } | |
| error(ELINE); | |
| } | |
| /* finds the line of a location */ | |
| address_t myline(address_t h) { | |
| address_t l=0; | |
| address_t l1=0; | |
| address_t here2; | |
| here2=here; | |
| here=0; | |
| gettoken(); | |
| while (here < top) { | |
| if (token == LINENUMBER) { | |
| l1=l; | |
| l=ax; | |
| } | |
| if (here >= h) break; | |
| gettoken(); | |
| } | |
| here=here2; | |
| if (token == LINENUMBER) | |
| return l1; | |
| else | |
| return l; | |
| } | |
| /* | |
| * Move a block of storage beginng at b ending at e | |
| * to destination d. No error handling here!! | |
| */ | |
| void moveblock(address_t b, address_t l, address_t d) { | |
| address_t i; | |
| if (d+l > himem) { | |
| error(EOUTOFMEMORY); | |
| return; | |
| } | |
| if (l<1) return; | |
| if (b < d) for (i=l; i>0; i--) memwrite2(d+i-1, memread2(b+i-1)); | |
| else for (i=0; i<l; i++) memwrite2(d+i, memread2(b+i)); | |
| } | |
| /* zero a block of memory */ | |
| void zeroblock(address_t b, address_t l){ | |
| address_t i; | |
| if (b+l > himem) { error(EOUTOFMEMORY); return; } | |
| if (l<1) return; | |
| for (i=0; i<l+1; i++) memwrite2(b+i, 0); | |
| } | |
| /* | |
| * Line editor: | |
| * | |
| * stage 1: no matter what the line number is - store at the top | |
| * remember the location in here. | |
| * stage 2: see if it is only an empty line - try to delete this line | |
| * stage 3: calculate lengthes and free memory and make room at the | |
| * appropriate place | |
| * stage 4: copy to the right place | |
| * | |
| * Very fragile code, con't change if you don't have to | |
| * | |
| * zeroblock statements commented out after EOL code was fixed | |
| */ | |
| /* diagnosis function */ | |
| void diag(){ | |
| outsc("top, here, y and x\n"); | |
| outnumber(top); outspc(); | |
| outnumber(here); outspc(); | |
| outcr(); | |
| } | |
| void storeline() { | |
| const index_t lnlength=addrsize+1; | |
| index_t linelength; | |
| number_t newline; | |
| address_t here2, here3; | |
| address_t t1, t2; | |
| address_t y; | |
| /* the data pointers becomes invalid once the code has been changed */ | |
| clrdata(); | |
| /* line cache is invalid on line storage */ | |
| clrlinecache(); | |
| if (DEBUG) { outsc("storeline "); outnumber(ax); outsc(" : "); outsc(ibuffer); outcr(); } | |
| /* | |
| * stage 1: append the line at the end of the memory, | |
| * remember the line number on the stack and the old top in here | |
| */ | |
| t1=ax; | |
| here=top; | |
| newline=here; | |
| token=LINENUMBER; | |
| do { | |
| storetoken(); | |
| if (er != 0 ) { | |
| top=newline; | |
| here=0; | |
| return; | |
| } | |
| nexttoken(); | |
| } while (token != EOL); | |
| ax=t1; /* recall the line number */ | |
| linelength=top-here; /* calculate the number of stored bytes */ | |
| /* | |
| * stage 2: check if only a linenumber stored - then delete this line | |
| */ | |
| if (linelength == (lnlength)) { | |
| top-=(lnlength); | |
| findline(ax); | |
| if (er) return; | |
| y=here-lnlength; | |
| nextline(); | |
| here-=lnlength; | |
| if (ax != 0) { | |
| moveblock(here, top-here, y); | |
| top=top-(here-y); | |
| } else { | |
| top=y; | |
| } | |
| return; | |
| } | |
| /* | |
| * stage 3, a nontrivial line with linenumber x is to be stored | |
| * try to find it first by walking through all lines | |
| */ | |
| else { | |
| y=ax; | |
| here2=here; | |
| here=lnlength; | |
| nextline(); | |
| /* there is no nextline after the first line, we are done */ | |
| if (ax == 0) return; | |
| /* go back to the beginning */ | |
| here=0; | |
| here2=0; | |
| while (here < top) { | |
| here3=here2; | |
| here2=here; | |
| nextline(); | |
| if (ax > y) break; | |
| } | |
| /* | |
| * at this point y contains the number of the line to be inserted | |
| * x contains the number of the first line with a higher line number | |
| * or 0 if the line is to be inserted at the end | |
| * here points to the following line and here2 points to the previous line | |
| */ | |
| if (ax == 0) { | |
| here=here3-lnlength; | |
| gettoken(); | |
| if (token == LINENUMBER && ax == y) { | |
| here2-=lnlength; | |
| here-=lnlength; | |
| moveblock(here2, linelength, here); | |
| top=here+linelength; | |
| } | |
| return; | |
| } | |
| here-=lnlength; | |
| t1=here; | |
| here=here2-lnlength; | |
| t2=here; | |
| gettoken(); | |
| if (ax == y) { /* the line already exists and has to be replaced */ | |
| here2=t2; /* this is the line we are dealing with */ | |
| here=t1; /* this is the next line */ | |
| y=here-here2; /* the length of the line as it is */ | |
| if (linelength == y) { /* no change in line length */ | |
| moveblock(top-linelength, linelength, here2); | |
| top=top-linelength; | |
| } else if (linelength > y) { /* the new line is longer than the old one */ | |
| moveblock(here, top-here, here+linelength-y); | |
| here=here+linelength-y; | |
| top=top+linelength-y; | |
| moveblock(top-linelength, linelength, here2); | |
| top=top-linelength; | |
| } else { /* the new line is short than the old one */ | |
| moveblock(top-linelength, linelength, here2); | |
| top=top-linelength; | |
| moveblock(here, top-here, here2+linelength); | |
| top=top-y+linelength; | |
| } | |
| } else { /* the line has to be inserted in between */ | |
| here=t1; | |
| moveblock(here, top-here, here+linelength); | |
| moveblock(top, linelength, here); | |
| } | |
| } | |
| } | |
| /* | |
| * Layer 1 - the code in this section calculates an expression | |
| * with a recursive descent algorithm | |
| * | |
| * all function use the stack to pass values back. We use the | |
| * Backus-Naur form of basic from here https: | |
| * implementing a C style logical expression model | |
| */ | |
| /* the terminal symbol it ends a statement list - ELSE is one too as it ends a statement list */ | |
| char termsymbol() { | |
| return (token == LINENUMBER || token == ':' || token == EOL || token == TELSE); | |
| } | |
| /* a little helpers - one token expect */ | |
| char expect(token_t t, mem_t e) { | |
| nexttoken(); | |
| if (token != t) {error(e); return 0; } else return 1; | |
| } | |
| /* a little helpers - expression expect */ | |
| char expectexpr() { | |
| nexttoken(); | |
| expression(); | |
| if (er != 0) return 0; else return 1; | |
| } | |
| /* parses a list of expression, this may be recursive! */ | |
| void parsearguments() { | |
| short argsl; | |
| /* begin counting */ | |
| argsl=0; | |
| /* having 0 args at the end of a command is legal */ | |
| if (!termsymbol()) { | |
| /* list of expressions separated by commas */ | |
| do { | |
| expression(); | |
| if (er != 0) break; | |
| argsl++; | |
| if (token == ',') nexttoken(); else break; | |
| } while (1); | |
| } | |
| /* because of the recursion ... */ | |
| args=argsl; | |
| } | |
| /* expect exactly n arguments */ | |
| void parsenarguments(char n) { | |
| parsearguments(); | |
| if (args != n) error(EARGS); | |
| } | |
| /* counts and parses the number of arguments given in brakets, this function | |
| should not advance to the next token because it is called in factor */ | |
| void parsesubscripts() { | |
| blocation_t l; | |
| args=0; | |
| if (DEBUG) { | |
| outsc("** in parsesubscripts "); outcr(); | |
| bdebug("token "); | |
| } | |
| /* remember where we where */ | |
| pushlocation(&l); | |
| /* parsesubscripts is called directly after the object in question, it does | |
| nexttoken() itself now */ | |
| nexttoken(); | |
| /* if we have no bracket here, we return with zero */ | |
| if (token != '(') { | |
| poplocation(&l); | |
| return; | |
| } | |
| nexttoken(); | |
| /* if () we return also with -1 */ | |
| if (token == ')') { | |
| args=-1; | |
| return; | |
| } | |
| /* now we are ready to parse a set of arguments */ | |
| parsearguments(); | |
| if (!USELONGJUMP && er) return; | |
| if (token != ')') { error(EARGS); return; } /* we return with ) as a last token on success */ | |
| /* we end with the ) still as active token */ | |
| } | |
| /* parse a function argument ae is the number of | |
| expected expressions in the argument list, parsesubscripts | |
| should not avance precisely because it is used in factor */ | |
| void parsefunction(void (*f)(), short ae){ | |
| parsesubscripts(); | |
| if (!USELONGJUMP && er) return; | |
| if (args == ae) f(); else error(EARGS); | |
| } | |
| /* helper function in the recursive decent parser */ | |
| void parseoperator(void (*f)()) { | |
| mem_t u=1; | |
| nexttoken(); | |
| /* unary minuses in front of an operator are consumed once! */ | |
| if (token == '-') { | |
| u=-1; | |
| nexttoken(); | |
| } | |
| /* the operator */ | |
| f(); | |
| if (er !=0 ) return; | |
| y=pop(); | |
| if (u == -1) y=-y; | |
| x=pop(); | |
| } | |
| /* | |
| * ABS absolute value | |
| */ | |
| void xabs(){ | |
| number_t x; | |
| if ((x=pop())<0) { x=-x; } | |
| push(x); | |
| } | |
| /* | |
| * SGN evaluates the sign | |
| */ | |
| void xsgn(){ | |
| number_t x; | |
| x=pop(); | |
| if (x>0) x=1; | |
| if (x<0) x=-1; | |
| push(x); | |
| } | |
| /* | |
| * PEEK on an arduino, negative values of peek address | |
| * the EEPROM range -1 .. -1024 on an UNO | |
| */ | |
| void xpeek(){ | |
| number_t a; | |
| /* get the argument from the stack because this is a function only */ | |
| a=pop(); | |
| /* the memory and EEPROM range */ | |
| if (a >= 0 && a<=memsize) | |
| push(memread2(a)); | |
| else if (a < 0 && -a <= elength()) | |
| push(eread(-a-1)); | |
| else { | |
| error(EORANGE); | |
| return; | |
| } | |
| } | |
| /* | |
| * MAP Arduino map function, we always cast to long, this | |
| * makes it potable for various integer sizes and the float. | |
| */ | |
| void xmap() { | |
| long v, in_min, in_max, out_min, out_max; | |
| out_max=pop(); | |
| out_min=pop(); | |
| in_max=pop(); | |
| in_min=pop(); | |
| v=pop(); | |
| push((v - in_min) * (out_max - out_min) / (in_max - in_min) + out_min); | |
| } | |
| /* | |
| * RND very basic random number generator with constant seed in 16 bit | |
| * for float systems, use glibc parameters https: | |
| */ | |
| void xrnd() { | |
| number_t r; | |
| mem_t base = randombase; | |
| /* the argument of the RND() function */ | |
| r=pop(); | |
| /* this is the microsoft mode, argument <0 resets the sequence, 0 always the same number, > 1 a number between 0 and 1 */ | |
| if (randombase < 0) { | |
| base=0; | |
| if (r < 0) { | |
| rd=-r; | |
| r=1; | |
| } else if (r == 0) { | |
| r=1; | |
| goto pushresult; | |
| } else { | |
| r=1; | |
| } | |
| } | |
| /* this is the congruence */ | |
| /* the original 16 bit congruence, the & is needed to make it work for all kinds of ints */ | |
| rd = (31421*rd + 6927) & 0xffff; | |
| /* glibc parameters */ | |
| rd = (110351245*rd + 12345) & 0x7fffffff; | |
| pushresult: | |
| /* the result is calculated with the right modulus */ | |
| if (r>=0) | |
| push((unsigned long)rd*r/0x10000+base); | |
| else | |
| push((unsigned long)rd*r/0x10000+1-base); | |
| if (r>=0) | |
| push(rd*r/0x80000000+base); | |
| else | |
| push(rd*r/0x80000000+1-base); | |
| } | |
| /* | |
| * SQR - a very simple approximate square root formula | |
| * for integers, for floats we use the library | |
| */ | |
| void sqr(){ | |
| number_t t,r; | |
| number_t l=0; | |
| r=pop(); | |
| t=r; | |
| while (t > 0) { | |
| t>>=1; | |
| l++; | |
| } | |
| l=l/2; | |
| t=1; | |
| t<<=l; | |
| do { | |
| l=t; | |
| t=(t+r/t)/2; | |
| } while (abs(t-l)>1); | |
| push(t); | |
| } | |
| void sqr(){ | |
| push(sqrt(pop())); | |
| } | |
| /* | |
| * POW(X, N) evaluates powers | |
| */ | |
| /* this function is called by POW(a, c) in BASIC */ | |
| void xpow(){ | |
| number_t n; | |
| number_t a; | |
| n=pop(); | |
| a=pop(); | |
| push(bpow(a, n)); | |
| } | |
| /* while this is needed by ^*/ | |
| number_t bpow(number_t x, number_t y) { | |
| return pow(x, y); | |
| number_t r; | |
| address_t i; | |
| r=1; | |
| if (y>=0) for(i=0; i<y; i++) r*=x; | |
| else r=0; | |
| return r; | |
| } | |
| /* | |
| * the reimplementation of parsestringvar, this code uses a lhsobject | |
| * to store the data in, which is somehow more natural with the new code. | |
| * | |
| * The need to handle substring and no substring situations makes thing | |
| * complex. | |
| * | |
| * A plain string has the format A$. | |
| * A substringed string has the format A$(i) or A$(i,j). | |
| * A string array has the format A$()(i2), A$(i)(i2), A$(i,j)(i2). | |
| * In the Microsoft world with substringmode==0 we only have | |
| * A$, A$(i2). | |
| * | |
| * All this is distiguished here. | |
| * | |
| */ | |
| void parsestringvar(string_t* strp, lhsobject_t* lhs) { | |
| blocation_t l; | |
| address_t temp; | |
| /* remember the variable name and prep the indices */ | |
| lhs->name=name; | |
| lhs->i=1; /* we start at 1 */ | |
| lhs->j=arraylimit; /* we assume a string array of length 1, all simple strings are like this */ | |
| lhs->i2=0; /* we want the full string length */ | |
| lhs->ps=1; /* we deal with a pure string */ | |
| /* the array index default, can vary */ | |
| /* array_index=arraylimit; this is i2 */ | |
| /* remember the location */ | |
| pushlocation(&l); | |
| /* and inspect the (first) brackets */ | |
| parsesubscripts(); | |
| if (!USELONGJUMP && er) return; | |
| if (DEBUG) { outsc("** in parsestringvar "); outnumber(args); outsc(" arguments \n"); } | |
| /* do we deal with a pure unindexed string or something more complicated */ | |
| if (args == 0) { | |
| /* pure string, we rewind and are done here */ | |
| poplocation(&l); | |
| } else if (!substringmode) { | |
| /* we have no substring interpretation hence the brackets can only be one array index */ | |
| if (args == 1) lhs->j=pop(); else { error(EORANGE); return; } | |
| } else { | |
| /* not a pure string */ | |
| lhs->ps=0; | |
| /* we are in the substring world here */ | |
| if (args == 2) { lhs->i2=popaddress(); args--; } /* A$(i,j) */ | |
| if (!USELONGJUMP && er) return; | |
| if (args == 1) { lhs->i=popaddress(); } /* A$(i) */ | |
| if (!USELONGJUMP && er) return; | |
| if (args == -1) {} /* A$(), ignore */ | |
| /* here we have parsed a full substring and remember where we are and look forward*/ | |
| pushlocation(&l); | |
| nexttoken(); | |
| if (token == '(') { | |
| /* a second pair of braces is coming, we parse an array index */ | |
| nexttoken(); | |
| expression(); | |
| if (!USELONGJUMP && er) return; | |
| if (token != ')') { error(EUNKNOWN); return; } | |
| lhs->j=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| } else | |
| poplocation(&l); | |
| } | |
| /* in pure parse mode we end here. This is used for the lefthandside code */ | |
| if (!strp) return; | |
| /* try to get the string */ | |
| getstring(strp, &lhs->name, lhs->i, lhs->j); | |
| if (!USELONGJUMP && er) return; | |
| /* look what we do with the upper index */ | |
| if (!lhs->i2) lhs->i2=strp->length; | |
| if (DEBUG) { | |
| outsc("** in parsestringvar lower is "); outnumber(lhs->i); outcr(); | |
| outsc("** in parsestringvar upper is "); outnumber(lhs->i2); outcr(); | |
| outsc("** in parsestringvar array_index is "); outnumber(lhs->j); outcr(); | |
| } | |
| /* find the length */ | |
| if (lhs->i2-lhs->i+1 > 0) strp->length=lhs->i2-lhs->i+1; else strp->length=0; | |
| /* done */ | |
| if (DEBUG) { | |
| outsc("** in parsestringvar, length "); | |
| outnumber(strp->length); | |
| outsc(" from "); outnumber(lhs->i); outspc(); outnumber(lhs->i2); | |
| outcr(); | |
| } | |
| /* restore the name */ | |
| name=lhs->name; | |
| return; | |
| } | |
| /* | |
| * stringvalue(string_t*) evaluates a string value, return 0 if there is no string, | |
| * 1 if there is a string. The pointer contains all the data needed to process the string. | |
| * | |
| */ | |
| char stringvalue(string_t* strp) { | |
| address_t k, l; | |
| address_t i; | |
| token_t t; | |
| mem_t args=1; | |
| lhsobject_t lhs; | |
| if (DEBUG) outsc("** entering stringvalue \n"); | |
| /* make sure everything is nice and clean */ | |
| strp->address=0; | |
| strp->arraydim=0; | |
| strp->length=0; | |
| strp->strdim=0; | |
| strp->ir=0; | |
| switch(token) { | |
| case STRING: | |
| /* sr has the string information from gettoken and nexttoken*/ | |
| strp->ir=sr.ir; | |
| strp->length=sr.length; | |
| strp->address=sr.address; | |
| break; | |
| case STRINGVAR: | |
| parsestringvar(strp, &lhs); | |
| break; | |
| case TSTR: | |
| nexttoken(); | |
| if (token == '$') nexttoken(); | |
| if (token != '(') { error(EARGS); return 0; } | |
| nexttoken(); | |
| expression(); | |
| if (er != 0) return 0; | |
| strp->length=writenumber2(sbuffer, pop()); | |
| strp->length=writenumber(sbuffer, pop()); | |
| strp->ir=sbuffer; | |
| if (er != 0) return 0; | |
| if (token != ')') {error(EARGS); return 0; } | |
| break; | |
| case TCHR: | |
| nexttoken(); | |
| if (token == '$') nexttoken(); | |
| if (token != '(') { error(EARGS); return 0; } | |
| nexttoken(); | |
| expression(); | |
| if (er != 0) return 0; | |
| *sbuffer=pop(); | |
| strp->ir=sbuffer; | |
| strp->length=1; | |
| if (token != ')') {error(EARGS); return 0; } | |
| break; | |
| case TRIGHT: | |
| case TMID: | |
| case TLEFT: | |
| t=token; | |
| nexttoken(); | |
| if (token == '$') nexttoken(); | |
| if (token != '(') { error(EARGS); return 0; } | |
| nexttoken(); | |
| if (token != STRINGVAR) { error(EARGS); return 0; } | |
| parsestringvar(strp, &lhs); | |
| if (er != 0) return 0; | |
| k=strp->length; /* the length of the original string variable */ | |
| nexttoken(); | |
| if (token != ',') { error(EARGS); return 0; } | |
| nexttoken(); | |
| expression(); | |
| if (er != 0) return 0; | |
| /* all the rest depends on the function */ | |
| switch (t) { | |
| case TRIGHT: | |
| l=popaddress(); | |
| if (k < l) l=k; | |
| if (strp->address) strp->address=strp->address+(k-l); | |
| if (strp->ir) strp->ir=strp->ir+(k-l); | |
| break; | |
| case TLEFT: | |
| l=popaddress(); | |
| if (k < l) l=k; | |
| break; | |
| case TMID: | |
| if (token == ',') { | |
| nexttoken(); | |
| expression(); | |
| if (er != 0) return 0; | |
| args++; | |
| } | |
| if (args == 1) { | |
| i=popaddress(); | |
| l=0; | |
| if (i <= k) l=k-i+1; | |
| } else { | |
| l=popaddress(); | |
| if (er != 0) return 0; | |
| i=popaddress(); | |
| } | |
| if (k < i+l) l=k-i+1; | |
| if (l < 0) l=0; | |
| if (strp->address != 0) strp->address=strp->address+i-1; | |
| if (strp->ir) strp->ir=strp->ir+i-1;; | |
| break; | |
| } | |
| strp->length=l; | |
| if (token != ')') {error(EARGS); return 0; } | |
| break; | |
| default: | |
| return 0; | |
| } | |
| return 1; | |
| } | |
| /* | |
| * (numerical) evaluation of a string expression, used for | |
| * comparison and for string rightvalues as numbers | |
| * the token rewind here is needed as streval is called in | |
| * factor - no factor function should nexttoken | |
| */ | |
| void streval(){ | |
| token_t t; | |
| address_t k; | |
| string_t s1, s2; | |
| char* ir; | |
| address_t a; | |
| blocation_t l; | |
| /* is the right side of the expression a string */ | |
| if (!stringvalue(&s1)) { error(EUNKNOWN); return; } | |
| if (!USELONGJUMP && er) return; | |
| if (DEBUG) { outsc("** in streval first string"); outcr(); } | |
| /* get ready for rewind to this location */ | |
| pushlocation(&l); | |
| t=token; | |
| /* is the next token a operator, hence we need to compare two strings? */ | |
| nexttoken(); | |
| if (token != '=' && token != NOTEQUAL) { | |
| /* if not, rewind one token and evaluate the string as a boolean */ | |
| poplocation(&l); | |
| token=t; | |
| /* a zero length string evaluates to zero else to the first character */ | |
| if (s1.length == 0) push(0); else { | |
| if (s1.ir) push(s1.ir[0]); | |
| else if (s1.address) push(memread2(s1.address)); | |
| else error(EGENERAL); | |
| } | |
| return; | |
| } | |
| /* remember which operator we use */ | |
| t=token; | |
| /* questionable !! */ | |
| nexttoken(); | |
| if (DEBUG) { | |
| outsc("** in streval second string"); outcr(); | |
| debugtoken(); outcr(); | |
| } | |
| /* get the second string */ | |
| if (!stringvalue(&s2)){ error(EUNKNOWN); return; } | |
| if (!USELONGJUMP && er) return; | |
| if (DEBUG) { outsc("** in streval result: "); outnumber(x); outcr(); } | |
| /* different length means unequal */ | |
| if (s2.length != s1.length) goto neq; | |
| /* and a different character somewhere is also unequal */ | |
| if (s1.ir && s2.ir) | |
| for (k=0; k < s1.length; k++) { if (s1.ir[k] != s2.ir[k]) goto neq; } | |
| else if (s1.address && s2.address) | |
| for (k=0; k < s1.length; k++) { if (memread2(s1.address+k) != memread2(s2.address+k)) goto neq; } | |
| else { | |
| if (s1.address) { a=s1.address; ir=s2.ir; } else { a=s2.address; ir=s1.ir; } | |
| for (k=0; k < s1.length; k++) { if (memread2(a+k) != ir[k] ) goto neq; } | |
| } | |
| for (k=0; k < s1.length; k++) if (s1.ir[k] != s2.ir[k]) goto neq; | |
| /* which operator did we use */ | |
| if (t == '=') push(booleanmode); else push(0); | |
| return; | |
| neq: | |
| if (t == '=') push(0); else push(booleanmode); | |
| return; | |
| } | |
| /* | |
| * floating point arithmetic | |
| * SIN, COS, TAN, ATAN, LOG, EXP, INT | |
| * INT is always there and is nop in integer BASICs | |
| * no handling of floating point errors yet. | |
| */ | |
| void xsin() { push(sin(pop())); } | |
| void xcos() { push(cos(pop())); } | |
| void xtan() { push(tan(pop())); } | |
| void xatan() { push(atan(pop())); } | |
| void xlog() { push(log(pop())); } | |
| void xexp() { push(exp(pop())); } | |
| void xint() { push(floor(pop())); } | |
| void xint() {} | |
| /* | |
| * Recursive expression parser functions | |
| * | |
| * factor(); term(); addexpression(); compexpression(); | |
| * notexpression(); andexpression(); expression() | |
| * | |
| * doing | |
| * | |
| * functions, numbers; *, /, %; +, -; =, <>, =>, <=; | |
| * NOT; AND; OR | |
| * | |
| */ | |
| /* | |
| * factor() - contrary to all other function | |
| * nothing here should end with a new token - this is handled | |
| * in factors calling function | |
| * | |
| * evaluates constants, variables and all functions | |
| */ | |
| /* helpers of factor - array access */ | |
| void factorarray() { | |
| lhsobject_t object; | |
| number_t v; | |
| /* remember the variable, because parsesubscript changes this */ | |
| object.name=name; | |
| /* parse the arguments */ | |
| parsesubscripts(); | |
| if (er != 0 ) return; | |
| switch(args) { | |
| case 1: | |
| object.i=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| object.j=arraylimit; | |
| break; | |
| case 2: | |
| object.j=popaddress(); | |
| object.i=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| break; | |
| default: | |
| error(EARGS); | |
| return; | |
| } | |
| array(&object, 'g', &v); | |
| push(v); | |
| } | |
| /* helpers of factor - string length */ | |
| void factorlen() { | |
| address_t a; | |
| string_t s; | |
| name_t n; | |
| lhsobject_t lhs; | |
| nexttoken(); | |
| if ( token != '(') { error(EARGS); return; } | |
| nexttoken(); | |
| switch(token) { | |
| case STRING: | |
| push(sr.length); | |
| nexttoken(); | |
| break; | |
| case STRINGVAR: | |
| parsestringvar(&s, &lhs); | |
| push(s.length); | |
| nexttoken(); | |
| break; | |
| case TRIGHT: | |
| case TLEFT: | |
| case TMID: | |
| case TCHR: | |
| case TSTR: | |
| error(EARGS); | |
| return; | |
| default: | |
| expression(); | |
| if (!USELONGJUMP && er) return; | |
| n.token=TBUFFER; | |
| a=pop(); | |
| n.c[0]=a%256; | |
| n.c[1]=a/256; | |
| push(blength(&n)); | |
| } | |
| if (!USELONGJUMP && er) return; | |
| if (token != ')') { error(EARGS); return; } | |
| push(0); | |
| } | |
| /* helpers of factor - the VAL command */ | |
| void factorval() { | |
| index_t y; | |
| number_t x; | |
| string_t s; | |
| address_t a; | |
| char *ir; | |
| nexttoken(); | |
| if (token != '(') { error(EARGS); return; } | |
| nexttoken(); | |
| if (!stringvalue(&s)) { error(EUNKNOWN); return; } | |
| if (!USELONGJUMP && er) return; | |
| /* the length of the strings consumed */ | |
| vlength=0; | |
| /* get the string if it is in serial memory */ | |
| if (!s.ir) getstringtobuffer(&s, spistrbuf1, SPIRAMSBSIZE); | |
| /* generate a 0 terminated string for use with parsenumber */ | |
| stringtobuffer(sbuffer, &s); | |
| ir=sbuffer; | |
| /* remove whitespaces */ | |
| while(*ir==' ' || *ir=='\t') { ir++; vlength++; } | |
| /* find a sign */ | |
| if(*ir=='-') { y=-1; ir++; vlength++; } else y=1; | |
| x=0; | |
| if ((a=parsenumber2(ir, &x)) > 0) {vlength+=a; ert=0; } else {vlength=0; ert=1;}; | |
| if ((a=parsenumber(ir, &x)) > 0) {vlength+=a; ert=0; } else {vlength=0; ert=1;}; | |
| push(x*y); | |
| nexttoken(); | |
| if (token != ')') { error(EARGS); return; } | |
| } | |
| /* helpers of factor - the INSTR command */ | |
| /* this is instring in a single character version, usefull to split strings */ | |
| void factorinstr() { | |
| char ch; | |
| address_t a; | |
| string_t s; | |
| nexttoken(); | |
| if (token != '(') { error(EARGS); return; } | |
| nexttoken(); | |
| if (!stringvalue(&s)) { error(EUNKNOWN); return; } | |
| if (!USELONGJUMP && er) return; | |
| nexttoken(); | |
| if (token != ',') { error(EARGS); return; } | |
| nexttoken(); | |
| expression(); | |
| if (!USELONGJUMP && er) return; | |
| ch=pop(); | |
| if (s.address) { | |
| for (a=1; a<=s.length; a++) {if (memread2(s.address+a-1) == ch) break; } | |
| } else { | |
| for (a=1; a<=s.length; a++) {if (s.ir[a-1] == ch) break; } | |
| } | |
| if (a > s.length) a=0; | |
| push(a); | |
| if (token != ')') { error(EARGS); return; } | |
| } | |
| /* the full instr command which can compare two strings */ | |
| void factorinstr() { | |
| char ch; | |
| address_t a=1; | |
| address_t i=1; | |
| string_t search; | |
| string_t s; | |
| nexttoken(); | |
| if (token != '(') { error(EARGS); return; } | |
| nexttoken(); | |
| /* the search string */ | |
| if(!stringvalue(&s)) { error(EUNKNOWN); return; } | |
| if (!USELONGJUMP && er) return; | |
| nexttoken(); | |
| if (token != ',') { error(EARGS); return; } | |
| nexttoken(); | |
| /* the string to be searched */ | |
| if (!stringvalue(&search)) { error(EUNKNOWN); return; } | |
| if (!USELONGJUMP && er) return; | |
| nexttoken(); | |
| /* potentially the start value */ | |
| if (token == ',') { | |
| nexttoken(); | |
| expression(); | |
| if (!USELONGJUMP && er) return; | |
| a=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| } | |
| if (token != ')') { | |
| error(EUNKNOWN); | |
| return; | |
| } | |
| /* health check */ | |
| if (search.length == 0 || search.length+a > s.length || a == 0) { | |
| push(0); | |
| return; | |
| } | |
| /* go through the search string */ | |
| while(i<=search.length) { | |
| /* get one character from the string */ | |
| if (search.address) { | |
| ch=memread2(search.address+i-1); | |
| } else { | |
| ch=search.ir[i-1]; | |
| } | |
| /* search the character */ | |
| if (s.address) { | |
| for (; a<=s.length; a++) {if (memread2(s.address+a-1) == ch) break; } | |
| } else { | |
| for (; a<=s.length; a++) {if ( s.ir[a-1] == ch ) break; } | |
| } | |
| /* we haven't found the character until the end of the string */ | |
| if (a > s.length) { | |
| a=0; | |
| break; | |
| } | |
| /* next character */ | |
| i+=1; | |
| } | |
| /* how did the search go? a rewind because we were at the end of the search part already */ | |
| if (i <= search.length) { | |
| a=0; | |
| } else { | |
| a=a-search.length+1; | |
| } | |
| push(a); | |
| } | |
| /* helpers of factor - the NETSTAT command */ | |
| void factornetstat() { | |
| address_t x=0; | |
| if (netconnected()) x=1; | |
| if (mqttstate() == 0) x+=2; | |
| push(x); | |
| } | |
| /* helpers of factor - the ASC command, really not needed but for completeness */ | |
| void factorasc() { | |
| string_t s; | |
| lhsobject_t lhs; | |
| nexttoken(); | |
| if ( token != '(') { error(EARGS); return; } | |
| nexttoken(); | |
| switch(token) { | |
| case STRING: | |
| if (sr.ir) push(sr.ir[0]); else push(memread2(sr.address)); | |
| nexttoken(); | |
| break; | |
| case STRINGVAR: | |
| parsestringvar(&s, &lhs); | |
| if (s.length > 0) { | |
| if (s.ir) push(s.ir[0]); else push(memread2(s.address)); | |
| } else | |
| push(0); | |
| nexttoken(); | |
| break; | |
| default: | |
| error(EARGS); | |
| return; | |
| } | |
| if (!USELONGJUMP && er) return; | |
| if (token != ')') { error(EARGS); return; } | |
| push(0); | |
| } | |
| void factor(){ | |
| if (DEBUG) bdebug("factor\n"); | |
| switch (token) { | |
| case NUMBER: | |
| push(x); | |
| break; | |
| case VARIABLE: | |
| push(getvar(&name)); | |
| break; | |
| case ARRAYVAR: | |
| factorarray(); | |
| break; | |
| case '(': | |
| nexttoken(); | |
| expression(); | |
| if (er != 0 ) return; | |
| if (token != ')') { error(EARGS); return; } | |
| break; | |
| /* Palo Alto BASIC functions */ | |
| case TABS: | |
| parsefunction(xabs, 1); | |
| break; | |
| case TRND: | |
| parsefunction(xrnd, 1); | |
| break; | |
| case TSIZE: | |
| push(himem-top); | |
| break; | |
| /* Apple 1 BASIC functions */ | |
| case TSGN: | |
| parsefunction(xsgn, 1); | |
| break; | |
| case TPEEK: | |
| parsefunction(xpeek, 1); | |
| break; | |
| case TLEN: | |
| factorlen(); | |
| break; | |
| case TAVAIL: | |
| parsefunction(xavail, 1); | |
| break; | |
| case TOPEN: | |
| parsefunction(xfopen, 1); | |
| break; | |
| case TSENSOR: | |
| parsefunction(xfsensor, 2); | |
| break; | |
| case TVAL: | |
| factorval(); | |
| break; | |
| case TINSTR: | |
| factorinstr(); | |
| break; | |
| case TWIRE: | |
| parsefunction(xfwire, 1); | |
| break; | |
| case TERROR: | |
| push(erh); | |
| break; | |
| case THIMEM: | |
| push(himem); | |
| break; | |
| /* Apple 1 string compare code */ | |
| case STRING: | |
| case STRINGVAR: | |
| case TSTR: | |
| case TLEFT: | |
| case TRIGHT: | |
| case TMID: | |
| case TCHR: | |
| streval(); | |
| if (er != 0 ) return; | |
| break; | |
| /* Stefan's tinybasic additions */ | |
| case TSQR: | |
| parsefunction(sqr, 1); | |
| break; | |
| case TMAP: | |
| parsefunction(xmap, 5); | |
| break; | |
| case TUSR: | |
| parsefunction(xusr, 2); | |
| break; | |
| case TPOW: | |
| parsefunction(xpow, 2); | |
| break; | |
| /* Arduino I/O */ | |
| case TAREAD: | |
| parsefunction(xaread, 1); | |
| break; | |
| case TDREAD: | |
| parsefunction(xdread, 1); | |
| break; | |
| case TMILLIS: | |
| parsefunction(bmillis, 1); | |
| break; | |
| case TPULSE: | |
| parsefunction(bpulsein, 3); | |
| break; | |
| case TAZERO: | |
| push(A0); | |
| push(0); | |
| break; | |
| case TLED: | |
| push(LED_BUILTIN); | |
| push(0); | |
| break; | |
| /* mathematical functions in case we have float */ | |
| case TSIN: | |
| parsefunction(xsin, 1); | |
| break; | |
| case TCOS: | |
| parsefunction(xcos, 1); | |
| break; | |
| case TTAN: | |
| parsefunction(xtan, 1); | |
| break; | |
| case TATAN: | |
| parsefunction(xatan, 1); | |
| break; | |
| case TLOG: | |
| parsefunction(xlog, 1); | |
| break; | |
| case TEXP: | |
| parsefunction(xexp, 1); | |
| break; | |
| /* int is always present to make programs compatible */ | |
| case TINT: | |
| parsefunction(xint, 1); | |
| break; | |
| case TFN: | |
| xfn(0); | |
| break; | |
| /* an arcane feature, DATA evaluates to the data record number */ | |
| case TDATA: | |
| push(datarc); | |
| break; | |
| case TMALLOC: | |
| parsefunction(xmalloc, 2); | |
| break; | |
| case TFIND: | |
| xfind(); | |
| break; | |
| case TNETSTAT: | |
| factornetstat(); | |
| break; | |
| case TASC: | |
| factorasc(); | |
| break; | |
| /* unknown function */ | |
| default: | |
| error(EUNKNOWN); | |
| return; | |
| } | |
| } | |
| /* this is how the power operator ^ is handled */ | |
| /* the recursive version */ | |
| void power() { | |
| if (DEBUG) bdebug("power\n"); | |
| factor(); | |
| if (!USELONGJUMP && er) return; | |
| nexttoken(); | |
| if (DEBUG) bdebug("in power\n"); | |
| if (token == '^'){ | |
| parseoperator(power); | |
| if (!USELONGJUMP && er) return; | |
| push(bpow(x,y)); | |
| } | |
| if (DEBUG) bdebug("leaving power\n"); | |
| } | |
| /* the left associative version */ | |
| void power() { | |
| if (DEBUG) bdebug("power\n"); | |
| factor(); | |
| if (!USELONGJUMP && er) return; | |
| nextpower: | |
| nexttoken(); | |
| if (DEBUG) bdebug("in power\n"); | |
| if (token == '^'){ | |
| parseoperator(factor); | |
| push(bpow(x,y)); | |
| goto nextpower; | |
| } | |
| if (DEBUG) bdebug("leaving power\n"); | |
| } | |
| /* | |
| * term() evaluates multiplication, division and mod | |
| */ | |
| void term(){ | |
| if (DEBUG) bdebug("term\n"); | |
| power(); | |
| if (!USELONGJUMP && er) return; | |
| nextfactor: | |
| if (DEBUG) bdebug("in term\n"); | |
| if (token == '*'){ | |
| parseoperator(power); | |
| if (!USELONGJUMP && er) return; | |
| push(x*y); | |
| goto nextfactor; | |
| } else if (token == '/'){ | |
| parseoperator(power); | |
| if (!USELONGJUMP && er) return; | |
| if (y != 0) | |
| push(x/y); | |
| if (forceint) push((int)x/(int)y); else push(x/y); | |
| else { | |
| error(EDIVIDE); | |
| return; | |
| } | |
| goto nextfactor; | |
| } else if (token == '%') { | |
| parseoperator(power); | |
| if (!USELONGJUMP && er) return; | |
| if (y != 0) | |
| push(x%y); | |
| push((int)x%(int)y); | |
| else { | |
| error(EDIVIDE); | |
| return; | |
| } | |
| goto nextfactor; | |
| } | |
| if (DEBUG) bdebug("leaving term\n"); | |
| } | |
| /* add and subtract */ | |
| void addexpression(){ | |
| if (DEBUG) bdebug("addexp\n"); | |
| if (token != '+' && token != '-') { | |
| term(); | |
| if (!USELONGJUMP && er) return; | |
| } else { | |
| push(0); | |
| } | |
| nextterm: | |
| if (token == '+' ) { | |
| parseoperator(term); | |
| if (!USELONGJUMP && er) return; | |
| push(x+y); | |
| goto nextterm; | |
| } else if (token == '-'){ | |
| parseoperator(term); | |
| if (!USELONGJUMP && er) return; | |
| push(x-y); | |
| goto nextterm; | |
| } | |
| } | |
| /* comparisions */ | |
| void compexpression() { | |
| if (DEBUG) bdebug("compexp\n"); | |
| addexpression(); | |
| if (!USELONGJUMP && er) return; | |
| switch (token){ | |
| case '=': | |
| parseoperator(compexpression); | |
| if (!USELONGJUMP && er) return; | |
| push(x == y ? booleanmode : 0); | |
| break; | |
| case NOTEQUAL: | |
| parseoperator(compexpression); | |
| if (!USELONGJUMP && er) return; | |
| push(x != y ? booleanmode : 0); | |
| break; | |
| case '>': | |
| parseoperator(compexpression); | |
| if (!USELONGJUMP && er) return; | |
| push(x > y ? booleanmode : 0); | |
| break; | |
| case '<': | |
| parseoperator(compexpression); | |
| if (!USELONGJUMP && er) return; | |
| push(x < y ? booleanmode : 0); | |
| break; | |
| case LESSEREQUAL: | |
| parseoperator(compexpression); | |
| if (!USELONGJUMP && er) return; | |
| push(x <= y ? booleanmode : 0); | |
| break; | |
| case GREATEREQUAL: | |
| parseoperator(compexpression); | |
| if (!USELONGJUMP && er) return; | |
| push(x >= y ? booleanmode : 0); | |
| break; | |
| } | |
| } | |
| /* boolean NOT */ | |
| void notexpression() { | |
| if (DEBUG) bdebug("notexp\n"); | |
| if (token == TNOT) { | |
| nexttoken(); | |
| expression(); | |
| if (!USELONGJUMP && er) return; | |
| if (booleanmode == -1) push(~(short)pop()); | |
| else if (pop() == 0) push(1); else push(0); | |
| } else | |
| compexpression(); | |
| } | |
| /* boolean AND and at the same time bitwise */ | |
| void andexpression() { | |
| if (DEBUG) bdebug("andexp\n"); | |
| notexpression(); | |
| if (!USELONGJUMP && er) return; | |
| if (token == TAND) { | |
| parseoperator(expression); | |
| if (!USELONGJUMP && er) return; | |
| push((short)x & (short)y); | |
| } | |
| } | |
| /* expression function and boolean OR at the same time bitwise !*/ | |
| void expression(){ | |
| if (DEBUG) bdebug("exp\n"); | |
| andexpression(); | |
| if (!USELONGJUMP && er) return; | |
| if (token == TOR) { | |
| parseoperator(expression); | |
| if (!USELONGJUMP && er) return; | |
| push((short)x | (short)y); | |
| } | |
| } | |
| /* expression function simplified */ | |
| void expression(){ | |
| if (DEBUG) bdebug("exp\n"); | |
| compexpression(); | |
| if (!USELONGJUMP && er) return; | |
| if (token == TOR) { | |
| parseoperator(expression); | |
| if (!USELONGJUMP && er) return; | |
| push((short)x | (short)y); | |
| } | |
| } | |
| /* | |
| * Layer 2 - The commands and their helpers | |
| * | |
| * Palo Alto BASIC languge set - PRINT, LET, INPUT, GOTO, GOSUB, RETURN, | |
| * IF, FOR, TO, NEXT, STEP, BREAK, STOP, END, LIST, NEW, RUN, REM | |
| * BREAK is not Palo ALto but fits here, eEND is identical to STOP. | |
| */ | |
| /* | |
| * PRINT command, extended by many features like file, wire, mqtt and radio i/o. | |
| * TAB added as part of the PRINT statement with C64 compatibility. | |
| */ | |
| void xprint(){ | |
| char semicolon = 0; | |
| char oldod; | |
| char modifier = 0; | |
| string_t s; | |
| stringlength_t i; | |
| form=0; | |
| oldod=od; | |
| nexttoken(); | |
| processsymbol: | |
| /* at the end of a print statement, do we need a newline, restore the defaults */ | |
| if (termsymbol()) { | |
| if (!semicolon) outcr(); | |
| od=oldod; | |
| form=0; | |
| return; | |
| } | |
| semicolon=0; | |
| /* output a string if we found it */ | |
| if (stringvalue(&s)) { | |
| if (!USELONGJUMP && er) return; | |
| /* buffer must be used here for machine code to work */ | |
| if (!s.ir) getstringtobuffer(&s, spistrbuf1, SPIRAMSBSIZE); | |
| outs(s.ir, s.length); | |
| nexttoken(); | |
| goto separators; | |
| } | |
| /* the tab command as part of print */ | |
| if (token == TTAB || token == TSPC) { | |
| xtab(); | |
| goto separators; | |
| } | |
| /* modifiers of the print statement */ | |
| if (token == '#' || token == '&') { | |
| modifier=token; | |
| nexttoken(); | |
| expression(); | |
| if (!USELONGJUMP && er) return; | |
| switch(modifier) { | |
| case '#': | |
| form=pop(); | |
| break; | |
| case '&': | |
| od=pop(); | |
| break; | |
| } | |
| goto separators; | |
| } | |
| if (token != ',' && token != ';') { | |
| expression(); | |
| if (!USELONGJUMP && er) return; | |
| outnumber(pop()); | |
| } | |
| /* commas and semicolons, all other symbols are accepted and no error is thrown */ | |
| separators: | |
| if (termsymbol()) goto processsymbol; | |
| switch (token) { | |
| case ',': | |
| if (!modifier) outspc(); | |
| case ';': | |
| semicolon=1; | |
| nexttoken(); | |
| break; | |
| } | |
| modifier=0; | |
| goto processsymbol; | |
| } | |
| /* | |
| * LET assigment code for various lefthand and righthand side. | |
| * | |
| * | |
| * lefthandside is a helper function for reuse in other | |
| * commands. It determines the address the value is to be | |
| * assigned to and whether the assignment target is a | |
| * "pure" i.e. subscriptless string expression | |
| * | |
| * assignnumber assigns a number to a given lefthandside | |
| * | |
| * In lefthandside the type of the object is determined and | |
| * possible subscripts are parsed. | |
| * | |
| * Variables have no subscripts. The arguments are unchanged. | |
| * Arrays may have two subscript which will go to i, j | |
| * Strings may have a string subscript, going to i and an array subscript | |
| * going to j | |
| * String arrays use i2 as the array index. | |
| * Note that the role of the variables differs for string array and normal | |
| * arrays. | |
| * Strings without a subscript i.e. pure strings, set the ps flag | |
| * | |
| */ | |
| /* the new lefthandsite code */ | |
| void lefthandside(lhsobject_t* lhs) { | |
| /* just to provide it for parsestringvar to reuse the righthandside code */ | |
| address_t temp; | |
| if (DEBUG) { | |
| outsc("assigning to variable "); | |
| outname(&lhs->name); outspc(); | |
| outsc(" type "); outnumber(lhs->name.token); | |
| outcr(); | |
| } | |
| /* prep it */ | |
| lhs->i=1; | |
| lhs->i2=0; | |
| lhs->j=arraylimit; | |
| lhs->ps=1; | |
| /* look at the variables and continue parsing */ | |
| switch (lhs->name.token) { | |
| case VARIABLE: | |
| nexttoken(); | |
| break; | |
| case ARRAYVAR: | |
| parsesubscripts(); | |
| if (!USELONGJUMP && er) return; | |
| switch(args) { | |
| case 1: | |
| lhs->i=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| lhs->j=arraylimit; | |
| break; | |
| case 2: | |
| lhs->j=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| lhs->i=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| break; | |
| default: | |
| error(EARGS); | |
| return; | |
| } | |
| nexttoken(); | |
| break; | |
| case STRINGVAR: | |
| parsestringvar(0, lhs); | |
| nexttoken(); | |
| break; | |
| /* here we could implement a string thing for a true Tinybasic */ | |
| default: | |
| error(EUNKNOWN); | |
| return; | |
| } | |
| if (DEBUG) { | |
| outsc("** in assignment lefthandside with (i,j,ps,i2) "); | |
| outnumber(lhs->i); outspc(); | |
| outnumber(lhs->j); outspc(); | |
| outnumber(lhs->ps); outspc(); | |
| outnumber(lhs->i2); outcr(); | |
| outsc(" token is "); outputtoken(); | |
| outsc(" at "); outnumber(here); outcr(); | |
| } | |
| } | |
| /* assign a number to a left hand side we have parsed */ | |
| void assignnumber2(lhsobject_t* lhs, number_t x) { | |
| string_t sr; | |
| /* depending on the variable type, assign the value */ | |
| switch (lhs->name.token) { | |
| case VARIABLE: | |
| setvar(&lhs->name, x); | |
| break; | |
| case ARRAYVAR: | |
| array(lhs, 's', &x); | |
| break; | |
| case STRINGVAR: | |
| /* find the string variable */ | |
| getstring(&sr, &lhs->name, lhs->i, lhs->j); | |
| if (!USELONGJUMP && er) return; | |
| /* the first character of the string is set to the number */ | |
| if (sr.ir) sr.ir[0]=x; else if (sr.address) memwrite2(ax, x); else error(EUNKNOWN); | |
| /* set the length */ | |
| if (lhs->ps) | |
| setstringlength(&lhs->name, 1, lhs->j); | |
| else | |
| if (sr.length < lhs->i && lhs->i <= sr.strdim) | |
| setstringlength(&lhs->name, lhs->i, lhs->j); | |
| break; | |
| } | |
| } | |
| /* | |
| * LET - the core assigment function, this is different from other BASICs | |
| */ | |
| void assignment() { | |
| address_t newlength, copybytes; | |
| mem_t s; | |
| index_t k; | |
| char tmpchar; /* for number conversion only */ | |
| string_t sr, sl; /* the right and left hand side strings */ | |
| /* the lefthandside identifier */ | |
| lhsobject_t lhs; | |
| /* this code evaluates the left hand side, we remember the object information first */ | |
| lhs.name=name; | |
| lefthandside(&lhs); | |
| if (!USELONGJUMP && er) return; | |
| /* the assignment part */ | |
| if (token != '=') { error(EUNKNOWN); return; } | |
| nexttoken(); | |
| /* here comes the code for the right hand side, the evaluation depends on the left hand side type */ | |
| switch (lhs.name.token) { | |
| /* the lefthandside is a scalar, evaluate the righthandside as a number, even if it is a string */ | |
| case VARIABLE: | |
| case ARRAYVAR: | |
| expression(); | |
| if (!USELONGJUMP && er) return; | |
| assignnumber2(&lhs, pop()); | |
| break; | |
| /* the lefthandside is a string variable, try evaluate the righthandside as a stringvalue */ | |
| case STRINGVAR: | |
| nextstring: | |
| /* do we deal with a string as righthand side */ | |
| s=stringvalue(&sr); | |
| if (!USELONGJUMP && er) return; | |
| /* and then as an expression if it is no string, any number appearing in a string expression terminates the addition loop */ | |
| if (!s) { | |
| expression(); | |
| if (!USELONGJUMP && er) return; | |
| tmpchar=pop(); | |
| sr.length=1; | |
| sr.ir=&tmpchar; | |
| } else | |
| nexttoken(); /* we do this here because expression also advances, this way we avoid double advance */ | |
| if (DEBUG) { outsc("* assigment stringcode at "); outnumber(here); outcr(); } | |
| /* at this point we have a stringvalue with ir2 pointing to the payload and the stack the length | |
| this is either coming from stringvalue or from the expression code */ | |
| /* we now process the source string */ | |
| /* getstring of the destination */ | |
| getstring(&sl, &lhs.name, lhs.i, lhs.j); | |
| if (!USELONGJUMP && er) return; | |
| /* this debug messes up sbuffer hence all functions that use it in stringvalue produce wrong results */ | |
| if (DEBUG) { | |
| outsc("* assigment stringcode "); outname(&lhs.name); outcr(); | |
| outsc("** assignment source string length "); outnumber(sr.length); outcr(); | |
| outsc("** assignment dest string length "); outnumber(sl.length); outcr(); | |
| outsc("** assignment dest string dimension "); outnumber(sl.strdim); outcr(); | |
| } | |
| /* does the source string fit into the destination if we have no destination second index*/ | |
| if ((lhs.i2 == 0) && ((lhs.i+sr.length-1)>sl.strdim)) { error(EORANGE); return; }; | |
| /* if we have a second index, is it in range */ | |
| if ((lhs.i2 != 0) && lhs.i2>sl.strdim) { error(EORANGE); return; }; | |
| /* calculate the number of bytes we truely want to copy */ | |
| if (lhs.i2 > 0) copybytes=((lhs.i2-lhs.i+1) > sr.length) ? sr.length : (lhs.i2-lhs.i+1); | |
| else copybytes=sr.length; | |
| if (DEBUG) { outsc("** assignment copybytes "); outnumber(copybytes); outcr(); } | |
| /* now do the heavy lifting in a seperate function to encasulate buffering */ | |
| assignstring(&sl, &sr, copybytes); | |
| /* | |
| * classical Apple 1 behaviour is string truncation in substring logic, with | |
| * two index destination string we follow another route. We extend the string | |
| * for the number of copied bytes | |
| */ | |
| if (lhs.i2 == 0) { | |
| newlength = lhs.i+sr.length-1; | |
| } else { | |
| if (lhs.i+copybytes > sl.length) newlength=lhs.i+copybytes-1; | |
| else newlength=sl.length; | |
| } | |
| setstringlength(&lhs.name, newlength, lhs.j); | |
| /* | |
| * we have processed one string and copied it fully to the destination | |
| * see if there is more to come. For inplace strings this is odd because | |
| * one term can change during adding (A$ = B$ + A$). | |
| */ | |
| addstring: | |
| if (token == '+') { | |
| lhs.i=lhs.i+copybytes; | |
| nexttoken(); | |
| goto nextstring; | |
| } | |
| break; /* case STRINGVAR */ | |
| } /* switch */ | |
| } | |
| /* | |
| * Try to copy one string to the other, assumes that getstring did its work | |
| * and that copybyte is correct. | |
| * BASICs in place strings make this a non trivial exercise as we need to | |
| * avoid overwrites. | |
| * Another complication is the mixed situation of BASIC memory strings | |
| * and C memory strings. | |
| */ | |
| void assignstring(string_t* sl, string_t* sr, stringlength_t copybytes) { | |
| stringlength_t k; | |
| /* if we have a memory model that needs the mem interface, go through the addresses by default | |
| else use just the pointers */ | |
| /* for a regular string variable as left hand side we know the address */ | |
| if (sl->address) { | |
| /* for a regular string variable as a source we need to take care of order */ | |
| if (sr->address) { | |
| if (sr->address > sl->address) | |
| for (k=0; k<copybytes; k++) memwrite2(sl->address+k, memread2(sr->address+k)); | |
| else | |
| for (k=1; k<=copybytes; k++) memwrite2(sl->address+copybytes-k, memread2(sr->address+copybytes-k)); | |
| } else { | |
| /* if the right hand side is a special string or a constant things are much simpler */ | |
| for (k=0; k<copybytes; k++) memwrite2(sl->address+k, sr->ir[k]); | |
| } | |
| } else { | |
| /* non regular string variables like @U$ and @T$ are never assignable */ | |
| error(EUNKNOWN); | |
| } | |
| /* we just go through the C memory here */ | |
| if (sr->ir && sl->ir) { | |
| if (sr->ir > sl->ir) | |
| for (k=0; k<copybytes; k++) sl->ir[k]=sr->ir[k]; | |
| else | |
| for (k=1; k<=copybytes; k++) sl->ir[copybytes-k]=sr->ir[copybytes-k]; | |
| } else { | |
| error(EUNKNOWN); | |
| } | |
| } | |
| /* | |
| * INPUT ["string",] variable [,["string",] variable] | |
| * | |
| * The original version of input only processes simple variables one at a time | |
| * and does not support arrays. The code is redudant to assignment and read. | |
| * It also does not support comma separated lists of values to be input. | |
| */ | |
| void showprompt() { | |
| outsc("? "); | |
| } | |
| /* | |
| * Reimplementation of input using the same pattern as read and print . | |
| */ | |
| void xinput() { | |
| mem_t oldid = id; /* remember the stream on modify */ | |
| mem_t prompt = 1; /* determine if we show the prompt */ | |
| number_t xv; /* for number conversion with innumber */ | |
| /* the identifier of the lefthandside */ | |
| lhsobject_t lhs; | |
| address_t maxlen, newlength; /* the maximum length of the string to be read */ | |
| int k=0; /* the result of the number conversion */ | |
| string_t s; | |
| char* buffer; /* the buffer we use for input */ | |
| address_t bufsize; /* the size of the buffer */ | |
| /* depending on the RUN state we use either the input buffer or the string buffer */ | |
| /* this ways we can process long inputs in RUN and don't need a lot of memory */ | |
| if (st == SRUN || st == SERUN) { | |
| buffer=ibuffer; | |
| bufsize=BUFSIZE; | |
| } else { | |
| buffer=sbuffer; | |
| bufsize=SBUFSIZE; | |
| } | |
| /* get the next token and check what we are dealing with */ | |
| nexttoken(); | |
| /* modifiers of the input statement (stream) */ | |
| if (token == '&') { | |
| if(!expectexpr()) return; | |
| oldid=id; | |
| id=pop(); | |
| if (id != ISERIAL || id !=IKEYBOARD) prompt=0; | |
| if (token != ',') { | |
| error(EUNKNOWN); | |
| return; | |
| } else | |
| nexttoken(); | |
| } | |
| /* unlike print, form can appear only once in input after the | |
| stream, it controls character counts in wire */ | |
| if (token == '#') { | |
| if(!expectexpr()) return; | |
| form=pop(); | |
| if (token != ',') { | |
| error(EUNKNOWN); | |
| return; | |
| } else | |
| nexttoken(); | |
| } | |
| /* we have a string to be printed to prompt the user */ | |
| nextstring: | |
| if (token == STRING && id != IFILE) { | |
| prompt=0; | |
| if (!sr.ir) getstringtobuffer(&sr, spistrbuf1, SPIRAMSBSIZE); | |
| outs(sr.ir, sr.length); | |
| nexttoken(); | |
| } | |
| /* now we check for a variable and parse it */ | |
| nextvariable: | |
| if (token == VARIABLE || token == ARRAYVAR || token == STRINGVAR) { | |
| /* check for a valid lefthandside expression */ | |
| lhs.name=name; | |
| lefthandside(&lhs); | |
| if (!USELONGJUMP && er) return; | |
| /* which data type do we input */ | |
| switch (lhs.name.token) { | |
| case VARIABLE: | |
| case ARRAYVAR: | |
| again: | |
| /* if we have no buffer or are at the end, read it and set cursor k to the beginning */ | |
| if (k == 0 || (address_t) buffer[0] < k) { | |
| if (prompt) showprompt(); | |
| (void) ins(buffer, bufsize); | |
| k=1; | |
| } | |
| /* read a number from the buffer and return it, advance the cursor k */ | |
| k=innumber(&xv, buffer, k); | |
| /* if we break, end it here */ | |
| if (k == -1) { | |
| st=SINT; | |
| token=EOL; | |
| goto resetinput; | |
| } | |
| /* if we have no valid number, ask again */ | |
| if (k == 0) { | |
| if (id == ISERIAL || id == IKEYBOARD) { | |
| printmessage(ENUMBER); | |
| outspc(); | |
| printmessage(EGENERAL); | |
| outcr(); | |
| xv=0; | |
| k=0; | |
| goto again; | |
| } else { | |
| ert=1; | |
| xv=0; | |
| goto resetinput; | |
| } | |
| } | |
| /* now assign the number */ | |
| assignnumber2(&lhs, xv); | |
| /* look if there is a comma coming in the buffer and keep it */ | |
| while (k < (address_t) buffer[0] && buffer[k] != 0) { | |
| if (buffer[k] == ',') { | |
| k++; | |
| break; | |
| } | |
| k++; | |
| } | |
| break; | |
| case STRINGVAR: | |
| /* the destination address of the lefthandside, on the fly create included */ | |
| getstring(&s, &lhs.name, lhs.i, lhs.j); | |
| if (!USELONGJUMP && er) return; | |
| /* the length of the lefthandside string */ | |
| if (lhs.i2 == 0) { | |
| maxlen=s.strdim-lhs.i+1; | |
| } else { | |
| maxlen=lhs.i2-lhs.i+1; | |
| if (maxlen > s.strdim) maxlen=s.strdim-lhs.i+1; | |
| } | |
| /* the number of bytes we want to read the form parameter in WIRE can be used | |
| to set the expected number of bytes */ | |
| if (form != 0 && form < maxlen) maxlen=form; | |
| /* what is going on */ | |
| if (DEBUG) { | |
| outsc("** input stringcode at "); outnumber(here); outcr(); | |
| outsc("** input stringcode "); outname(&lhs.name); outcr(); | |
| outsc("** input stringcode maximum length "); outnumber(maxlen); outcr(); | |
| } | |
| /* now read the string inplace */ | |
| if (prompt) showprompt(); | |
| newlength=ins(s.ir-1, maxlen); | |
| newlength=ins(spistrbuf1, maxlen); | |
| /* if we have a string variable, we need to copy the buffer to the string */ | |
| if (newlength > 0) { | |
| if (s.ir) { | |
| for (k=0; k<newlength; k++) s.ir[k]=spistrbuf1[k+1]; | |
| } else { | |
| for (k=0; k<newlength; k++) memwrite2(s.address+k, spistrbuf1[k+1]); | |
| } | |
| } | |
| /* if we have a string variable, we need to copy the buffer to the string */ | |
| /* set the right string length */ | |
| /* classical Apple 1 behaviour is string truncation in substring logic */ | |
| newlength = lhs.i+newlength-1; | |
| setstringlength(&lhs.name, newlength, lhs.j); | |
| break; | |
| } | |
| } | |
| /* seperators and termsymbols */ | |
| if (token == ',' || token == ';') { | |
| nexttoken(); | |
| goto nextstring; | |
| } | |
| /* no further data */ | |
| if (!termsymbol()) { | |
| error(EUNKNOWN); | |
| } | |
| resetinput: | |
| id=oldid; | |
| form=0; | |
| } | |
| /* | |
| * GOTO, GOSUB, RETURN and their helpers | |
| * | |
| * GOTO and GOSUB function for a simple one statement goto | |
| */ | |
| void xgoto() { | |
| token_t t=token; | |
| number_t x; | |
| if (!expectexpr()) return; | |
| if (t == TGOSUB) pushgosubstack(0); | |
| if (!USELONGJUMP && er) return; | |
| x=pop(); | |
| if (DEBUG) { outsc("** goto/gosub evaluated line number "); outnumber(x); outcr(); } | |
| findline((address_t) x); | |
| if (!USELONGJUMP && er) return; | |
| if (DEBUG) { outsc("** goto/gosub branches to "); outnumber(here); outcr(); } | |
| /* goto in interactive mode switched to RUN mode | |
| no clearing of variables and stacks */ | |
| if (st == SINT) st=SRUN; | |
| } | |
| /* | |
| * RETURN retrieves here from the gosub stack | |
| */ | |
| void xreturn(){ | |
| popgosubstack(); | |
| if (DEBUG) { outsc("** restored location "); outnumber(here); outcr(); } | |
| if (!USELONGJUMP && er) return; | |
| nexttoken(); | |
| /* we return from an interrupt and reenable them */ | |
| if (gosubarg[gosubsp] == TEVENT) events_enabled=1; | |
| } | |
| /* | |
| * IF statement together with THEN | |
| */ | |
| void xif() { | |
| mem_t nl=0; | |
| if (!expectexpr()) return; | |
| x=pop(); | |
| if (DEBUG) { outsc("** in if, condition "); outnumber(x); outcr(); } | |
| /* if can have a new line after the expression in this BASIC */ | |
| if (token == LINENUMBER) nexttoken(); | |
| /* we only check false which is 0 */ | |
| if (x == 0) { | |
| /* on condition false skip the entire line and all : until a potential ELSE */ | |
| while(token != LINENUMBER && token != EOL && token != TELSE) nexttoken(); | |
| /* in the structured language set, we need to look for a DO close to the IF and skip it*/ | |
| /* a THEN or not and then a line number expects a block */ | |
| if (token == TTHEN) nexttoken(); | |
| if (token == LINENUMBER) { nexttoken(); nl=1; } | |
| /* skip the block */ | |
| if (token == TDO) { | |
| nexttoken(); | |
| findbraket(TDO, TDEND); | |
| nexttoken(); | |
| goto processelse; | |
| } | |
| /* skip the line */ | |
| if (!nl) while(token != LINENUMBER && token != EOL && token != TELSE) nexttoken(); | |
| processelse: | |
| /* if we have ELSE at this point we want to execute this part of the line as the condition | |
| was false, isolated ELSE is GOTO, otherwise just execute the code */ | |
| /* look if ELSE is at the next line */ | |
| if (token == LINENUMBER) nexttoken(); | |
| /* now process ELSE */ | |
| if (token == TELSE) { | |
| nexttoken(); | |
| if (token == NUMBER) { | |
| findline((address_t) x); | |
| return; | |
| } | |
| } | |
| } | |
| /* a THEN is interpreted as simple one statement goto if it is followed by a line number*/ | |
| /* then can be on a new line */ | |
| if (token == TTHEN) { | |
| nexttoken(); | |
| if (token == NUMBER) { | |
| findline((address_t) x); | |
| } | |
| } | |
| } | |
| /* if else is encountered in the statement line, the rest of the code is skipped | |
| as else code execution is triggered in the xif function */ | |
| void xelse() { | |
| mem_t nl=0; | |
| /* skip the entire line */ | |
| while(token != LINENUMBER && token != EOL) nexttoken(); | |
| nexttoken(); | |
| /* else in a single line */ | |
| if (token == LINENUMBER) { | |
| nexttoken(); | |
| nl=1; | |
| } | |
| /* the block after the else on a new line or the current line */ | |
| if (token == TDO) { | |
| nexttoken(); | |
| findbraket(TDO, TDEND); | |
| } | |
| /* single line else, skip the line */ | |
| if (!nl) while(token != LINENUMBER && token != EOL) nexttoken(); | |
| } | |
| /* | |
| * FOR, NEXT and the apocryphal BREAK | |
| * | |
| * find the NEXT token or the end of the program | |
| */ | |
| /* | |
| * The generic block scanner, used for structured code and in FOR NEXT. | |
| * The closing symbol of a symbol is found. Symbol pairs are: | |
| * | |
| * FOR NEXT | |
| * WHILE WEND | |
| * REPEAT UNTIL | |
| * DO DEND | |
| * SWITCH SWEND | |
| * | |
| */ | |
| void findbraket(token_t bra, token_t ket){ | |
| address_t fnc = 0; | |
| while (1) { | |
| if (DEBUG) { | |
| outsc("** skpping braket "); | |
| outputtoken(); outspc(); | |
| outnumber(here); outspc(); | |
| outnumber(fnc); outcr(); | |
| } | |
| if (token == ket) { | |
| if (fnc == 0) return; else fnc--; | |
| } | |
| if (token == bra) fnc++; | |
| /* no closing symbol found */ | |
| if (token == EOL) { | |
| error(bra); | |
| return; | |
| } | |
| nexttoken(); | |
| } | |
| } | |
| /* | |
| * FOR variable [= expression [to expression]] [STEP expression] | |
| * for stores the variable, the increment and the boudary on the | |
| * for stack. Changing steps and boundaries during the execution | |
| * of a loop has no effect. | |
| * | |
| * This is different from many other BASICS as FOR can be used | |
| * as an open loop with no boundary | |
| */ | |
| void xfor(){ | |
| name_t variable; | |
| number_t begin=1; | |
| number_t to=maxnum; | |
| number_t step=1; | |
| /* we need at least a name */ | |
| if (!expect(VARIABLE, EUNKNOWN)) return; | |
| variable=name; | |
| /* | |
| * This is not standard BASIC. | |
| * All combinations of FOR TO STEP are allowed. | |
| * FOR X : NEXT is an infinite loop. | |
| * FOR X=1 : NEXT is a loop with X=1 and an infinite loop. | |
| * FOR X=1 TO 10 : NEXT is a loop with X=1 to 10. | |
| * FOR X=1 TO 10 STEP 2 : NEXT is a loop with X=1 to 10 in steps of 2. | |
| * A variable must always be supplyed to indentify the loop. | |
| * FOR : NEXT is illegal. | |
| */ | |
| nexttoken(); | |
| if (token == '=') { | |
| if (!expectexpr()) return; | |
| begin=pop(); | |
| setvar(&variable, begin); | |
| } | |
| if (token == TTO) { | |
| if (!expectexpr()) return; | |
| to=pop(); | |
| } | |
| if (token == TSTEP) { | |
| if (!expectexpr()) return; | |
| step=pop(); | |
| } | |
| if (!termsymbol()) { | |
| error(EUNKNOWN); | |
| return; | |
| } | |
| /* in interactive mode we reuse here to store the offset in the buffer */ | |
| if (st == SINT) here=bi-ibuffer; | |
| /* here we know everything to set up the loop */ | |
| if (DEBUG) { | |
| outsc("** for loop with parameters var begin end step: "); | |
| outname(&variable); | |
| outspc(); outnumber(begin); | |
| outspc(); outnumber(to); | |
| outspc(); outnumber(step); | |
| outcr(); | |
| outsc("** for loop target location "); outnumber(here); outcr(); | |
| } | |
| pushloop(&variable, TFOR, here, to, step); | |
| if (!USELONGJUMP && er) return; | |
| /* | |
| * This tests the condition and stops if it is fulfilled already from start. | |
| * There is another apocryphal feature here: STEP 0 is legal triggers an infinite loop. | |
| */ | |
| if ((step > 0 && getvar(&variable) > to) || (step < 0 && getvar(&variable) < to)) { | |
| droploop(); | |
| findbraket(TFOR, TNEXT); | |
| nexttoken(); | |
| if (token == VARIABLE) nexttoken(); /* This BASIC does not check. */ | |
| } | |
| } | |
| /* | |
| * BREAK - an apocryphal feature here is the BREAK command ending a loop | |
| */ | |
| void xbreak(){ | |
| bloop_t* loop; | |
| loop=activeloop(); | |
| if (!USELONGJUMP && er) return; | |
| switch (loop->var.token) { | |
| case TWHILE: | |
| findbraket(TWHILE, TWEND); | |
| nexttoken(); | |
| break; | |
| case TREPEAT: | |
| findbraket(TREPEAT, TUNTIL); | |
| while (!termsymbol()) nexttoken(); | |
| break; | |
| default: /* a FOR loop is the default */ | |
| findbraket(TFOR, TNEXT); | |
| nexttoken(); | |
| if (token == VARIABLE) nexttoken(); /* we are at next and skip the variable check */ | |
| break; | |
| } | |
| droploop(); | |
| } | |
| void xbreak(){ | |
| droploop(); | |
| if (!USELONGJUMP && er) return; | |
| findbraket(TFOR, TNEXT); | |
| nexttoken(); | |
| if (token == VARIABLE) nexttoken(); /* we are at next and skip the variable check */ | |
| } | |
| /* | |
| * CONT as a loop control statement, as apocryphal as BREAK, simply | |
| * advance to next and the continue to process | |
| */ | |
| void xcont() { | |
| bloop_t* loop; | |
| loop=activeloop(); | |
| if (!USELONGJUMP && er) return; | |
| switch (loop->var.token) { | |
| case TWHILE: | |
| findbraket(TWHILE, TWEND); | |
| break; | |
| case TREPEAT: | |
| findbraket(TREPEAT, TUNTIL); | |
| break; | |
| default: /* a FOR loop is the default */ | |
| findbraket(TFOR, TNEXT); | |
| break; | |
| } | |
| } | |
| void xcont() { | |
| findbraket(TFOR, TNEXT); | |
| } | |
| /* | |
| * NEXT variable statement. | |
| * | |
| * This code uses the global name variable right now for processing of | |
| * the variable in FOR. The variable name in next is stored in a local variable. | |
| */ | |
| /* reimplementation of xnext without change of the stack in a running loop */ | |
| void xnext(){ | |
| name_t variable; /* this is a potential variable argument of next */ | |
| name_t* forname; /* the name of the variable in the FOR loop */ | |
| token_t t; | |
| address_t h; | |
| number_t value; | |
| bloop_t* loop; | |
| /* check is we have the variable argument */ | |
| nexttoken(); | |
| /* one variable is accepted as an argument, no list */ | |
| if (token == VARIABLE) { | |
| if (DEBUG) { outsc("** variable argument "); outname(&name); outcr(); } | |
| variable=name; | |
| nexttoken(); | |
| if (!termsymbol()) { | |
| error(EUNKNOWN); | |
| return; | |
| } | |
| } else { | |
| variable.c[0]=0; | |
| } | |
| /* see whats going on */ | |
| loop=activeloop(); | |
| if (!USELONGJUMP && er) return; | |
| /* check if this is really a FOR loop */ | |
| if (loop->var.token == TWHILE || loop->var.token == TREPEAT) { error(ELOOP); return; } | |
| /* a variable argument in next clears the for stack | |
| down as BASIC programs can and do jump out to an outer next */ | |
| if (variable.c[0] != 0) { | |
| /* | |
| while (!cmpname(&variable, forname)) { | |
| dropforstack(); | |
| if (!USELONGJUMP && er) return; | |
| forname=peekforstack2(&t, &h, &end, &step); | |
| if (!USELONGJUMP && er) return; | |
| } | |
| */ | |
| while(!cmpname(&variable, &loop->var)) { | |
| droploop(); | |
| if (!USELONGJUMP && er) return; | |
| loop=activeloop(); | |
| if (!USELONGJUMP && er) return; | |
| } | |
| } | |
| /* y=0 an infinite loop with step 0 */ | |
| value=getvar(&loop->var)+loop->step; | |
| setvar(&loop->var, value); | |
| if (DEBUG) { | |
| outsc("** next loop variable "); outname(&loop->var); outspc(); | |
| outsc(" value "); outnumber(value); outcr(); | |
| } | |
| /* do we need another iteration, STEP 0 always triggers an infinite loop */ | |
| if ((loop->step == 0) || (loop->step > 0 && value <= loop->to) || (loop->step < 0 && value >= loop->to)) { | |
| /* iterate in the loop */ | |
| here=loop->here; | |
| /* in interactive mode, jump to the right buffer location */ | |
| if (st == SINT) bi=ibuffer+here; | |
| } else { | |
| /* last iteration completed we stay here after the next, | |
| no precaution for SINT needed as bi unchanged */ | |
| droploop(); | |
| } | |
| nexttoken(); | |
| if (DEBUG) { outsc("** after next found token "); debugtoken(); } | |
| } | |
| /* | |
| * TOKEN output - this is also used in save. | |
| * list does a minimal formatting with a simple heuristic. | |
| * | |
| */ | |
| void outputtoken() { | |
| address_t i; | |
| if (token == EOL) return; | |
| if (token == LINENUMBER) outliteral=0; | |
| if (token == TREM) outliteral=1; | |
| if (spaceafterkeyword) { | |
| if (token != '(' && | |
| token != LINENUMBER && | |
| token !=':' && | |
| token !='$') outspc(); | |
| spaceafterkeyword=0; | |
| } | |
| switch (token) { | |
| case NUMBER: | |
| outnumber(x); | |
| break; | |
| case LINENUMBER: | |
| outnumber(ax); | |
| outspc(); | |
| break; | |
| case ARRAYVAR: | |
| case STRINGVAR: | |
| case VARIABLE: | |
| if (lastouttoken == NUMBER) outspc(); | |
| outname(&name); | |
| if (token == STRINGVAR) outch('$'); | |
| break; | |
| case STRING: | |
| outch('"'); | |
| if (!sr.ir) getstringtobuffer(&sr, spistrbuf1, SPIRAMSBSIZE); | |
| outs(sr.ir, sr.length); | |
| outch('"'); | |
| break; | |
| default: | |
| if ( (token < 32 && token >= BASEKEYWORD) || token < -127) { | |
| if ((token == TTHEN || | |
| token == TELSE || | |
| token == TTO || | |
| token == TSTEP || | |
| token == TGOTO || | |
| token == TGOSUB || | |
| token == TOR || | |
| token == TAND) && lastouttoken != LINENUMBER) outspc(); | |
| else | |
| if (lastouttoken == NUMBER || lastouttoken == VARIABLE) outspc(); | |
| for(i=0; gettokenvalue(i)!=0 && gettokenvalue(i)!=token; i++); | |
| outsc(getkeyword(i)); | |
| if (token != GREATEREQUAL && | |
| token != NOTEQUAL && | |
| token != LESSEREQUAL && | |
| token != TREM && | |
| token != TFN) spaceafterkeyword=1; | |
| break; | |
| } | |
| if (token >= 32) { | |
| outch(token); | |
| if (token == ':' && !outliteral) outspc(); | |
| break; | |
| } | |
| outch(token); outspc(); outnumber(token); | |
| } | |
| lastouttoken=token; | |
| } | |
| /* | |
| * LIST programs to an output device. | |
| * | |
| * The output is formatted to fit the screen, the heuristic is simple. | |
| * | |
| */ | |
| void listlines(address_t b, address_t e) { | |
| mem_t oflag=0; | |
| /* global variables controlling outputtoken, reset to default */ | |
| lastouttoken=0; | |
| spaceafterkeyword=0; | |
| /* if there is a programm ... */ | |
| if ( top != 0 ) { | |
| here=0; | |
| gettoken(); | |
| while (here < top) { | |
| if (token == LINENUMBER && ax >= b) oflag=1; | |
| if (token == LINENUMBER && ax > e) oflag=0; | |
| if (oflag) outputtoken(); | |
| gettoken(); | |
| if (token == LINENUMBER && oflag) { | |
| outcr(); | |
| /* wait after every line on small displays | |
| removed if ( dspactive() && (dsp_rows < 10) ){ if ( inch() == 27 ) break;} */ | |
| if (dspactive() && dspwaitonscroll() == 27) break; | |
| } | |
| } | |
| if (here == top && oflag) outputtoken(); | |
| if (e == maxaddr || b != e) outcr(); /* supress newlines in "list 50" - a little hack */ | |
| } | |
| } | |
| void xlist(){ | |
| number_t b, e; | |
| /* get the argument */ | |
| nexttoken(); | |
| parsearguments(); | |
| if (!USELONGJUMP && er) return; | |
| /* parse the arguments */ | |
| switch (args) { | |
| case 0: | |
| b=0; | |
| e=maxaddr; | |
| break; | |
| case 1: | |
| b=pop(); | |
| e=b; | |
| break; | |
| case 2: | |
| e=pop(); | |
| b=pop(); | |
| break; | |
| default: | |
| error(EARGS); | |
| return; | |
| } | |
| /* list the line from b to e in the default output device */ | |
| listlines(b, e); | |
| /* we are done */ | |
| nexttoken(); | |
| } | |
| /* | |
| * The progam line editor, first version. The code is | |
| * not save for BUFSIZE greater than 127. A cast to | |
| * unsigned char aka uint8_t is needed for the string | |
| * length as some platforms have a signed char and some | |
| * don't. | |
| */ | |
| void xedit(){ | |
| mem_t ood = od; | |
| address_t line; | |
| address_t l; | |
| int i, k, j; | |
| char ch; | |
| /* we edit only in interactive mode */ | |
| if (st != SINT) { error(EUNKNOWN); return; } | |
| /* currently only one line number */ | |
| if(!expectexpr()) { error(EARGS); return; } | |
| /* this uses the input buffer now */ | |
| line=pop(); | |
| undo: /* this is the undo point */ | |
| ibuffer[0]=0; | |
| ibuffer[1]=0; | |
| od=0; | |
| listlines(line, line); | |
| if (ibuffer[0] == 0) { | |
| outnumber(line); | |
| outspc(); | |
| } | |
| od=ood; | |
| /* set the cursor to the first character */ | |
| l=1; | |
| /* editing loop for blocking and non blocking terminals */ | |
| while (-1) { | |
| /* show the line and indicate the cursor */ | |
| for(i=1; i<=(unsigned char)ibuffer[0]; i++) outch(ibuffer[i]); | |
| outcr(); | |
| for(i=0; i<l-1; i++) outspc(); | |
| outch('^'); | |
| outcr(); | |
| /* get a bunch of editing commands and process them*/ | |
| i=ins(sbuffer, SBUFSIZE); | |
| for (k=1; k<=i; k++) { | |
| ch=sbuffer[k]; | |
| switch (ch) { | |
| case 'q': /* quit the editor*/ | |
| goto done; | |
| case 'Q': /* end the editor without saving */ | |
| goto endnosave; | |
| break; | |
| case 'X': /* delete from cursor until the end of the line */ | |
| ibuffer[0]=(char)l; | |
| ibuffer[l]=0; | |
| break; | |
| case 'j': /* vi style left */ | |
| if (l > 1) l--; | |
| break; | |
| case 'k': /* vi style right */ | |
| if (l<(unsigned char)ibuffer[0]) l++; | |
| break; | |
| case 'x': /* delete the cursor character */ | |
| if ((unsigned char)ibuffer[0]>0) { | |
| for (j=l; j<(unsigned char)ibuffer[0]; j++) ibuffer[j]=ibuffer[j+1]; | |
| ibuffer[j]=0; | |
| ibuffer[0]=(unsigned char)ibuffer[0]-1; | |
| } | |
| if ((unsigned char)ibuffer[0]<l) l=(unsigned char)ibuffer[0]; | |
| break; | |
| case 's': /* substitute one character at the cursor position */ | |
| if (k<i) { | |
| k++; | |
| ibuffer[l]=sbuffer[k]; | |
| } | |
| break; | |
| case 'a': /* append multiple characters at the end of the line */ | |
| l=(unsigned char)ibuffer[0]+1; | |
| case 'i': /* insert multiple characters at the cursor position */ | |
| if (i-k+(unsigned char)ibuffer[0] < BUFSIZ) { | |
| for (j=i-k+(unsigned char)ibuffer[0]; j>=l; j--) { | |
| ibuffer[j+i-k]=ibuffer[j]; | |
| if (j<=l+i-k) ibuffer[j]=sbuffer[k+1+(j-l)]; | |
| } | |
| } | |
| ibuffer[0]=(unsigned char)ibuffer[0]+i-k; | |
| k=i; | |
| break; | |
| case '^': /* vi style start of line */ | |
| l=1; | |
| break; | |
| case '$': /* vi style end of line */ | |
| l=(unsigned char)ibuffer[0]+1; | |
| break; | |
| case 'h': /* vi style backspace */ | |
| if (l > 1) { | |
| for (j=l-1; j<(unsigned char)ibuffer[0]; j++) ibuffer[j]=ibuffer[j+1]; | |
| ibuffer[j]=0; | |
| ibuffer[0]=(unsigned char)ibuffer[0]-1; | |
| l--; | |
| } | |
| break; | |
| case 'u': /* vi style undo */ | |
| goto undo; | |
| break; | |
| case ':': /* find the next colon : character*/ | |
| if (l <= (unsigned char)ibuffer[0]) { | |
| while (l <= (unsigned char)ibuffer[0] && ibuffer[l] != ':') l++; | |
| if (l <= (unsigned char)ibuffer[0]) l++; | |
| } | |
| break; | |
| default: /* do nothing if the character is not recogized */ | |
| break; | |
| } | |
| } | |
| } | |
| /* try to store the line, may heaven help us */ | |
| done: | |
| bi=ibuffer; | |
| st=SINT; | |
| nexttoken(); | |
| if (token == NUMBER) { | |
| ax=x; | |
| storeline(); | |
| } | |
| /* and we are done, restore the output device and clean the buffer */ | |
| endnosave: | |
| ibuffer[0]=0; | |
| ibuffer[1]=0; | |
| bi=ibuffer+1; | |
| nexttoken(); | |
| } | |
| /* | |
| * RUN and CONTINUE are the same function | |
| */ | |
| void xrun(){ | |
| if (token == TCONT) { | |
| st=SRUN; | |
| nexttoken(); | |
| } else { | |
| nexttoken(); | |
| parsearguments(); | |
| if (er != 0 ) return; | |
| if (args > 1) { error(EARGS); return; } | |
| if (args == 0) { | |
| here=0; | |
| } else { | |
| findline(pop()); | |
| } | |
| if (!USELONGJUMP && er) return; | |
| if (st == SINT) st=SRUN; | |
| /* all reset on run */ | |
| clrvars(); | |
| clrgosubstack(); | |
| clrforstack(); | |
| clrdata(); | |
| clrlinecache(); | |
| ert=0; | |
| ioer=0; | |
| fncontext=0; | |
| resettimer(&every_timer); | |
| resettimer(&after_timer); | |
| events_enabled=1; | |
| nexttoken(); | |
| } | |
| /* once statement is called it stays into a loop until the token stream | |
| is exhausted. Then we return to interactive mode. */ | |
| statement(); | |
| st=SINT; | |
| /* flush the EEPROM when changing to interactive mode */ | |
| eflush(); | |
| /* if called from command line with file arg - exit after run */ | |
| if (bnointafterrun) restartsystem(); | |
| } | |
| /* | |
| * NEW the general cleanup function - new deletes everything | |
| * | |
| * restbasicstate() is a helper, it keeps the memory intact | |
| * this is needed for EEPROM direct memory | |
| */ | |
| void resetbasicstate() { | |
| if (DEBUG) { outsc("** BASIC state reset \n"); } | |
| /* all stacks are purged */ | |
| clearst(); | |
| clrgosubstack(); | |
| clrforstack(); | |
| clrdata(); | |
| clrvars(); | |
| clrlinecache(); | |
| /* error status reset */ | |
| reseterror(); | |
| /* let here point to the beginning of the program */ | |
| here=0; | |
| /* function context back to zero */ | |
| fncontext=0; | |
| /* interactive mode */ | |
| st=SINT; | |
| /* switch off timers and interrupts */ | |
| resettimer(&after_timer); | |
| resettimer(&every_timer); | |
| } | |
| void xnew(){ | |
| /* reset the state of the interpreter */ | |
| resetbasicstate(); | |
| if (DEBUG) { | |
| outsc("** clearing memory "); | |
| outnumber(memsize); | |
| outsc(" bytes \n"); | |
| } | |
| /* program memory back to zero and variable heap cleared */ | |
| himem=memsize; | |
| zeroblock(0, memsize); | |
| top=0; | |
| if (DEBUG) outsc("** clearing EEPROM state \n "); | |
| /* on EEPROM systems also clear the stored state and top */ | |
| eupdate(0, 0); | |
| setaddress(1, eupdate, top); | |
| } | |
| /* | |
| * REM - skip everything | |
| */ | |
| void xrem() { | |
| if (debuglevel == -1) outsc("REM: "); | |
| while (token != LINENUMBER && token != EOL && here <= top) | |
| { | |
| nexttoken(); | |
| if (debuglevel == -1) { | |
| if (token != LINENUMBER) outputtoken(); else outcr(); | |
| } | |
| } | |
| } | |
| /* | |
| * The Apple 1 BASIC additions | |
| * CLR, DIM, POKE, TAB | |
| */ | |
| /* | |
| * CLR - clearing variable space | |
| */ | |
| void xclr() { | |
| name_t variable; | |
| nexttoken(); | |
| if (termsymbol()) { | |
| clrvars(); | |
| clrgosubstack(); | |
| clrforstack(); | |
| clrdata(); | |
| clrlinecache(); | |
| ert=0; | |
| ioer=0; | |
| } else { | |
| variable=name; | |
| switch (variable.token) { | |
| case VARIABLE: | |
| if (variable.c[0] == '@') { return; } | |
| break; | |
| case ARRAYVAR: | |
| nexttoken(); | |
| if (token != '(') { error(EVARIABLE); return; } | |
| nexttoken(); | |
| if (token != ')') { error(EVARIABLE); return; } | |
| break; | |
| case STRINGVAR: | |
| if (variable.c[0] == '@') { error(EVARIABLE); return; } | |
| break; | |
| case TGOSUB: | |
| clrgosubstack(); | |
| goto next; | |
| case TFOR: | |
| clrforstack(); | |
| goto next; | |
| case TEVERY: | |
| resettimer(&every_timer); | |
| goto next; | |
| case TAFTER: | |
| resettimer(&after_timer); | |
| goto next; | |
| default: | |
| expression(); | |
| if (!USELONGJUMP && er) return; | |
| ax=pop(); | |
| variable.c[0]=ax%256; | |
| variable.c[1]=ax/256; | |
| variable.token=TBUFFER; | |
| } | |
| /* we have to clear an object, call free */ | |
| ax=bfree(&variable); | |
| if (ax == 0) { | |
| if (variable.token != TBUFFER) { error(EVARIABLE); return; } | |
| else ert=1; | |
| } | |
| } | |
| clrvars(); | |
| clrgosubstack(); | |
| clrforstack(); | |
| clrdata(); | |
| clrlinecache(); | |
| ert=0; | |
| ioer=0; | |
| next: | |
| nexttoken(); | |
| } | |
| /* | |
| * DIM - the dimensioning of arrays and strings from Apple 1 BASIC | |
| */ | |
| void xdim(){ | |
| name_t variable; | |
| address_t x; | |
| address_t y=1; | |
| /* which object should be dimensioned or created */ | |
| nexttoken(); | |
| nextvariable: | |
| if (token == ARRAYVAR || token == STRINGVAR ){ | |
| /* remember the object, direct assignment of struct for the moment */ | |
| variable=name; | |
| if (DEBUG) { | |
| outsc("** in xdim "); outname(&variable); outspc(); outnumber(variable.token); | |
| outspc(); outname(&name); outspc(); outnumber(name.token); outcr(); | |
| } | |
| /* parse the arguments */ | |
| parsesubscripts(); | |
| if (!USELONGJUMP && er) return; | |
| if (args != 1) {error(EARGS); return; } | |
| x=popaddress(); | |
| if (args != 1 && args != 2) {error(EARGS); return; } | |
| if (args == 2) y=popaddress(); | |
| x=popaddress(); | |
| /* we create at least one element */ | |
| if (x<1 || y<1) {error(EORANGE); return; } | |
| /* various checks - do we have enough space in buffers and string indices */ | |
| if (variable.token == STRINGVAR) { | |
| if ((x>255) && (strindexsize==1)) {error(EORANGE); return; } | |
| if (x>SPIRAMSBSIZE-1) {error(EORANGE); return; } | |
| /* With the substringmode switched off, if only one argument is given | |
| we interpret the argument as the string array dimension and not as | |
| the length two arguments are allowed and work as always. This makes | |
| things more compatible to the Microsoft BASIC world. */ | |
| if (!substringmode) | |
| if (args == 1) { | |
| y=x; | |
| x=defaultstrdim; | |
| } | |
| (void) createstring(&variable, x, y); | |
| } else { | |
| (void) createarray(&variable, x, y); | |
| } | |
| if (!USELONGJUMP && er) return; | |
| } else if (token == VARIABLE) { | |
| (void) bmalloc(&name, 0); /* this is a local variable, currently no safety net */ | |
| } else { | |
| error(EUNKNOWN); | |
| return; | |
| } | |
| nexttoken(); | |
| if (token == ',') { | |
| nexttoken(); | |
| goto nextvariable; | |
| } | |
| nexttoken(); | |
| } | |
| /* | |
| * POKE - low level poke to the basic memory. | |
| * on 16bit systems, the address is signed, so we can only go up to 32767. | |
| * If the address is negative, we poke into the EEPROM. | |
| */ | |
| void xpoke(){ | |
| number_t a, v; | |
| /* get the address and the value */ | |
| nexttoken(); | |
| parsenarguments(2); | |
| if (!USELONGJUMP && er) return; | |
| v=pop(); /* the value */ | |
| a=pop(); /* the address */ | |
| /* catch memsize here because memwrite doesn't do it */ | |
| if (a >= 0 && a<=memsize) | |
| memwrite2(a, v); | |
| else if (a < 0 && a >= -elength()) | |
| eupdate(-a-1, v); | |
| else { | |
| error(EORANGE); | |
| } | |
| } | |
| /* | |
| * TAB - spaces command of Apple 1 BASIC | |
| * charcount mechanism for relative tab if HASMSTAB is set | |
| */ | |
| void xtab(){ | |
| address_t a; | |
| token_t t = token; | |
| /* get the number of spaces, we allow brackets here to use xtab also in PRINT */ | |
| nexttoken(); | |
| if (token == '(') nexttoken(); | |
| parsenarguments(1); | |
| if (!USELONGJUMP && er) return; | |
| if (token == ')') nexttoken(); | |
| a=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| /* the runtime environment can do a true tab then ... */ | |
| if (t != TSPC && reltab && od <= OPRT && od > 0) { | |
| if (charcount[od-1] >= a) a=0; else a=a-charcount[od-1]-1; | |
| } | |
| while (a-- > 0) outspc(); | |
| } | |
| /* | |
| * locate the curor on the screen | |
| */ | |
| void xlocate() { | |
| address_t cx, cy; | |
| nexttoken(); | |
| parsenarguments(2); | |
| if (!USELONGJUMP && er) return; | |
| cy=popaddress(); | |
| cx=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| /* for locate we go through the VT52 interface for cursor positioning*/ | |
| if (cx > 0 && cy > 0 && cx < 224 && cy < 224) { | |
| outch(27); outch('Y'); | |
| outch(31+(unsigned int) cy); | |
| outch(31+(unsigned int) cx); | |
| } | |
| /* set the charcount, this is half broken on escape sequences */ | |
| if (od > 0 && od <= OPRT) charcount[od-1]=cx; | |
| } | |
| /* | |
| * Stefan's additions to Palo Alto BASIC | |
| * DUMP, SAVE, LOAD, GET, PUT, SET | |
| * | |
| */ | |
| /* | |
| * DUMP - memory dump program | |
| */ | |
| void xdump() { | |
| address_t a, x; | |
| char eflag = 0; | |
| nexttoken(); | |
| if (token == '!') { eflag=1; nexttoken(); } | |
| parsearguments(); | |
| if (!USELONGJUMP && er) return; | |
| switch (args) { | |
| case 0: | |
| x=0; | |
| a=memsize; | |
| break; | |
| case 1: | |
| x=pop(); | |
| a=memsize; | |
| break; | |
| case 2: | |
| a=pop(); | |
| x=pop(); | |
| break; | |
| default: | |
| error(EARGS); | |
| return; | |
| } | |
| form=6; | |
| if (a>x) dumpmem((a-x)/8+1, x, eflag); | |
| form=0; | |
| } | |
| /* | |
| * helper of DUMP, wrote the memory out | |
| */ | |
| void dumpmem(address_t r, address_t b, char eflag) { | |
| address_t j, i; | |
| address_t k; | |
| mem_t c; | |
| address_t end; | |
| k=b; | |
| i=r; | |
| if (eflag) end=elength(); else end=memsize; | |
| while (i>0) { | |
| outnumber(k); outspc(); | |
| for (j=0; j<8; j++) { | |
| if (eflag) c=eread(k); else c=memread(k); | |
| k++; | |
| outnumber(c); outspc(); | |
| if (k > end) break; | |
| } | |
| outcr(); | |
| i--; | |
| if (k > end) break; | |
| } | |
| if (eflag) { | |
| outsc("elength: "); outnumber(elength()); outcr(); | |
| } else { | |
| outsc("top: "); outnumber(top); outcr(); | |
| outsc("himem: "); outnumber(himem); outcr(); | |
| } | |
| } | |
| /* | |
| * creates a C string from a BASIC string | |
| * after reading a BASIC string | |
| */ | |
| void stringtobuffer(char *buffer, string_t* s) { | |
| index_t i = s->length; | |
| if (i >= SBUFSIZE) i=SBUFSIZE-1; | |
| buffer[i--]=0; | |
| while (i >= 0) { buffer[i]=s->ir[i]; i--; } | |
| } | |
| /* helper for the memintercase code */ | |
| void getstringtobuffer(string_t* strp, char *buffer, stringlength_t maxlen) { | |
| stringlength_t i; | |
| for (i=0; i<strp->length && i<maxlen; i++) buffer[i]=memread2(strp->address+i); | |
| strp->ir=buffer; | |
| } | |
| /* get a file argument */ | |
| void getfilename(char *buffer, char d) { | |
| index_t s; | |
| char *sbuffer; | |
| string_t sr; | |
| /* do we have a string argument? */ | |
| s=stringvalue(&sr); | |
| if (!USELONGJUMP && er) return; | |
| if (DEBUG) {outsc("** in getfilename2 stringvalue delivered "); outnumber(s); outcr(); } | |
| if (s) { | |
| if (DEBUG) {outsc("** in getfilename2 copying string of length "); outnumber(x); outcr(); } | |
| if (!sr.ir) getstringtobuffer(&sr, spistrbuf1, SPIRAMSBSIZE); | |
| stringtobuffer(buffer, &sr); | |
| if (DEBUG) {outsc("** in getfilename2 stringvalue string "); outsc(buffer); outcr(); } | |
| nexttoken(); | |
| } else if (termsymbol()) { | |
| if (d) { | |
| sbuffer=getmessage(MFILE); | |
| s=0; | |
| while ((sbuffer[s] != 0) && (s < SBUFSIZE-1)) { buffer[s]=sbuffer[s]; s++; } | |
| buffer[s]=0; | |
| x=s; | |
| } else { | |
| buffer[0]=0; | |
| x=0; | |
| } | |
| nexttoken(); | |
| } else { | |
| expression(); | |
| if (!USELONGJUMP && er) return; | |
| buffer[0]=pop(); | |
| buffer[1]=0; | |
| } | |
| } | |
| /* | |
| * an alternative getfilename implementation that simply gives back a buffer | |
| * with the filename in it. We avoid a string buffer in the calling commands | |
| * like SAVE and LOAD. | |
| */ | |
| char* getfilename2(char d) { | |
| mem_t s; | |
| string_t sr; | |
| /* we have no argument and use the default */ | |
| if (termsymbol()) { | |
| if (d) return getmessage(MFILE); | |
| else return 0; | |
| } | |
| /* we have a string argument or an expression */ | |
| s=stringvalue(&sr); | |
| if (!USELONGJUMP && er) return 0; | |
| if (s) { | |
| if (!sr.ir) getstringtobuffer(&sr, sbuffer, SBUFSIZE); | |
| nexttoken(); /* undo the rewind of stringvalue, only then use the buffer */ | |
| for (s=0; s<sr.length && s<SBUFSIZE-1; s++) sbuffer[s]=sr.ir[s]; | |
| sbuffer[s]=0; | |
| return sbuffer; | |
| } else { | |
| expression(); | |
| if (!USELONGJUMP && er) return 0; | |
| sbuffer[0]=pop(); | |
| sbuffer[1]=0; | |
| return sbuffer; | |
| } | |
| } | |
| /* | |
| * SAVE a file either to disk or to EEPROM | |
| */ | |
| void xsave() { | |
| char *filename; | |
| address_t here2; | |
| token_t t; | |
| nexttoken(); | |
| filename=getfilename2(1); | |
| if (!USELONGJUMP && er) return; | |
| t=token; | |
| if (filename[0] == '!') { | |
| esave(); | |
| } else { | |
| if (DEBUG) { outsc("** Opening the file "); outsc(filename); outcr(); }; | |
| if (!ofileopen(filename, "w")) { | |
| error(EFILE); | |
| return; | |
| } | |
| /* save the output mode and then save */ | |
| push(od); | |
| od=OFILE; | |
| /* the core save - xlist() not used any more */ | |
| ert=0; /* reset error to trap file problems */ | |
| here2=here; | |
| here=0; | |
| gettoken(); | |
| while (here < top) { | |
| outputtoken(); | |
| if (ert) break; | |
| gettoken(); | |
| if (token == LINENUMBER) outcr(); | |
| } | |
| if (here == top) outputtoken(); | |
| outcr(); | |
| /* back to where we were */ | |
| here=here2; | |
| /* restore the output mode */ | |
| od=pop(); | |
| /* clean up */ | |
| ofileclose(); | |
| /* did an accident happen */ | |
| if (ert) { printmessage(EGENERAL); outcr(); ert=0; } | |
| } | |
| /* and continue remembering, where we were */ | |
| token=t; | |
| } | |
| /* | |
| * LOAD a file, LOAD can either be invoked with a filename argument | |
| * or without, in the latter case the filename is read from the token stream | |
| * with getfilename. | |
| */ | |
| void xload(const char* f) { | |
| char* filename; | |
| char ch; | |
| address_t here2; | |
| mem_t chain = 0; | |
| if (f == 0) { | |
| nexttoken(); | |
| filename=getfilename2(1); | |
| if (!USELONGJUMP && er) return; | |
| } else { | |
| filename=(char*)f; | |
| } | |
| if (filename[0] == '!') { | |
| eload(); | |
| } else { | |
| /* if load is called during runtime it chains | |
| * load the program as new but perserve the variables | |
| * gosub and for stacks are cleared | |
| */ | |
| if (st == SRUN) { | |
| chain=1; | |
| st=SINT; | |
| top=0; | |
| clrgosubstack(); | |
| clrforstack(); | |
| clrdata(); | |
| } | |
| if (!f) | |
| if (!ifileopen(filename)) { error(EFILE); return; } | |
| bi=ibuffer+1; | |
| while (fileavailable()) { | |
| ch=fileread(); | |
| if (ch == '\n' || ch == '\r' || cheof(ch)) { | |
| *bi=0; | |
| bi=ibuffer+1; | |
| if (*bi != '#') { /* lines starting with a # are skipped - Unix style shell startup */ | |
| nexttoken(); | |
| if (token == NUMBER) { | |
| ax=x; | |
| storeline(); | |
| } | |
| if (er != 0 ) break; | |
| bi=ibuffer+1; | |
| } | |
| } else { | |
| *bi++=ch; | |
| } | |
| if ((bi-ibuffer) > BUFSIZE) { | |
| error(EOUTOFMEMORY); | |
| break; | |
| } | |
| } | |
| ifileclose(); | |
| /* after a successful load we save top to the EEPROM header */ | |
| setaddress(1, eupdate, top); | |
| /* go back to run mode and start from the first line */ | |
| if (chain) { | |
| st=SRUN; | |
| here=0; | |
| nexttoken(); | |
| } | |
| } | |
| } | |
| /* | |
| * SAVE a file to EEPROM - minimal version for small Arduinos | |
| */ | |
| void xsave() { | |
| esave(); | |
| nexttoken(); | |
| } | |
| /* | |
| * LOAD a file from EEPROM - minimal version for small Arduinos | |
| */ | |
| void xload(const char* f) { | |
| eload(); | |
| nexttoken(); | |
| } | |
| /* | |
| * GET just one character from input | |
| */ | |
| void xget(){ | |
| /* identifier of the lefthandside */ | |
| lhsobject_t lhs; | |
| mem_t oid=id; /* remember the input stream */ | |
| char ch; | |
| nexttoken(); | |
| /* modifiers of the get statement */ | |
| if (token == '&') { | |
| if (!expectexpr()) return; | |
| id=pop(); | |
| if (token != ',') { | |
| error(EUNKNOWN); | |
| return; | |
| } | |
| nexttoken(); | |
| } | |
| /* this code evaluates the left hand side - remember type and name */ | |
| lhs.name=name; | |
| lefthandside(&lhs); | |
| if (!USELONGJUMP && er) return; | |
| /* get the data, non blocking on Arduino */ | |
| if (availch()) ch=inch(); else ch=0; | |
| /* store the data element as a number expect for */ | |
| assignnumber2(&lhs, ch); | |
| /* but then, strings where we deliver a string with length 0 if there is no data */ | |
| if (lhs.name.token == STRINGVAR && ch == 0 && lhs.ps) setstringlength(&lhs.name, 0, arraylimit); | |
| /* restore the output device */ | |
| id=oid; | |
| } | |
| /* | |
| * PUT writes one character to an output stream | |
| */ | |
| void xput(){ | |
| mem_t ood=od; | |
| index_t i; | |
| nexttoken(); | |
| /* modifiers of the put statement */ | |
| if (token == '&') { | |
| if(!expectexpr()) return; | |
| od=pop(); | |
| if (token != ',') { | |
| error(EUNKNOWN); | |
| return; | |
| } | |
| nexttoken(); | |
| } | |
| parsearguments(); | |
| if (!USELONGJUMP && er) return; | |
| for (i=args-1; i>=0; i--) sbuffer[i]=pop(); | |
| outs(sbuffer, args); | |
| od=ood; | |
| } | |
| /* setpersonality is a helper of xset */ | |
| void setpersonality(index_t p) { | |
| switch(p) { | |
| /* a Microsoft like BASIC have arrays starting at 0 with n+1 elements and no substrings, MS type RND */ | |
| case 'm': | |
| case 'M': | |
| msarraylimits=1; | |
| arraylimit=0; | |
| substringmode=0; | |
| booleanmode=-1; | |
| randombase=-1; | |
| reltab=1; | |
| break; | |
| /* an Apple 1 like BASIC have arrays starting at 1 with n elements and substrings */ | |
| case 'a': | |
| case 'A': | |
| msarraylimits=0; | |
| arraylimit=1; | |
| substringmode=1; | |
| booleanmode=1; | |
| randombase=0; | |
| reltab=0; | |
| break; | |
| /* PaloAlto BASIC is an integer basic with slightly different behaviour */ | |
| case 'p': | |
| case 'P': | |
| msarraylimits=0; | |
| arraylimit=0; | |
| substringmode=1; | |
| booleanmode=1; | |
| forceint=1; | |
| randombase=1; | |
| reltab=0; | |
| break; | |
| } | |
| } | |
| /* | |
| * SET - the command itself is also apocryphal it is a low level | |
| * control command setting certain properties | |
| * syntax, currently it is only SET expression, expression | |
| */ | |
| void xset(){ | |
| address_t function; | |
| index_t argument; | |
| nexttoken(); | |
| parsenarguments(2); | |
| if (!USELONGJUMP && er) return; | |
| argument=pop(); | |
| function=pop(); | |
| switch (function) { | |
| /* runtime debug level */ | |
| case 0: | |
| debuglevel=argument; | |
| break; | |
| /* autorun/run flag of the EEPROM 255 for clear, 0 for prog, 1 for autorun */ | |
| case 1: | |
| eupdate(0, argument); | |
| break; | |
| /* change the output device */ | |
| case 2: | |
| switch (argument) { | |
| case 0: | |
| od=OSERIAL; | |
| break; | |
| case 1: | |
| od=ODSP; | |
| break; | |
| } | |
| break; | |
| /* change the default output device */ | |
| case 3: | |
| switch (argument) { | |
| case 0: | |
| od=(odd=OSERIAL); | |
| break; | |
| case 1: | |
| od=(odd=ODSP); | |
| break; | |
| } | |
| break; | |
| /* change the input device */ | |
| case 4: | |
| switch (argument) { | |
| case 0: | |
| id=ISERIAL; | |
| break; | |
| case 1: | |
| id=IKEYBOARD; | |
| break; | |
| } | |
| break; | |
| /* change the default input device */ | |
| case 5: | |
| switch (argument) { | |
| case 0: | |
| idd=(id=ISERIAL); | |
| break; | |
| case 1: | |
| idd=(id=IKEYBOARD); | |
| break; | |
| } | |
| break; | |
| /* set the cr behaviour */ | |
| case 6: | |
| sendcr=(char)argument; | |
| break; | |
| /* set the blockmode behaviour */ | |
| case 7: | |
| blockmode=argument; | |
| break; | |
| /* set the second serial ports baudrate */ | |
| case 8: | |
| prtset(argument); | |
| break; | |
| /* set the power amplifier level of the radio module */ | |
| case 9: | |
| radioset(argument); | |
| break; | |
| /* display update control for paged displays */ | |
| case 10: | |
| dspsetupdatemode(argument); | |
| break; | |
| /* change the output device to a true TAB */ | |
| case 11: | |
| reltab=argument; | |
| break; | |
| /* change the lower array limit */ | |
| case 12: | |
| if (argument>=0) arraylimit=argument; else error(EORANGE); | |
| break; | |
| /* the keyboard repeat frequency */ | |
| case 13: | |
| kbdrepeat=argument; | |
| break; | |
| /* the units the pulse command is using */ | |
| case 14: | |
| bpulseunit=argument; | |
| break; | |
| /* switch on the vt52 emulation an a POSIX system with an ANSI terminal */ | |
| case 15: | |
| vt52active=argument; | |
| break; | |
| /* change the default size of a string at autocreate */ | |
| case 16: | |
| if (argument>0) defaultstrdim=argument; else error(EORANGE); | |
| break; | |
| /* set the boolean mode */ | |
| case 17: | |
| if (argument==-1 || argument==1) booleanmode=argument; else error(EORANGE); | |
| break; | |
| /* set the integer mode */ | |
| case 18: | |
| forceint=(argument != 0); | |
| break; | |
| /* set the random number behaviour */ | |
| case 19: | |
| randombase=argument; | |
| break; | |
| /* the substring mode on and off */ | |
| case 20: | |
| substringmode=(argument != 0); | |
| break; | |
| /* the MS array behaviour, creates n+1 elements when on */ | |
| case 21: | |
| msarraylimits=(argument != 0); | |
| break; | |
| /* set many settings at once to change the entire personality of the interpreter */ | |
| case 22: | |
| setpersonality(argument); | |
| break; | |
| case 23: | |
| lowercasenames=(argument != 0); | |
| break; | |
| } | |
| } | |
| /* | |
| * NETSTAT - network status command, rudimentary | |
| */ | |
| void xnetstat(){ | |
| nexttoken(); | |
| parsearguments(); | |
| if (!USELONGJUMP && er) return; | |
| switch (args) { | |
| case 0: | |
| if (netconnected()) outsc("Network connected \n"); else outsc("Network not connected \n"); | |
| outsc("MQTT state "); outnumber(mqttstate()); outcr(); | |
| outsc("MQTT out topic "); outsc(mqtt_otopic); outcr(); | |
| outsc("MQTT inp topic "); outsc(mqtt_itopic); outcr(); | |
| outsc("MQTT name "); outsc(mqttname); outcr(); | |
| break; | |
| case 1: | |
| ax=pop(); | |
| switch (ax) { | |
| case 0: | |
| netstop(); | |
| break; | |
| case 1: | |
| netbegin(); | |
| break; | |
| case 2: | |
| if (!mqttreconnect()) ert=1; | |
| break; | |
| default: | |
| error(EARGS); | |
| return; | |
| } | |
| break; | |
| default: | |
| error(EARGS); | |
| return; | |
| } | |
| nexttoken(); | |
| } | |
| /* | |
| * The arduino io functions. | |
| * | |
| */ | |
| /* | |
| * rtaread and rtdread are wrappers coming from runtime | |
| * This is done for portability for raspberry pi and other systems | |
| */ | |
| void xaread(){ | |
| push(aread(popaddress())); | |
| } | |
| void xdread(){ | |
| push(dread(popaddress())); | |
| } | |
| /* | |
| * DWRITE - digital write | |
| */ | |
| void xdwrite(){ | |
| address_t x,y; | |
| nexttoken(); | |
| parsenarguments(2); | |
| if (!USELONGJUMP && er) return; | |
| x=popaddress(); | |
| y=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| dwrite(y, x); | |
| } | |
| /* | |
| * AWRITE - analog write | |
| */ | |
| void xawrite(){ | |
| address_t x,y; | |
| nexttoken(); | |
| parsenarguments(2); | |
| if (!USELONGJUMP && er) return; | |
| x=popaddress(); | |
| if (x > 255) error(EORANGE); | |
| y=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| awrite(y, x); | |
| } | |
| /* | |
| * PINM - pin mode | |
| */ | |
| void xpinm(){ | |
| address_t x,y; | |
| nexttoken(); | |
| parsenarguments(2); | |
| if (!USELONGJUMP && er) return; | |
| x=popaddress(); | |
| if (x > 1) error(EORANGE); | |
| y=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| pinm(y, x); | |
| } | |
| /* | |
| * DELAY in milliseconds | |
| * | |
| * this must call bdelay() and not delay() as bdelay() | |
| * handles all the yielding and timing functions | |
| * | |
| */ | |
| void xdelay(){ | |
| nexttoken(); | |
| parsenarguments(1); | |
| if (!USELONGJUMP && er) return; | |
| bdelay(pop()); | |
| } | |
| /* tone if the platform has it -> BASIC command PLAY */ | |
| /* play a tone */ | |
| void xtone(){ | |
| address_t d = 0; | |
| address_t v = 100; | |
| address_t f, p; | |
| /* get minimum of 2 and maximum of 4 args */ | |
| nexttoken(); | |
| parsearguments(); | |
| if (!USELONGJUMP && er) return; | |
| if (args > 4 || args < 2) { error(EARGS); return; } | |
| /* a switch would be more elegant but needs more progspace ;-) */ | |
| if (args == 4) v=popaddress(); | |
| if (args >= 3) d=popaddress(); | |
| f=popaddress(); | |
| p=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| playtone(p, f, d, v); | |
| } | |
| /* pulse output - pin, duration, [value], [repetitions, delay] */ | |
| void xpulse(){ | |
| address_t pin, duration; | |
| address_t val = 1; | |
| address_t interval = 0; | |
| address_t repetition = 1; | |
| /* do we have at least 2 and not more than 5 arguments */ | |
| nexttoken(); | |
| parsearguments(); | |
| if (!USELONGJUMP && er) return; | |
| if (args>5 || args<2) { error(EARGS); return; } | |
| /* get the data from stack */ | |
| if (args == 5) { interval=popaddress(); repetition=popaddress(); } | |
| if (args == 4) { error(EARGS); return; } | |
| if (args > 2) val=popaddress(); | |
| duration=popaddress(); | |
| pin=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| /* low level run time function for the pulse */ | |
| pulseout(bpulseunit, pin, duration, val, repetition, interval); | |
| } | |
| /* read a pulse, units given by bpulseunit - default 10 microseconds */ | |
| void bpulsein() { | |
| address_t x,y; | |
| unsigned long t, pt; | |
| t=((unsigned long) popaddress())*1000; | |
| y=popaddress(); | |
| x=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| push(pulsein(x, y, t)/bpulseunit); | |
| } | |
| /* | |
| * COLOR setting, accepting one or 3 arguments | |
| */ | |
| void xcolor() { | |
| int r, g, b; | |
| nexttoken(); | |
| parsearguments(); | |
| if (!USELONGJUMP && er) return; | |
| switch(args) { | |
| case 1: | |
| vgacolor(pop()); | |
| break; | |
| case 3: | |
| b=pop(); | |
| g=pop(); | |
| r=pop(); | |
| rgbcolor(r, g, b); | |
| break; | |
| default: | |
| error(EARGS); | |
| break; | |
| } | |
| } | |
| /* | |
| * PLOT a pixel on the screen | |
| */ | |
| void xplot() { | |
| int x0, y0; | |
| nexttoken(); | |
| parsenarguments(2); | |
| if (!USELONGJUMP && er) return; | |
| y0=pop(); | |
| x0=pop(); | |
| plot(x0, y0); | |
| } | |
| /* | |
| * LINE draws a line | |
| */ | |
| void xline() { | |
| int x0, y0, x1, y1; | |
| nexttoken(); | |
| parsenarguments(4); | |
| if (!USELONGJUMP && er) return; | |
| y1=pop(); | |
| x1=pop(); | |
| y0=pop(); | |
| x0=pop(); | |
| line(x0, y0, x1, y1); | |
| } | |
| void xrect() { | |
| int x0, y0, x1, y1; | |
| nexttoken(); | |
| parsenarguments(4); | |
| if (!USELONGJUMP && er) return; | |
| y1=pop(); | |
| x1=pop(); | |
| y0=pop(); | |
| x0=pop(); | |
| rect(x0, y0, x1, y1); | |
| } | |
| void xcircle() { | |
| int x0, y0, r; | |
| nexttoken(); | |
| parsenarguments(3); | |
| if (!USELONGJUMP && er) return; | |
| r=pop(); | |
| y0=pop(); | |
| x0=pop(); | |
| circle(x0, y0, r); | |
| } | |
| void xfrect() { | |
| int x0, y0, x1, y1; | |
| nexttoken(); | |
| parsenarguments(4); | |
| if (!USELONGJUMP && er) return; | |
| y1=pop(); | |
| x1=pop(); | |
| y0=pop(); | |
| x0=pop(); | |
| frect(x0, y0, x1, y1); | |
| } | |
| void xfcircle() { | |
| int x0, y0, r; | |
| nexttoken(); | |
| parsenarguments(3); | |
| if (!USELONGJUMP && er) return; | |
| r=pop(); | |
| y0=pop(); | |
| x0=pop(); | |
| fcircle(x0, y0, r); | |
| } | |
| /* | |
| * MALLOC allocates a chunk of memory | |
| */ | |
| void xmalloc() { | |
| address_t s; | |
| address_t a; | |
| name_t name; | |
| /* size and identifier */ | |
| s=popaddress(); | |
| a=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| /* create a name */ | |
| name.token=TBUFFER; | |
| name.l=2; | |
| name.c[0]=a%256; | |
| name.c[1]=a/256; | |
| /* allocate the memory */ | |
| push(bmalloc(&name, s)); | |
| } | |
| /* | |
| * FIND an object on the heap | |
| * xfind can find things in the variable name space and the buffer space | |
| */ | |
| void xfind() { | |
| address_t a; | |
| address_t n; | |
| /* is there a ( */ | |
| if (!expect('(', EUNKNOWN)) return; | |
| /* after that, try to find the object on the heap */ | |
| nexttoken(); | |
| if (token == TFN) { | |
| nexttoken(); | |
| name.token=TFN; | |
| } | |
| a=bfind(&name); | |
| /* depending on the object, interpret the result */ | |
| switch (token) { | |
| case ARRAYVAR: | |
| case TFN: | |
| if (!expect('(', EUNKNOWN)) return; | |
| if (!expect(')', EUNKNOWN)) return; | |
| case VARIABLE: | |
| case STRINGVAR: | |
| nexttoken(); | |
| break; | |
| default: | |
| expression(); /* do not use expectexpr here because of the token sequence */ | |
| if (!USELONGJUMP && er) return; | |
| n=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| name.token=TBUFFER; | |
| name.l=2; | |
| name.c[0]=n%256; | |
| name.c[1]=n/256; | |
| a=bfind(&name); | |
| } | |
| /* closing braket, dont use expect here because of the token sequence */ | |
| if (token != ')') { error(EUNKNOWN); return; } | |
| push(a); | |
| } | |
| /* | |
| * EVAL can modify a program, there are serious side effects | |
| * which are not caught (and cannot be). All FOR loops and RETURN | |
| * vectors break if EVAL inserts in their range | |
| */ | |
| void xeval(){ | |
| address_t i, l; | |
| address_t mline, line; | |
| string_t s; | |
| /* get the line number to store */ | |
| if (!expectexpr()) return; | |
| line=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| if (token != ',') { error(EUNKNOWN); return; } | |
| /* the line to be stored */ | |
| nexttoken(); | |
| if (!stringvalue(&s)) { error(EARGS); return; } | |
| /* here we have the string to evaluate it to the ibuffer | |
| only one line allowed, BUFSIZE is the limit */ | |
| l=s.length; | |
| if (!USELONGJUMP && er) return; | |
| if (l>BUFSIZE-1) {error(EORANGE); return; } | |
| if (!s.ir) getstringtobuffer(&s, spistrbuf1, SPIRAMSBSIZE); | |
| for (i=0; i<l; i++) ibuffer[i+1]=s.ir[i]; | |
| ibuffer[l+1]=0; | |
| if (DEBUG) {outsc("** Preparing to store line "); outnumber(line); outspc(); outsc(ibuffer+1); outcr(); } | |
| /* we find the line we are currently at */ | |
| if (st != SINT) { | |
| mline=myline(here); | |
| if (DEBUG) {outsc("** myline is "); outnumber(mline); outcr(); } | |
| } | |
| /* go to interactive mode and try to store the line */ | |
| ax=line; | |
| push(st); st=SINT; | |
| bi=ibuffer; | |
| storeline(); | |
| st=pop(); | |
| /* find my line - side effects not checked here */ | |
| if (st != SINT) { | |
| findline(mline); | |
| nextline(); | |
| } | |
| } | |
| /* | |
| * AVAIL of a stream - are there characters in the stream | |
| */ | |
| void xavail() { | |
| mem_t oid=id; | |
| id=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| push(availch()); | |
| id=oid; | |
| } | |
| /* | |
| * IoT functions - sensor reader, experimentral | |
| */ | |
| void xfsensor() { | |
| address_t s, a; | |
| a=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| s=popaddress(); | |
| if (!USELONGJUMP && er) return; | |
| push(sensorread(s, a)); | |
| } | |
| /* | |
| * Going to sleep for battery saving - implemented for ESP8266 and ESP32 | |
| * in hardware-*.h | |
| */ | |
| void xsleep() { | |
| nexttoken(); | |
| parsenarguments(1); | |
| if (!USELONGJUMP && er) return; | |
| activatesleep(pop()); | |
| } | |
| /* | |
| * single byte wire access - keep it simple | |
| */ | |
| void xwire() { | |
| short port, data1, data2; | |
| nexttoken(); | |
| parsearguments(); | |
| if (!USELONGJUMP && er) return; | |
| if (args == 3) { | |
| data2=pop(); | |
| data1=pop(); | |
| port=pop(); | |
| wirewriteword(port, data1, data2); | |
| } else if (args == 2) { | |
| data1=pop(); | |
| port=pop(); | |
| wirewritebyte(port, data1); | |
| } else { | |
| error(EARGS); | |
| return; | |
| } | |
| } | |
| void xfwire() { | |
| push(wirereadbyte(pop())); | |
| } | |
| /* | |
| * Error handling function. | |
| */ | |
| void xerror() { | |
| berrorh.type=0; | |
| erh=0; | |
| nexttoken(); | |
| switch (token) { | |
| case TGOTO: | |
| if (!expectexpr()) return; | |
| berrorh.type=TGOTO; | |
| berrorh.linenumber=pop(); | |
| break; | |
| case TCONT: | |
| berrorh.type=TCONT; | |
| case TSTOP: | |
| nexttoken(); | |
| break; | |
| default: | |
| error(EARGS); | |
| return; | |
| } | |
| } | |
| /* | |
| * After and every trigger timing GOSUBS and GOTOS. | |
| */ | |
| void resettimer(btimer_t* t) { | |
| t->enabled=0; | |
| t->interval=0; | |
| t->last=0; | |
| t->type=0; | |
| t->linenumber=0; | |
| } | |
| void xtimer() { | |
| token_t t; | |
| btimer_t* timer; | |
| /* do we deal with every or after */ | |
| if (token == TEVERY) timer=&every_timer; else timer=&after_timer; | |
| /* one argument expected, the time intervall */ | |
| if (!expectexpr()) return; | |
| /* after that, a command GOTO or GOSUB with a line number | |
| more commands thinkable */ | |
| switch(token) { | |
| case TGOSUB: | |
| case TGOTO: | |
| t=token; | |
| if (!expectexpr()) return; | |
| timer->last=millis(); | |
| timer->type=t; | |
| timer->linenumber=pop(); | |
| timer->interval=pop(); | |
| timer->enabled=1; | |
| break; | |
| default: | |
| if (termsymbol()) { | |
| x=pop(); | |
| if (x == 0) | |
| timer->enabled=0; | |
| else { | |
| if (timer->linenumber) { | |
| timer->enabled=1; | |
| timer->interval=x; | |
| timer->last=millis(); | |
| } else | |
| error(EARGS); | |
| } | |
| } else | |
| error(EUNKNOWN); | |
| return; | |
| } | |
| } | |
| /* | |
| * The events API for Arduino with interrupt service routines | |
| * analogous to the timer API. | |
| * | |
| * we use raw modes here | |
| * | |
| * #define CHANGE 1 | |
| * #define FALLING 2 | |
| * #define RISING 3 | |
| * | |
| * detach und attach are wrappers around the original Arduino functions. | |
| * Runtime needs to be compiled with #define ARDUINOINTERRUPTS for this to | |
| * work. | |
| * | |
| */ | |
| /* interrupts in BASIC fire once and then disable themselves, BASIC reenables them */ | |
| void bintroutine0() { | |
| eventlist[0].active=1; | |
| detachinterrupt(eventlist[0].pin); | |
| } | |
| void bintroutine1() { | |
| eventlist[1].active=1; | |
| detachinterrupt(eventlist[1].pin); | |
| } | |
| void bintroutine2() { | |
| eventlist[2].active=1; | |
| detachinterrupt(eventlist[2].pin); | |
| } | |
| void bintroutine3() { | |
| eventlist[3].active=1; | |
| detachinterrupt(eventlist[3].pin); | |
| } | |
| mem_t eventindex(mem_t pin) { | |
| mem_t i; | |
| for(i=0; i<EVENTLISTSIZE; i++ ) if (eventlist[i].pin == pin) return i; | |
| return -1; | |
| } | |
| mem_t enableevent(mem_t pin){ | |
| mem_t inter; | |
| mem_t i; | |
| /* do we have the data */ | |
| if ((i=eventindex(pin))<0) return 0; | |
| /* can we use this pin? */ | |
| inter=pintointerrupt(eventlist[i].pin); | |
| if (inter < 0) return 0; | |
| /* attach the interrupt function to this pin */ | |
| switch(i) { | |
| case 0: | |
| attachinterrupt(inter, bintroutine0, eventlist[i].mode); | |
| break; | |
| case 1: | |
| attachinterrupt(inter, bintroutine1, eventlist[i].mode); | |
| break; | |
| case 2: | |
| attachinterrupt(inter, bintroutine2, eventlist[i].mode); | |
| break; | |
| case 3: | |
| attachinterrupt(inter, bintroutine3, eventlist[i].mode); | |
| break; | |
| default: | |
| return 0; | |
| } | |
| /* now set it enabled in BASIC */ | |
| eventlist[i].enabled=1; | |
| return 1; | |
| } | |
| void disableevent(mem_t pin) { | |
| detachinterrupt(pin); | |
| } | |
| /* the event BASIC commands */ | |
| void initevents() { | |
| mem_t i; | |
| for(i=0; i<EVENTLISTSIZE; i++) eventlist[i].pin=-1; | |
| } | |
| void xevent() { | |
| mem_t pin, mode; | |
| mem_t type=0; | |
| address_t line=0; | |
| /* in this version two arguments are neded, one is the pin, the second the mode */ | |
| nexttoken(); | |
| /* debug code, display the event list */ | |
| if (termsymbol()) { | |
| for (ax=0; ax<EVENTLISTSIZE; ax++) { | |
| if (eventlist[ax].pin >= 0) { | |
| outnumber(eventlist[ax].pin); outspc(); | |
| outnumber(eventlist[ax].mode); outspc(); | |
| outnumber(eventlist[ax].type); outspc(); | |
| outnumber(eventlist[ax].linenumber); outspc(); | |
| outcr(); | |
| } | |
| } | |
| outnumber(nevents); outcr(); | |
| nexttoken(); | |
| return; | |
| } | |
| /* control of events */ | |
| if (token == TSTOP) { | |
| events_enabled=0; | |
| nexttoken(); | |
| return; | |
| } | |
| if (token == TCONT) { | |
| events_enabled=1; | |
| nexttoken(); | |
| return; | |
| } | |
| /* argument parsing */ | |
| parsearguments(); | |
| if (!USELONGJUMP && er) return; | |
| switch(args) { | |
| case 2: | |
| mode=pop(); | |
| if (mode > 3) { | |
| error(EARGS); | |
| return; | |
| } | |
| case 1: | |
| pin=pop(); | |
| break; | |
| default: | |
| error(EARGS); | |
| } | |
| /* followed by termsymbol, GOTO or GOSUB */ | |
| if (token == TGOTO || token == TGOSUB) { | |
| type=token; | |
| /* which line to go to */ | |
| if (!expectexpr()) return; | |
| line=pop(); | |
| } | |
| /* all done either set the interrupt up or delete it*/ | |
| if (type) { | |
| if (!addevent(pin, mode, type, line)) { | |
| error(EARGS); | |
| return; | |
| } | |
| } else { | |
| disableevent(pin); | |
| deleteevent(pin); | |
| return; | |
| } | |
| /* enable the interrupt */ | |
| if (!enableevent(pin)) { | |
| deleteevent(pin); | |
| error(EARGS); | |
| return; | |
| } | |
| } | |
| /* handling the event list */ | |
| mem_t addevent(mem_t pin, mem_t mode, mem_t type, address_t linenumber) { | |
| int i; | |
| /* is the event already there */ | |
| for (i=0; i<EVENTLISTSIZE; i++) | |
| if (pin == eventlist[i].pin) goto slotfound; | |
| /* if not, look for a free slot */ | |
| if (nevents >= EVENTLISTSIZE) return 0; | |
| for (i=0; i<EVENTLISTSIZE; i++) | |
| if (eventlist[i].pin == -1) goto slotfound; | |
| /* no free event slot */ | |
| return 0; | |
| /* we have a slot */ | |
| slotfound: | |
| eventlist[i].enabled=0; | |
| eventlist[i].pin=pin; | |
| eventlist[i].mode=mode; | |
| eventlist[i].type=type; | |
| eventlist[i].linenumber=linenumber; | |
| eventlist[i].active=0; | |
| nevents++; | |
| return 1; | |
| } | |
| void deleteevent(mem_t pin) { | |
| int i; | |
| /* do we have the event? */ | |
| i=eventindex(pin); | |
| if (i >= 0) { | |
| eventlist[i].enabled=0; | |
| eventlist[i].pin=-1; | |
| eventlist[i].mode=0; | |
| eventlist[i].type=0; | |
| eventlist[i].linenumber=0; | |
| eventlist[i].active=0; | |
| nevents--; | |
| } | |
| } | |
| /* | |
| * BASIC DOS - disk access programs, to control mass storage from BASIC | |
| */ | |
| /* string match helper in catalog */ | |
| char streq(const char *s, char *m){ | |
| short i=0; | |
| while (m[i]!=0 && s[i]!=0 && i < SBUFSIZE){ | |
| if (s[i] != m[i]) return 0; | |
| i++; | |
| } | |
| return 1; | |
| } | |
| /* | |
| * CATALOG - basic directory function | |
| */ | |
| void xcatalog() { | |
| char filename[SBUFSIZE]; | |
| const char *name; | |
| nexttoken(); | |
| getfilename(filename, 0); | |
| if (!USELONGJUMP && er) return; | |
| rootopen(); | |
| while (rootnextfile()) { | |
| if (rootisfile()) { | |
| name=rootfilename(); | |
| if (*name != '_' && *name !='.' && streq(name, filename)){ | |
| outscf(name, 14); outspc(); | |
| if (rootfilesize()>0) outnumber(rootfilesize()); | |
| outcr(); | |
| if (dspwaitonscroll() == 27) break; | |
| } | |
| } | |
| rootfileclose(); | |
| } | |
| rootclose(); | |
| nexttoken(); | |
| } | |
| /* | |
| * DELETE a file | |
| */ | |
| void xdelete() { | |
| char filename[SBUFSIZE]; | |
| nexttoken(); | |
| getfilename(filename, 0); | |
| if (!USELONGJUMP && er) return; | |
| removefile(filename); | |
| nexttoken(); | |
| } | |
| /* | |
| * OPEN a file or I/O stream - very raw mix of different functions | |
| */ | |
| void xopen() { | |
| char stream = IFILE; /* default is file operation */ | |
| char* filename; | |
| int mode; | |
| /* which stream do we open? default is FILE */ | |
| nexttoken(); | |
| if (token == '&') { | |
| if (!expectexpr()) return; | |
| stream=pop(); | |
| if (token != ',') {error(EUNKNOWN); return; } | |
| nexttoken(); | |
| } | |
| /* the filename and its length */ | |
| filename = getfilename2(0); | |
| if (!USELONGJUMP && er) return; | |
| /* and the arguments */ | |
| args=0; | |
| if (token == ',') { | |
| nexttoken(); | |
| parsearguments(); | |
| } | |
| /* getting an argument, no argument is read, i.e. mode 0 */ | |
| if (args == 0 ) { | |
| mode=0; | |
| } else if (args == 1) { | |
| mode=pop(); | |
| } else { | |
| error(EARGS); | |
| return; | |
| } | |
| /* open the stream */ | |
| switch(stream) { | |
| case ISERIAL1: | |
| prtclose(); | |
| if (mode == 0) mode=9600; | |
| if (prtopen(filename, mode)) ert=0; else ert=1; | |
| break; | |
| case IFILE: | |
| switch (mode) { | |
| case 1: | |
| ofileclose(); | |
| if (ofileopen(filename, "w")) ert=0; else ert=1; | |
| break; | |
| case 2: | |
| ofileclose(); | |
| if (ofileopen(filename, "a")) ert=0; else ert=1; | |
| break; | |
| default: | |
| case 0: | |
| ifileclose(); | |
| if (ifileopen(filename)) ert=0; else ert=1; | |
| break; | |
| } | |
| break; | |
| case IRADIO: | |
| if (mode == 0) { | |
| iradioopen(filename); | |
| } else if (mode == 1) { | |
| oradioopen(filename); | |
| } | |
| break; | |
| case IWIRE: | |
| wireopen(filename[0], mode); | |
| break; | |
| case IMQTT: | |
| if (mode == 0) { | |
| mqttsubscribe(filename); | |
| } else if (mode == 1) { | |
| mqttsettopic(filename); | |
| } | |
| break; | |
| default: | |
| error(EORANGE); | |
| return; | |
| } | |
| nexttoken(); | |
| } | |
| /* | |
| * OPEN as a function, currently only implemented for MQTT | |
| */ | |
| void xfopen() { | |
| address_t stream = popaddress(); | |
| if (stream == 9) push(mqttstate()); else push(0); | |
| } | |
| /* | |
| * CLOSE a file or stream | |
| */ | |
| void xclose() { | |
| char stream = IFILE; | |
| char mode; | |
| nexttoken(); | |
| if (token == '&') { | |
| if (!expectexpr()) return; | |
| stream=pop(); | |
| if (token != ',' && ! termsymbol()) {error(EUNKNOWN); return; } | |
| nexttoken(); | |
| } | |
| parsearguments(); | |
| if (args == 0) { | |
| mode=0; | |
| } else if (args == 1) { | |
| mode=pop(); | |
| } else { | |
| error(EARGS); | |
| return; | |
| } | |
| /* currently only close of files is implemented, should be also implemented for Wire */ | |
| switch(stream) { | |
| case IFILE: | |
| if (mode == 1 || mode == 2) ofileclose(); else if (mode == 0) ifileclose(); | |
| break; | |
| } | |
| nexttoken(); | |
| } | |
| /* | |
| * FDISK - format internal disk storages of RP2040, ESP and the like | |
| */ | |
| void xfdisk() { | |
| nexttoken(); | |
| parsearguments(); | |
| if (!USELONGJUMP && er) return; | |
| if (args > 1) error(EORANGE); | |
| if (args == 0) push(0); | |
| outsc("Format disk (y/N)?"); | |
| (void) consins(sbuffer, SBUFSIZE); | |
| if (sbuffer[1] == 'y') formatdisk(pop()); | |
| if (fsstat(1) > 0) outsc("ok\n"); else outsc("fail\n"); | |
| nexttoken(); | |
| } | |
| /* | |
| * USR low level function access of the interpreter | |
| * for each group of functions there is a call vector | |
| * and and argument. | |
| * | |
| * USR arguments from 0 to 31 are reserved for the | |
| * interpreter status and the input output stream | |
| * mechanisms. All other values are free and can be used | |
| * for individual functions - see case 32 for an example. | |
| */ | |
| void xusr() { | |
| address_t fn; | |
| number_t v; | |
| int arg; | |
| address_t a; | |
| v=pop(); | |
| arg=(int)v; /* a bit paranoid here */ | |
| fn=pop(); | |
| switch(fn) { | |
| /* USR(0,y) delivers all the internal constants and variables or the interpreter */ | |
| case 0: | |
| switch(arg) { | |
| case 0: | |
| push(bsystype); | |
| break; | |
| case 1: /* language set identifier, odd because USR is part of STEFANSEXT*/ | |
| a=0; | |
| a|=1; | |
| a|=2; | |
| a|=4; | |
| a|=8; | |
| a|=16; | |
| a|=32; | |
| a|=64; | |
| push(a); break; | |
| case 2: | |
| push(0); break; /* reserved for system speed identifier */ | |
| case 3: push(-1); break; | |
| case 3: push(0); break; | |
| case 4: push(numsize); break; | |
| case 5: push(maxnum); break; | |
| case 6: push(addrsize); break; | |
| case 7: push(maxaddr); break; | |
| case 8: push(strindexsize); break; | |
| case 9: push(memsize+1); break; | |
| case 10: push(elength()); break; | |
| case 11: push(GOSUBDEPTH); break; | |
| case 12: push(FORDEPTH); break; | |
| case 13: push(STACKSIZE); break; | |
| case 14: push(BUFSIZE); break; | |
| case 15: push(SBUFSIZE); break; | |
| case 16: push(ARRAYSIZEDEF); break; | |
| case 17: push(defaultstrdim); break; | |
| /* - 24 reserved, don't use */ | |
| case 24: push(top); break; | |
| case 25: push(here); break; | |
| case 26: push(himem); break; | |
| case 27: push(0); break; | |
| case 28: push(freeRam()); break; | |
| case 29: push(gosubsp); break; | |
| case 30: push(loopsp); break; | |
| case 31: push(0); break; /* fnc removed as interpreter variable */ | |
| case 32: push(sp); break; | |
| case 33: push(data); break; | |
| case 33: push(0); break; | |
| case 34: push(0); break; | |
| case 35: push(avgfasttick); break; | |
| /* - 48 reserved, don't use */ | |
| case 48: push(id); break; | |
| case 49: push(idd); break; | |
| case 50: push(od); break; | |
| case 51: push(odd); break; | |
| default: push(0); | |
| } | |
| break; | |
| /* access to properties of stream 1 - serial */ | |
| case 1: | |
| push(serialstat(arg)); | |
| break; | |
| /* access to properties of stream 2 - display and keyboard */ | |
| case 2: | |
| push(dspstat(arg)); | |
| push(vgastat(arg)); | |
| push(0); | |
| break; | |
| /* access to properties of stream 4 - printer */ | |
| case 4: | |
| push(prtstat(arg)); | |
| break; | |
| /* access to properties of stream 7 - wire */ | |
| case 7: | |
| push(wirestat(arg)); | |
| break; | |
| /* access to properties of stream 8 - radio */ | |
| case 8: | |
| push(radiostat(arg)); | |
| break; | |
| /* access to properties of stream 9 - mqtt */ | |
| case 9: | |
| push(mqttstat(arg)); | |
| break; | |
| /* access to properties of stream 16 - file */ | |
| case 16: | |
| push(fsstat(arg)); | |
| break; | |
| case 32: | |
| /* user function 32 and beyond can be used freely */ | |
| /* all USR values not assigned return 0 */ | |
| default: | |
| if (fn>31) push(usrfunction(fn, v)); else push(0); | |
| } | |
| } | |
| /* | |
| * CALL currently only to exit the interpreter | |
| */ | |
| void xcall() { | |
| int r; | |
| if (!expectexpr()) return; | |
| r=pop(); | |
| switch(r) { | |
| case 0: | |
| /* flush the EEPROM dummy and the output file and then exit */ | |
| eflush(); | |
| ofileclose(); | |
| vgaend(); /* clean up if you have played with the framebuffer */ | |
| restartsystem(); | |
| break; | |
| /* restart the filesystem - only test code */ | |
| case 1: | |
| fsbegin(); | |
| break; | |
| /* call values to 31 reserved! */ | |
| default: | |
| /* your custom code into usrcall() */ | |
| if (r > 31) usrcall(r); else { error(EORANGE); return; } | |
| nexttoken(); | |
| return; | |
| } | |
| } | |
| /* the dartmouth stuff */ | |
| /* | |
| * DATA is simply skipped when encountered as a command | |
| */ | |
| void xdata() { | |
| while (!termsymbol()) nexttoken(); | |
| } | |
| /* | |
| * for READ find the next data record, helper of READ | |
| */ | |
| void nextdatarecord() { | |
| address_t h; | |
| mem_t s=1; | |
| /* save the location of the interpreter and the token we are processing */ | |
| h=here; | |
| /* data at zero means we need to init it, by searching the first data record */ | |
| if (data == 0) { | |
| here=0; | |
| while (here<top && token!=TDATA) gettoken(); | |
| data=here; | |
| datarc=1; | |
| } | |
| processdata: | |
| /* | |
| * data at top means we have exhausted all data, | |
| * nothing more to be done here, however we simulate | |
| * a number value of 0 here and don't throw an error | |
| * this is not Dartmouth style, more consistent with | |
| * iterables | |
| */ | |
| if (data == top) { | |
| token=NUMBER; | |
| x=0; | |
| ert=1; | |
| here=h; | |
| return; | |
| } | |
| /* we process the data record by setting the here pointer to data and search with gettoken */ | |
| here=data; | |
| gettoken(); | |
| if (token == '-') {s=-1; gettoken();} | |
| if (token == NUMBER || token == STRING) goto enddatarecord; | |
| if (token == ',') { | |
| gettoken(); | |
| if (token == '-') {s=-1; gettoken();} | |
| if (token != NUMBER && token != STRING) { | |
| error(EUNKNOWN); | |
| here=h; | |
| return; | |
| } | |
| goto enddatarecord; | |
| } | |
| if (termsymbol()) { | |
| while (here<top && token!=TDATA) gettoken(); | |
| data=here; | |
| goto processdata; | |
| } | |
| if (DEBUG) { outsc("** error in nextdata after termsymbol "); outnumber(data); outcr(); } | |
| error(EUNKNOWN); | |
| enddatarecord: | |
| if (token == NUMBER && s == -1) { x=-x; s=1; } /* this is needed because we tokenize only positive numbers */ | |
| data=here; | |
| datarc++; | |
| here=h; | |
| if (DEBUG) { | |
| outsc("** leaving nextdata with data and here "); | |
| outnumber(data); outspc(); | |
| outnumber(here); outcr(); | |
| } | |
| } | |
| /* | |
| * READ - find data records and insert them to variables | |
| */ | |
| void xread(){ | |
| token_t t0; /* remember the left hand side token until the end of the statement, type of the lhs */ | |
| lhsobject_t lhs; | |
| mem_t datat; /* the type of the data element */ | |
| address_t lendest, lensource, newlength; | |
| int k; | |
| string_t s; | |
| nextdata: | |
| /* look for the variable */ | |
| nexttoken(); | |
| /* this code evaluates the left hand side - remember type and name */ | |
| lhs.name=name; | |
| lefthandside(&lhs); | |
| if (!USELONGJUMP && er) return; | |
| if (DEBUG) { | |
| outsc("** read lefthandside "); | |
| outname(&lhs.name); | |
| outsc(" at here "); | |
| outnumber(here); | |
| outsc(" and data pointer "); | |
| outnumber(data); | |
| outcr(); | |
| } | |
| /* if the token after lhs is not a termsymbol or a comma, something is wrong */ | |
| if (!termsymbol() && token != ',') { error(EUNKNOWN); return; } | |
| /* remember the token we have draw from the stream */ | |
| t0=token; | |
| /* find the data and assign */ | |
| nextdatarecord(); | |
| if (!USELONGJUMP && er) return; | |
| /* assign the value to the lhs - somewhat redundant code to assignment */ | |
| switch (token) { | |
| case NUMBER: | |
| /* a number is stored on the stack */ | |
| assignnumber2(&lhs, x); | |
| break; | |
| case STRING: | |
| if (lhs.name.token != STRINGVAR) { | |
| /* we read a string into a numerical variable */ | |
| if (sr.address) assignnumber2(&lhs, memread2(sr.address)); | |
| else assignnumber2(&lhs, *sr.ir); | |
| } else { | |
| /* we have all we need in sr */ | |
| /* the destination address of the lefthandside, on the fly create included */ | |
| getstring(&s, &lhs.name, lhs.i, lhs.j); | |
| if (!USELONGJUMP && er) return; | |
| /* the length of the lefthandside string */ | |
| lendest=s.length; | |
| if (DEBUG) { | |
| outsc("* read stringcode "); outname(&lhs.name); outcr(); | |
| outsc("** read source string length "); outnumber(sr.length); outcr(); | |
| outsc("** read dest string length "); outnumber(s.length); outcr(); | |
| outsc("** read dest string dimension "); outnumber(s.strdim); outcr(); | |
| } | |
| /* does the source string fit into the destination */ | |
| if ((lhs.i+sr.length-1) > s.strdim) { error(EORANGE); return; } | |
| /* now write the string */ | |
| assignstring(&s, &sr, sr.length); | |
| /* classical Apple 1 behaviour is string truncation in substring logic */ | |
| newlength = lhs.i+sr.length-1; | |
| setstringlength(&lhs.name, newlength, lhs.j); | |
| } | |
| break; | |
| default: | |
| error(EUNKNOWN); | |
| return; | |
| } | |
| /* next list item */ | |
| if (t0 == ',') goto nextdata; | |
| /* no nexttoken here as we have already a termsymbol */ | |
| if (DEBUG) { | |
| outsc("** leaving xread with "); outnumber(token); outcr(); | |
| outsc("** at here "); outnumber(here); outcr(); | |
| outsc("** and data pointer "); outnumber(data); outcr(); | |
| } | |
| /* restore the token for further processing */ | |
| token=t0; | |
| } | |
| /* | |
| * RESTORE sets the data pointer to zero right now | |
| */ | |
| void xrestore(){ | |
| short rec; | |
| nexttoken(); | |
| /* a plain restore */ | |
| if (termsymbol()) { | |
| data=0; | |
| datarc=1; | |
| return; | |
| } | |
| /* something with an argument */ | |
| expression(); | |
| if (!USELONGJUMP && er) return; | |
| /* we search a record */ | |
| rec=pop(); | |
| /* if we need to search backward, back to the beginning */ | |
| if (rec < datarc) { | |
| data=0; | |
| datarc=1; | |
| } | |
| /* advance to the record or top */ | |
| while (datarc < rec && data < top) nextdatarecord(); | |
| /* token is poisoned after nextdatarecord, need to cure this here */ | |
| nexttoken(); | |
| } | |
| /* | |
| * DEF a function, functions are tokenized as FN ARRAYVAR to make | |
| * name processing easy. | |
| */ | |
| void xdef(){ | |
| address_t a; | |
| name_t function; /* the name of the function */ | |
| name_t variable; /* the name of the argument */ | |
| /* do we define a function */ | |
| if (!expect(TFN, EUNKNOWN)) return; | |
| /* the name of the function, it is tokenized as an array */ | |
| if (!expect(ARRAYVAR, EUNKNOWN)) return; | |
| function=name; | |
| function.token=TFN; /* set the right type here */ | |
| /* the argument variable */ | |
| if (!expect('(', EUNKNOWN)) return; | |
| nexttoken(); | |
| if (token == ')') { | |
| zeroname(&variable); | |
| } else if (token == VARIABLE) { | |
| variable=name; | |
| nexttoken(); | |
| } else { | |
| error(EUNKNOWN); | |
| return; | |
| } | |
| if (token != ')') { error(EUNKNOWN); return; } | |
| /* which type of function do we store is found in token */ | |
| nexttoken(); | |
| /* ready to store the function */ | |
| if (DEBUG) { | |
| outsc("** DEF FN with function "); | |
| outname(&function); | |
| outsc(" and argument "); | |
| outname(&variable); | |
| outsc(" at here "); | |
| outnumber(here); | |
| outsc(" and token is "); | |
| outnumber(token); | |
| outcr(); | |
| } | |
| /* find the function, we allow redefinition, currently only functions with 1 argument */ | |
| if ((a=bfind(&function))==0) a=bmalloc(&function, 1); | |
| if (DEBUG) {outsc("** found function structure at "); outnumber(a); outcr(); } | |
| if (!USELONGJUMP && er) return; | |
| /* no more memory */ | |
| if (a == 0) { error(EVARIABLE); return; } | |
| /* store the payload */ | |
| /* first the jump address */ | |
| setaddress(a, memwrite2, here); | |
| a=a+addrsize; | |
| /* the type of the return value - at the moment only numbers */ | |
| if (token == '=') | |
| memwrite2(a++, VARIABLE); | |
| else | |
| memwrite2(a++, 0); | |
| /* store the number of variables */ | |
| memwrite2(a++, 1); | |
| /* store the type and the name of the variables */ | |
| memwrite2(a++, variable.token); | |
| setname_pgm(a, &variable); | |
| a=a+sizeof(name_t)-1; /* useless now but needed later */ | |
| /* skip the function body during defintion */ | |
| if (token == '=') { | |
| while (!termsymbol()) nexttoken(); | |
| } else { | |
| while (token != TFEND) { | |
| nexttoken(); | |
| if (token == TDEF || token == EOL) { error(EFUN); return; } | |
| } | |
| nexttoken(); | |
| error(EFUN); | |
| return; | |
| } | |
| } | |
| /* | |
| * FN function evaluation, this is a call from factor or directly from | |
| * statement, the variable m tells xfn which one it is. 0 is from | |
| * factor and 1 is from statement. | |
| * | |
| * This mechanism is only needed in multiline functions. In this case, | |
| * a new interpreter instance is started with statement(). The variable | |
| * m decides whether the stack should contain a return value (call from factor) | |
| * or should be empty. | |
| * | |
| * The new function code has local variable capability of the new heap. | |
| */ | |
| void xfn(mem_t m) { | |
| address_t a; | |
| address_t h1, h2; | |
| name_t variable; | |
| token_t type; | |
| /* the name of the function and its address */ | |
| if (!expect(ARRAYVAR, EUNKNOWN)) return; | |
| name.token=TFN; | |
| a=bfind(&name); | |
| if (a == 0) {error(EUNKNOWN); return; } | |
| if (DEBUG) { outsc("** in xfn found function "); outname(&name); outsc(" at "); outnumber(a); outcr(); } | |
| /* and the argument */ | |
| if (!expect('(', EUNKNOWN)) return; | |
| nexttoken(); | |
| /* if there is no argument, set it to zero */ | |
| if (token == ')') { | |
| push(0); | |
| } else { | |
| expression(); | |
| if (!USELONGJUMP && er) return; | |
| } | |
| if (token != ')') {error(EUNKNOWN); return; } | |
| /* where is the function code */ | |
| h1=getaddress(a, memread2); | |
| a=a+addrsize; | |
| if (DEBUG) { outsc("** found function address "); outnumber(h1); outcr(); } | |
| /* which type of function do we have*/ | |
| type=memread2(a++); | |
| if (DEBUG) { outsc("** found function type "); outnumber(type); outcr(); } | |
| /* the number of variables is always one here */ | |
| a++; | |
| /* what is the name of the variable, direct read as getname also gets a token */ | |
| /* skip the type here as not needed*/ | |
| variable.token=memread2(a++); | |
| (void) getname(a, &variable, memread2); | |
| a=a+sizeof(name_t)-1; | |
| if (DEBUG) { outsc("** found function variable "); outname(&variable); outcr(); } | |
| /* create a local variable and store the value in it if there is a variable */ | |
| if (variable.c[0]) { | |
| if (!bmalloc(&variable, 0)) { error(EVARIABLE); return; } | |
| setvar(&variable, pop()); | |
| } | |
| /* store here and then evaluate the function */ | |
| h2=here; | |
| here=h1; | |
| /* for simple singleline function, we directly do experession evaluation */ | |
| if (type == VARIABLE) { | |
| if (DEBUG) {outsc("** evaluating expression at "); outnumber(here); outcr(); } | |
| if (!expectexpr()) return; | |
| } else { | |
| /* here comes the tricky part, we start a new interpreter instance */ | |
| if (DEBUG) {outsc("** starting a new interpreter instance "); outcr();} | |
| nexttoken(); | |
| fncontext++; | |
| if (fncontext > FNLIMIT) { error(EFUN); return; } | |
| statement(); | |
| if (!USELONGJUMP && er) return; | |
| if (fncontext > 0) fncontext--; else error(EFUN); | |
| error(EFUN); | |
| return; | |
| } | |
| /* now that all the function stuff is done, return to here and set the variable right */ | |
| here=h2; | |
| (void) bfree(&variable); | |
| /* now, depending on how this was called, make things right, we remove | |
| the return value from the stack and call nexttoken */ | |
| if (m == 1) { | |
| pop(); | |
| nexttoken(); | |
| } | |
| } | |
| /* | |
| * ON is a bit like IF | |
| */ | |
| void xon(){ | |
| number_t cr, tmp; | |
| int ci; | |
| token_t t; | |
| int line = 0; | |
| /* ON can do the ON ERROR and ON EVENT commands as well, in this BASIC | |
| ERROR and EVENT can also be used without the ON */ | |
| nexttoken(); | |
| switch(token) { | |
| case TERROR: | |
| xerror(); | |
| return; | |
| case TEVENT: | |
| xevent(); | |
| return; | |
| default: | |
| expression(); | |
| if (!USELONGJUMP && er) return; | |
| } | |
| /* the result of the condition, can be any number even large */ | |
| cr=pop(); | |
| if (DEBUG) { outsc("** in on condition found "); outnumber(cr); outcr(); } | |
| /* is there a goto or gosub */ | |
| if (token != TGOSUB && token != TGOTO) { error(EUNKNOWN); return; } | |
| /* remember if we do gosub or goto */ | |
| t=token; | |
| /* how many arguments have we got here */ | |
| nexttoken(); | |
| parsearguments(); | |
| if (!USELONGJUMP && er) return; | |
| if (args == 0) { error(EARGS); return; } | |
| /* do we have more arguments then the condition? */ | |
| if (cr > args && cr <= 0) ci=0; else ci=(int)cr; | |
| /* now find the line to jump to and clean the stack, reuse cr | |
| * we need to clean the stack here completely, therefore complete the loop | |
| * ERROR handling is needed for the trapping mechanism. No using popaddress() | |
| * here, because of the needed stack cleanup. Unclear is this precaution is | |
| * really needed. | |
| */ | |
| while (args) { | |
| tmp=pop(); | |
| if (args == ci) { | |
| if (tmp < 0) er=ELINE; | |
| line=tmp; | |
| } | |
| args--; | |
| } | |
| if (!USELONGJUMP && er) return; | |
| if (DEBUG) { outsc("** in on found line as target "); outnumber(line); outcr(); } | |
| /* no line found to jump to */ | |
| if (line == 0) { | |
| nexttoken(); | |
| return; | |
| } | |
| /* prepare for the jump */ | |
| if (t == TGOSUB) pushgosubstack(0); | |
| if (!USELONGJUMP && er) return; | |
| findline(line); | |
| if (!USELONGJUMP && er) return; | |
| /* goto in interactive mode switched to RUN mode | |
| no clearing of variables and stacks */ | |
| if (st == SINT) st=SRUN; | |
| } | |
| /* the structured BASIC extensions, WHILE, UNTIL, and SWITCH */ | |
| void xwhile() { | |
| /* what? */ | |
| if (DEBUG) { outsc("** in while "); outnumber(here); outspc(); outnumber(token); outcr(); } | |
| /* interactively we need to save the buffer location */ | |
| if (st == SINT) here=bi-ibuffer; | |
| /* save the current location and token type, here points to the condition, name is irrelevant */ | |
| pushloop(0, TWHILE, here, 0, 0); | |
| /* is there a valid condition */ | |
| if (!expectexpr()) return; | |
| /* if false, seek WEND and clear the stack*/ | |
| if (!pop()) { | |
| droploop(); | |
| if (st == SINT) bi=ibuffer+here; | |
| nexttoken(); | |
| findbraket(TWHILE, TWEND); | |
| nexttoken(); | |
| } | |
| } | |
| void xwend() { | |
| blocation_t l; | |
| bloop_t* loop; | |
| /* remember where we are */ | |
| pushlocation(&l); | |
| /* back to the condition */ | |
| loop = activeloop(); | |
| if (!USELONGJUMP && er) return; | |
| /* is this a while loop */ | |
| if (loop->var.token != TWHILE ) { error(TWEND); return; } | |
| /* interactive run or program run */ | |
| if (st == SINT) bi=ibuffer+loop->here; else here=loop->here; | |
| /* is there a valid condition */ | |
| if (!expectexpr()) return; | |
| /* if false, seek WEND */ | |
| if (!pop()) { | |
| droploop(); | |
| poplocation(&l); | |
| nexttoken(); | |
| } | |
| } | |
| void xrepeat() { | |
| /* what? */ | |
| if (DEBUG) { outsc("** in repeat "); outnumber(here); outspc(); outnumber(token); outcr(); } | |
| /* interactively we need to save the buffer location */ | |
| if (st == SINT) here=bi-ibuffer; | |
| /* save the current location and token type, here points statement after repeat */ | |
| pushloop(0, TREPEAT, here, 0, 0); | |
| /* we are done here */ | |
| nexttoken(); | |
| } | |
| void xuntil() { | |
| blocation_t l; | |
| bloop_t* loop; | |
| /* is there a valid condition */ | |
| if (!expectexpr()) return; | |
| /* remember the location */ | |
| pushlocation(&l); | |
| /* look on the stack */ | |
| loop = activeloop(); | |
| if (!USELONGJUMP && er) return; | |
| /* if false, go back to the repeat */ | |
| if (!pop()) { | |
| /* the right loop type ? */ | |
| if (loop->var.token != TREPEAT) { | |
| error(TUNTIL); | |
| return; | |
| } | |
| /* correct for interactive */ | |
| if (st == SINT) bi=ibuffer+loop->here; else here=loop->here; | |
| } else { | |
| /* back to where we were */ | |
| droploop(); | |
| poplocation(&l); | |
| } | |
| nexttoken(); /* a bit of evil here, hobling over termsymbols */ | |
| } | |
| void xswitch() { | |
| number_t r; | |
| mem_t match = 0; | |
| mem_t swcount = 0; | |
| blocation_t l; | |
| /* lets look at the condition */ | |
| if (!expectexpr()) return; | |
| r=pop(); | |
| /* remember where we are */ | |
| pushlocation(&l); | |
| /* seek the first case to match the condition */ | |
| while (token != EOL) { | |
| if (token == TSWEND) break; | |
| /* nested SWITCH - skip them all*/ | |
| if (token == TSWITCH) { | |
| if (DEBUG) { outsc("** in SWITCH - nesting found "); outcr(); } | |
| nexttoken(); | |
| findbraket(TSWITCH, TSWEND); | |
| if (DEBUG) { outsc("** in SWITCH SWEND found at "); outnumber(here); outcr(); } | |
| if (!USELONGJUMP && er) return; | |
| } | |
| /* a true case */ | |
| if (token == TCASE) { | |
| /* more sophisticated, case can have an argument list */ | |
| nexttoken(); | |
| parsearguments(); | |
| if (DEBUG) { outsc("** in CASE found "); outnumber(args); outsc(" arguments"); outcr(); } | |
| if (!USELONGJUMP && er) return; | |
| if (args == 0) { | |
| error(TCASE); | |
| return; | |
| } | |
| while (args > 0) { | |
| if (pop() == r) match=1; | |
| args--; | |
| } | |
| if (match) { | |
| return; | |
| } | |
| } | |
| nexttoken(); | |
| } | |
| /* return to the original location and continue if no case is found */ | |
| poplocation(&l); | |
| } | |
| /* a nacked case statement always seeks the end of the switch */ | |
| void xcase() { | |
| while (token != EOL) { | |
| nexttoken(); | |
| if (token == TSWEND) break; | |
| /* broken if switch is nested deeper then once, need the braket mechanism here */ | |
| /* | |
| if (token == TSWITCH) { | |
| nexttoken(); | |
| findbraket(TSWITCH, TSWEND); | |
| } | |
| */ | |
| } | |
| } | |
| /* | |
| * statement processes an entire basic statement until the end | |
| * of the line. | |
| * | |
| * The statement loop is a bit odd and requires some explanation. | |
| * A statement function called in the central switch here must either | |
| * call nexttoken as its last action to feed the loop with a new token | |
| * and then break or it must return which means that the rest of the | |
| * line is ignored. A function that doesn't call nexttoken and just | |
| * breaks causes an infinite loop. | |
| * | |
| * statement is called once in interactive mode and terminates | |
| * at end of a line. | |
| */ | |
| void statement(){ | |
| mem_t xc; | |
| if (DEBUG) bdebug("statement \n"); | |
| /* we can long jump out out any function now, making error handling easier */ | |
| /* if we return here with a long jump, only the error handler is triggered */ | |
| /* this mechanism always branches to the highest context */ | |
| if (fncontext == 0) if (setjmp(sthook)) goto errorhandler; | |
| /* the core loop processing commands */ | |
| while (token != EOL) { | |
| /* debug level 1 happens only in the statement loop */ | |
| if (debuglevel == 1) { debugtoken(); outcr(); } | |
| switch(token){ | |
| case ':': | |
| case LINENUMBER: | |
| nexttoken(); | |
| break; | |
| /* Palo Alto BASIC language set + BREAK */ | |
| case TPRINT: | |
| xprint(); | |
| break; | |
| case TLET: | |
| nexttoken(); | |
| if ((token != ARRAYVAR) && (token != STRINGVAR) && (token != VARIABLE)) { | |
| error(EUNKNOWN); | |
| break; | |
| } | |
| case STRINGVAR: | |
| case ARRAYVAR: | |
| case VARIABLE: | |
| assignment(); | |
| break; | |
| case TINPUT: | |
| xinput(); | |
| break; | |
| case TRETURN: | |
| xreturn(); | |
| if (fncontext > 0) { | |
| nexttoken(); | |
| if (termsymbol()) { push(0); } | |
| else expression(); | |
| return; | |
| } else | |
| xreturn(); | |
| break; | |
| case TGOSUB: | |
| case TGOTO: | |
| xgoto(); | |
| break; | |
| case TGOSUB: | |
| if (fncontext > 0) { error(EFUN); return; } | |
| case TGOTO: | |
| xgoto(); | |
| break; | |
| case TIF: | |
| xif(); | |
| break; | |
| case TFOR: | |
| xfor(); | |
| break; | |
| case TNEXT: | |
| xnext(); | |
| break; | |
| case TBREAK: | |
| xbreak(); | |
| break; | |
| case TSTOP: | |
| case TEND: /* return here because new input is needed, end as a block end is handles elsewhere */ | |
| *ibuffer=0; /* clear ibuffer - this is a hack */ | |
| st=SINT; /* switch to interactive mode */ | |
| eflush(); /* if there is an EEPROM dummy, flush it here (protects flash storage!) */ | |
| ofileclose(); | |
| nexttoken(); | |
| if (token == TSTOP) { | |
| restartsystem(); | |
| } | |
| *ibuffer=0; /* clear ibuffer - this is a hack */ | |
| st=SINT; /* switch to interactive mode */ | |
| return; | |
| case TLIST: | |
| xlist(); | |
| break; | |
| case TNEW: /* return here because new input is needed */ | |
| xnew(); | |
| return; | |
| case TCONT: /* cont behaves differently in interactive and in run mode */ | |
| if (st==SRUN || st==SERUN) { | |
| xcont(); | |
| break; | |
| } /* no break here, because interactively CONT=RUN minus CLR */ | |
| case TRUN: | |
| xrun(); | |
| return; | |
| case TREM: | |
| xrem(); | |
| break; | |
| /* Apple 1 language set */ | |
| case TDIM: | |
| xdim(); | |
| break; | |
| case TCLR: | |
| xclr(); | |
| break; | |
| case TTAB: | |
| case TSPC: | |
| xtab(); | |
| break; | |
| case TPOKE: | |
| xpoke(); | |
| break; | |
| /* Stefan's tinybasic additions */ | |
| case TSAVE: | |
| xsave(); | |
| break; | |
| case TLOAD: | |
| xload(0); | |
| if (st == SINT) return; /* interactive load doesn't like break as the ibuffer is messed up; */ | |
| else break; | |
| case TDUMP: | |
| xdump(); | |
| break; | |
| case TGET: | |
| xget(); | |
| break; | |
| case TPUT: | |
| xput(); | |
| break; | |
| case TSET: | |
| xset(); | |
| break; | |
| case TNETSTAT: | |
| xnetstat(); | |
| break; | |
| case TCLS: | |
| ax=od; | |
| /* if we have a display it is the default for CLS */ | |
| od=ODSP; | |
| outch(12); | |
| od=ax; | |
| nexttoken(); | |
| break; | |
| case TLOCATE: | |
| xlocate(); | |
| break; | |
| /* low level functions as part of Stefan's extension */ | |
| case TCALL: | |
| xcall(); | |
| break; | |
| /* Arduino IO */ | |
| case TDWRITE: | |
| xdwrite(); | |
| break; | |
| case TAWRITE: | |
| xawrite(); | |
| break; | |
| case TPINM: | |
| xpinm(); | |
| break; | |
| case TDELAY: | |
| xdelay(); | |
| break; | |
| case TTONE: | |
| xtone(); | |
| break; | |
| case TPULSE: | |
| xpulse(); | |
| break; | |
| /* BASIC DOS function */ | |
| case TCATALOG: | |
| xcatalog(); | |
| break; | |
| case TDELETE: | |
| xdelete(); | |
| break; | |
| case TOPEN: | |
| xopen(); | |
| break; | |
| case TCLOSE: | |
| xclose(); | |
| break; | |
| case TFDISK: | |
| xfdisk(); | |
| break; | |
| /* graphics */ | |
| case TCOLOR: | |
| xcolor(); | |
| break; | |
| case TPLOT: | |
| xplot(); | |
| break; | |
| case TLINE: | |
| xline(); | |
| break; | |
| case TRECT: | |
| xrect(); | |
| break; | |
| case TCIRCLE: | |
| xcircle(); | |
| break; | |
| case TFRECT: | |
| xfrect(); | |
| break; | |
| case TFCIRCLE: | |
| xfcircle(); | |
| break; | |
| case TDATA: | |
| xdata(); | |
| break; | |
| case TREAD: | |
| xread(); | |
| break; | |
| case TRESTORE: | |
| xrestore(); | |
| break; | |
| case TDEF: | |
| xdef(); | |
| break; | |
| case TON: | |
| xon(); | |
| break; | |
| case TFN: | |
| xfn(1); | |
| break; | |
| case TFEND: | |
| /* we leave the statement loop and return to the calling expression() */ | |
| /* if the function is ended with FEND we return 0 */ | |
| if (fncontext == 0) { error(EFUN); return; } | |
| else { push(0); return; } | |
| break; | |
| case TELSE: | |
| xelse(); | |
| break; | |
| case TEVAL: | |
| xeval(); | |
| break; | |
| case TERROR: | |
| xerror(); | |
| break; | |
| case TSLEEP: | |
| xsleep(); | |
| break; | |
| case TWIRE: | |
| xwire(); | |
| break; | |
| case TAFTER: | |
| case TEVERY: | |
| xtimer(); | |
| break; | |
| case TEVENT: | |
| xevent(); | |
| break; | |
| case TWHILE: | |
| xwhile(); | |
| break; | |
| case TWEND: | |
| xwend(); | |
| break; | |
| case TREPEAT: | |
| xrepeat(); | |
| break; | |
| case TUNTIL: | |
| xuntil(); | |
| break; | |
| case TSWITCH: | |
| xswitch(); | |
| break; | |
| case TCASE: | |
| xcase(); | |
| break; | |
| case TSWEND: | |
| case TDO: | |
| case TDEND: | |
| nexttoken(); | |
| break; | |
| case TEDIT: | |
| xedit(); | |
| break; | |
| default: | |
| /* strict syntax checking */ | |
| error(EUNKNOWN); | |
| goto errorhandler; | |
| } | |
| /* | |
| * after each statement we check on a break character | |
| * on an Arduino entering "#" at runtime stops the program | |
| * for POSIXNONBLOCKING we do this in the background loop | |
| * to avoid slowing down | |
| */ | |
| if (checkch() == BREAKCHAR) { | |
| st=SINT; | |
| if (od == 1) serialflush(); else xc=inch(); | |
| return; | |
| } | |
| if (breakcondition) { | |
| breakcondition=0; | |
| st=SINT; | |
| if (od == 1) serialflush(); else xc=inch(); | |
| return; | |
| } | |
| /* and after each statement, check the break pin */ | |
| if (getbreakpin() == 0) { | |
| st=SINT; | |
| return; | |
| }; | |
| /* and then there is also signal handling on some platforms */ | |
| if (breaksignal) { | |
| st=SINT; | |
| breaksignal=0; | |
| serialflush(); | |
| outcr(); | |
| return; | |
| } | |
| /* yield after each statement which is a 10-100 microsecond cycle | |
| on Arduinos and the like, all background tasks are handled in byield */ | |
| byield(); | |
| /* if error handling is compiled into the code, errors can be trapped here */ | |
| errorhandler: | |
| if (er) { | |
| if (st != SINT) { | |
| erh=er; | |
| er=0; | |
| switch(berrorh.type) { | |
| case TCONT: | |
| while(!termsymbol()) nexttoken(); | |
| break; | |
| case TGOTO: | |
| findline(berrorh.linenumber); | |
| berrorh.type=0; | |
| berrorh.linenumber=0; | |
| if (er) return; | |
| break; | |
| case 0: | |
| return; | |
| default: | |
| nexttoken(); | |
| } | |
| } else | |
| return; | |
| } | |
| /* when an error is encountered the statement loop is ended */ | |
| if (er) return; | |
| /* | |
| * if we run error free, interrupts and times can be processed | |
| * | |
| * We can savely interrupt and return only if here points either to | |
| * a termsymbol : or LINENUMBER. NEXT is a special case. We need to | |
| * catch this here because empty FOR loops never even have a termsymbol | |
| * a : is swallowed after FOR. | |
| * | |
| * the interrupts are only triggered in fncontext 0, i.e. in the | |
| * main loop. While in functions, all interrupts are disabled. | |
| * | |
| */ | |
| if ((token == LINENUMBER || token == ':' || token == TNEXT) && (st == SERUN || st == SRUN)) { | |
| /* after is always processed before every */ | |
| if (after_timer.enabled && fncontext == 0) { | |
| if (millis() > after_timer.last + after_timer.interval) { | |
| after_timer.enabled=0; | |
| if (after_timer.type == TGOSUB) { | |
| if (token == TNEXT || token == ':') here--; | |
| if (token == LINENUMBER) here-=(1+sizeof(address_t)); | |
| pushgosubstack(0); | |
| } | |
| findline(after_timer.linenumber); | |
| if (er) return; | |
| } | |
| } | |
| /* periodic events */ | |
| if (every_timer.enabled && fncontext == 0) { | |
| if (millis() > every_timer.last + every_timer.interval) { | |
| every_timer.last=millis(); | |
| if (every_timer.type == TGOSUB) { | |
| if (token == TNEXT || token == ':') here--; | |
| if (token == LINENUMBER) here-=(1+sizeof(address_t)); | |
| pushgosubstack(0); | |
| if (er) return; | |
| } | |
| findline(every_timer.linenumber); | |
| if (er) return; | |
| } | |
| } | |
| } | |
| /* the branch code for interrupts, we round robin through the event list */ | |
| if ((token == LINENUMBER || token == ':' || token == TNEXT) && (st == SERUN || st == SRUN)) { | |
| /* interrupts */ | |
| if (events_enabled && fncontext == 0) { | |
| for (ax=0; ax<EVENTLISTSIZE; ax++) { | |
| if (eventlist[ievent].pin && eventlist[ievent].enabled && eventlist[ievent].active) { | |
| if (eventlist[ievent].type == TGOSUB) { | |
| if (token == TNEXT || token == ':') here--; | |
| if (token == LINENUMBER) here-=(1+sizeof(address_t)); | |
| pushgosubstack(TEVENT); | |
| if (er) return; | |
| } | |
| findline(eventlist[ievent].linenumber); | |
| if (er) return; | |
| eventlist[ievent].active=0; | |
| enableevent(eventlist[ievent].pin); /* events are disabled in the interrupt function, here they are activated again */ | |
| events_enabled=0; /* once we have jumped, we keep the events in BASIC off until reenabled by the program*/ | |
| break; | |
| } | |
| ievent=(ievent+1)%EVENTLISTSIZE; | |
| } | |
| } | |
| } | |
| } | |
| } | |
| /* | |
| * the setup routine - Arduino style | |
| */ | |
| void setup() { | |
| /* start measureing time */ | |
| timeinit(); | |
| /* initialize the event system */ | |
| initevents(); | |
| /* init all io functions */ | |
| ioinit(); | |
| /* setup for all non BASIC stuff */ | |
| bsetup(); | |
| /* get the BASIC memory, either as memory array with | |
| ballocmem() or as an SPI serical memory */ | |
| himem=memsize=spirambegin(); | |
| /* | |
| * for an EEPROMMEM system, the memory consists of the | |
| * EEPROM from 0 to elength()-eheadersize and then the RAM. | |
| */ | |
| himem=memsize=ballocmem()+(elength()-eheadersize); | |
| himem=memsize=ballocmem(); | |
| if (DEBUG) { outsc("** on startup, memsize is "); outnumber(memsize); outcr(); } | |
| /* be ready for a new program if we run on RAM*/ | |
| xnew(); | |
| if (DEBUG) { outsc("** on startup, ran xnew "); outcr(); } | |
| /* if we run on an EEPROM system, more work is needed */ | |
| if (eread(0) == 0 || eread(0) == 1) { /* have we stored a program and don't do new */ | |
| top=getaddress(1, eread); | |
| resetbasicstate(); /* the little brother of new, reset the state but let the program memory be */ | |
| for (address_t a=elength(); a<memsize; a++) memwrite2(a, 0); /* clear the heap i.e. the basic RAM*/ | |
| } else { | |
| eupdate(0, 0); /* now we have stored a program of length 0 */ | |
| setaddress(1, eupdate, 0); | |
| xnew(); | |
| } | |
| /* check if there is something to autorun and prepare | |
| the interpreter to got into autorun once loop is reached */ | |
| if (!autorun()) { | |
| printmessage(MGREET); outspc(); | |
| printmessage(EOUTOFMEMORY); outspc(); | |
| if (memsize < maxnum) outnumber(memsize+1); else { outnumber(memsize/1024+1); outch('k'); } | |
| outspc(); | |
| outnumber(elength()); outcr(); | |
| } | |
| /* activate the BREAKPIN */ | |
| breakpinbegin(); | |
| } | |
| /* | |
| * the loop routine for interactive input | |
| */ | |
| void loop() { | |
| /* | |
| * autorun state was found in setup, autorun now but only once | |
| * autorun BASIC programs return to interactive after completion | |
| * autorun code always must loop in itself | |
| */ | |
| if (st == SERUN) { | |
| xrun(); | |
| /* on an EEPROM system we don't set top to 0 here */ | |
| top=0; | |
| st=SINT; | |
| } else if (st == SRUN) { | |
| here=0; | |
| xrun(); | |
| st=SINT; | |
| } | |
| /* always return to default io channels once interactive mode is reached */ | |
| iodefaults(); | |
| form=0; | |
| /* the prompt and the input request */ | |
| printmessage(MPROMPT); | |
| (void) ins(ibuffer, BUFSIZE-2); | |
| /* tokenize first token from the input buffer */ | |
| bi=ibuffer; | |
| nexttoken(); | |
| /* a number triggers the line storage, anything else is executed */ | |
| if (token == NUMBER) { | |
| ax=x; | |
| storeline(); | |
| /* on an EEPROM system we store top after each succesful line insert */ | |
| setaddress(1, eupdate, top); | |
| } else { | |
| /* st=SINT; */ | |
| statement(); | |
| st=SINT; | |
| } | |
| /* here, at last, all errors need to be catched and back to interactive input*/ | |
| if (er) reseterror(); | |
| } | |
| /* if we are not on an Arduino, we need a main */ | |
| int main(int argc, char* argv[]){ | |
| /* save the arguments if there are any */ | |
| bargc=argc; | |
| bargv=argv; | |
| /* do what an Arduino would do, this loops for every interactive input */ | |
| setup(); | |
| while (1) | |
| loop(); | |
| } | |
| /* | |
| * Arduino style function for non BASIC code to run on the MCU. | |
| * All code that needs to run on the MCU independently can be put here. | |
| * It works more or less just like the normal loop() and setup(). | |
| * | |
| * This is meant for robotics and other device control type of stuff | |
| * on the Arduino platform. | |
| * | |
| * On POSIX systems I/O is blocking and therefore bloop() is not called | |
| * consistently. | |
| * | |
| * Rules of the game: | |
| * - bsetup() is called once during interpreter startup after all | |
| * IO subsystems are started and before the BASIC main memory is | |
| * allocated. Allocate memory here. Do not allocate a lot of memory | |
| * in bloop(). | |
| * - Never start or restart I/O functions of BASIC in bsetup(), | |
| * no Wire.begin(), Serial.begin() etc. If BASIC also uses this | |
| * BASIC handles the I/O startup. Things that BASIC does not use | |
| * can be started here. | |
| * - bloop() is called after every token, during I/O polling and in | |
| * DELAY functions. | |
| * - The typical call frequency of bloop() is 20 microseconds or faster. | |
| * This is fairly constant and reliable. | |
| * - The interpreter is robust against code in bloop() that needs a lot | |
| * of CPU time. It will simply slow down but it will not break, | |
| * unless(!) other I/O systems like network also need background CPU time | |
| * and you block bloop for a long time. As a rule of thumb, on network systems | |
| * bloop() should return after 1 ms. After bloop() has returned, the interpreter | |
| * tries to handle network and USB update stuff. | |
| * - Never use delay() in bloop(). Set a counter. Look at the tone | |
| * emulation code for examples. | |
| * - Never ever call BASIC functions from bloop(). BASIC function will | |
| * eventually call byield() which calls bloop() and so forth. | |
| * If you need to communicate data into BASIC, use the BASIC main | |
| * memory, variables or the USR function mechanism. | |
| * - Avoid allocating a lot of memory in bloop(). | |
| */ | |
| void bsetup() { | |
| /* put your setup code here, to run once: */ | |
| } | |
| void bloop() { | |
| /* put your main code here, to run repeatedly: */ | |
| } | |