Download Fore_End/cpcn2/node_modules/htmlparser2/lib/Tokenizer.js from zxy0307/JavaWeb: direct link, hf CLI and curl.
- Browser
- Download file 35.6 kB
-
https://huggingface.co/zxy0307/JavaWeb/resolve/main/Fore_End/cpcn2/node_modules/htmlparser2/lib/Tokenizer.js
- Command line
-
hf download hf://zxy0307/JavaWeb/Fore_End/cpcn2/node_modules/htmlparser2/lib/Tokenizer.js
-
curl -L -o Tokenizer.js https://huggingface.co/zxy0307/JavaWeb/resolve/main/Fore_End/cpcn2/node_modules/htmlparser2/lib/Tokenizer.js
35.6 kB
| ; | |
| var __importDefault = (this && this.__importDefault) || function (mod) { | |
| return (mod && mod.__esModule) ? mod : { "default": mod }; | |
| }; | |
| Object.defineProperty(exports, "__esModule", { value: true }); | |
| var decode_codepoint_1 = __importDefault(require("entities/lib/decode_codepoint")); | |
| var entities_json_1 = __importDefault(require("entities/lib/maps/entities.json")); | |
| var legacy_json_1 = __importDefault(require("entities/lib/maps/legacy.json")); | |
| var xml_json_1 = __importDefault(require("entities/lib/maps/xml.json")); | |
| function whitespace(c) { | |
| return c === " " || c === "\n" || c === "\t" || c === "\f" || c === "\r"; | |
| } | |
| function isASCIIAlpha(c) { | |
| return (c >= "a" && c <= "z") || (c >= "A" && c <= "Z"); | |
| } | |
| function ifElseState(upper, SUCCESS, FAILURE) { | |
| var lower = upper.toLowerCase(); | |
| if (upper === lower) { | |
| return function (t, c) { | |
| if (c === lower) { | |
| t._state = SUCCESS; | |
| } | |
| else { | |
| t._state = FAILURE; | |
| t._index--; | |
| } | |
| }; | |
| } | |
| return function (t, c) { | |
| if (c === lower || c === upper) { | |
| t._state = SUCCESS; | |
| } | |
| else { | |
| t._state = FAILURE; | |
| t._index--; | |
| } | |
| }; | |
| } | |
| function consumeSpecialNameChar(upper, NEXT_STATE) { | |
| var lower = upper.toLowerCase(); | |
| return function (t, c) { | |
| if (c === lower || c === upper) { | |
| t._state = NEXT_STATE; | |
| } | |
| else { | |
| t._state = 3 /* InTagName */; | |
| t._index--; // Consume the token again | |
| } | |
| }; | |
| } | |
| var stateBeforeCdata1 = ifElseState("C", 24 /* BeforeCdata2 */, 16 /* InDeclaration */); | |
| var stateBeforeCdata2 = ifElseState("D", 25 /* BeforeCdata3 */, 16 /* InDeclaration */); | |
| var stateBeforeCdata3 = ifElseState("A", 26 /* BeforeCdata4 */, 16 /* InDeclaration */); | |
| var stateBeforeCdata4 = ifElseState("T", 27 /* BeforeCdata5 */, 16 /* InDeclaration */); | |
| var stateBeforeCdata5 = ifElseState("A", 28 /* BeforeCdata6 */, 16 /* InDeclaration */); | |
| var stateBeforeScript1 = consumeSpecialNameChar("R", 35 /* BeforeScript2 */); | |
| var stateBeforeScript2 = consumeSpecialNameChar("I", 36 /* BeforeScript3 */); | |
| var stateBeforeScript3 = consumeSpecialNameChar("P", 37 /* BeforeScript4 */); | |
| var stateBeforeScript4 = consumeSpecialNameChar("T", 38 /* BeforeScript5 */); | |
| var stateAfterScript1 = ifElseState("R", 40 /* AfterScript2 */, 1 /* Text */); | |
| var stateAfterScript2 = ifElseState("I", 41 /* AfterScript3 */, 1 /* Text */); | |
| var stateAfterScript3 = ifElseState("P", 42 /* AfterScript4 */, 1 /* Text */); | |
| var stateAfterScript4 = ifElseState("T", 43 /* AfterScript5 */, 1 /* Text */); | |
| var stateBeforeStyle1 = consumeSpecialNameChar("Y", 45 /* BeforeStyle2 */); | |
| var stateBeforeStyle2 = consumeSpecialNameChar("L", 46 /* BeforeStyle3 */); | |
| var stateBeforeStyle3 = consumeSpecialNameChar("E", 47 /* BeforeStyle4 */); | |
| var stateAfterStyle1 = ifElseState("Y", 49 /* AfterStyle2 */, 1 /* Text */); | |
| var stateAfterStyle2 = ifElseState("L", 50 /* AfterStyle3 */, 1 /* Text */); | |
| var stateAfterStyle3 = ifElseState("E", 51 /* AfterStyle4 */, 1 /* Text */); | |
| var stateBeforeSpecialT = consumeSpecialNameChar("I", 54 /* BeforeTitle1 */); | |
| var stateBeforeTitle1 = consumeSpecialNameChar("T", 55 /* BeforeTitle2 */); | |
| var stateBeforeTitle2 = consumeSpecialNameChar("L", 56 /* BeforeTitle3 */); | |
| var stateBeforeTitle3 = consumeSpecialNameChar("E", 57 /* BeforeTitle4 */); | |
| var stateAfterSpecialTEnd = ifElseState("I", 58 /* AfterTitle1 */, 1 /* Text */); | |
| var stateAfterTitle1 = ifElseState("T", 59 /* AfterTitle2 */, 1 /* Text */); | |
| var stateAfterTitle2 = ifElseState("L", 60 /* AfterTitle3 */, 1 /* Text */); | |
| var stateAfterTitle3 = ifElseState("E", 61 /* AfterTitle4 */, 1 /* Text */); | |
| var stateBeforeEntity = ifElseState("#", 63 /* BeforeNumericEntity */, 64 /* InNamedEntity */); | |
| var stateBeforeNumericEntity = ifElseState("X", 66 /* InHexEntity */, 65 /* InNumericEntity */); | |
| var Tokenizer = /** @class */ (function () { | |
| function Tokenizer(options, cbs) { | |
| var _a; | |
| /** The current state the tokenizer is in. */ | |
| this._state = 1 /* Text */; | |
| /** The read buffer. */ | |
| this.buffer = ""; | |
| /** The beginning of the section that is currently being read. */ | |
| this.sectionStart = 0; | |
| /** The index within the buffer that we are currently looking at. */ | |
| this._index = 0; | |
| /** | |
| * Data that has already been processed will be removed from the buffer occasionally. | |
| * `_bufferOffset` keeps track of how many characters have been removed, to make sure position information is accurate. | |
| */ | |
| this.bufferOffset = 0; | |
| /** Some behavior, eg. when decoding entities, is done while we are in another state. This keeps track of the other state type. */ | |
| this.baseState = 1 /* Text */; | |
| /** For special parsing behavior inside of script and style tags. */ | |
| this.special = 1 /* None */; | |
| /** Indicates whether the tokenizer has been paused. */ | |
| this.running = true; | |
| /** Indicates whether the tokenizer has finished running / `.end` has been called. */ | |
| this.ended = false; | |
| this.cbs = cbs; | |
| this.xmlMode = !!(options === null || options === void 0 ? void 0 : options.xmlMode); | |
| this.decodeEntities = (_a = options === null || options === void 0 ? void 0 : options.decodeEntities) !== null && _a !== void 0 ? _a : true; | |
| } | |
| Tokenizer.prototype.reset = function () { | |
| this._state = 1 /* Text */; | |
| this.buffer = ""; | |
| this.sectionStart = 0; | |
| this._index = 0; | |
| this.bufferOffset = 0; | |
| this.baseState = 1 /* Text */; | |
| this.special = 1 /* None */; | |
| this.running = true; | |
| this.ended = false; | |
| }; | |
| Tokenizer.prototype.write = function (chunk) { | |
| if (this.ended) | |
| this.cbs.onerror(Error(".write() after done!")); | |
| this.buffer += chunk; | |
| this.parse(); | |
| }; | |
| Tokenizer.prototype.end = function (chunk) { | |
| if (this.ended) | |
| this.cbs.onerror(Error(".end() after done!")); | |
| if (chunk) | |
| this.write(chunk); | |
| this.ended = true; | |
| if (this.running) | |
| this.finish(); | |
| }; | |
| Tokenizer.prototype.pause = function () { | |
| this.running = false; | |
| }; | |
| Tokenizer.prototype.resume = function () { | |
| this.running = true; | |
| if (this._index < this.buffer.length) { | |
| this.parse(); | |
| } | |
| if (this.ended) { | |
| this.finish(); | |
| } | |
| }; | |
| /** | |
| * The current index within all of the written data. | |
| */ | |
| Tokenizer.prototype.getAbsoluteIndex = function () { | |
| return this.bufferOffset + this._index; | |
| }; | |
| Tokenizer.prototype.stateText = function (c) { | |
| if (c === "<") { | |
| if (this._index > this.sectionStart) { | |
| this.cbs.ontext(this.getSection()); | |
| } | |
| this._state = 2 /* BeforeTagName */; | |
| this.sectionStart = this._index; | |
| } | |
| else if (this.decodeEntities && | |
| c === "&" && | |
| (this.special === 1 /* None */ || this.special === 4 /* Title */)) { | |
| if (this._index > this.sectionStart) { | |
| this.cbs.ontext(this.getSection()); | |
| } | |
| this.baseState = 1 /* Text */; | |
| this._state = 62 /* BeforeEntity */; | |
| this.sectionStart = this._index; | |
| } | |
| }; | |
| /** | |
| * HTML only allows ASCII alpha characters (a-z and A-Z) at the beginning of a tag name. | |
| * | |
| * XML allows a lot more characters here (@see https://www.w3.org/TR/REC-xml/#NT-NameStartChar). | |
| * We allow anything that wouldn't end the tag. | |
| */ | |
| Tokenizer.prototype.isTagStartChar = function (c) { | |
| return (isASCIIAlpha(c) || | |
| (this.xmlMode && !whitespace(c) && c !== "/" && c !== ">")); | |
| }; | |
| Tokenizer.prototype.stateBeforeTagName = function (c) { | |
| if (c === "/") { | |
| this._state = 5 /* BeforeClosingTagName */; | |
| } | |
| else if (c === "<") { | |
| this.cbs.ontext(this.getSection()); | |
| this.sectionStart = this._index; | |
| } | |
| else if (c === ">" || | |
| this.special !== 1 /* None */ || | |
| whitespace(c)) { | |
| this._state = 1 /* Text */; | |
| } | |
| else if (c === "!") { | |
| this._state = 15 /* BeforeDeclaration */; | |
| this.sectionStart = this._index + 1; | |
| } | |
| else if (c === "?") { | |
| this._state = 17 /* InProcessingInstruction */; | |
| this.sectionStart = this._index + 1; | |
| } | |
| else if (!this.isTagStartChar(c)) { | |
| this._state = 1 /* Text */; | |
| } | |
| else { | |
| this._state = | |
| !this.xmlMode && (c === "s" || c === "S") | |
| ? 32 /* BeforeSpecialS */ | |
| : !this.xmlMode && (c === "t" || c === "T") | |
| ? 52 /* BeforeSpecialT */ | |
| : 3 /* InTagName */; | |
| this.sectionStart = this._index; | |
| } | |
| }; | |
| Tokenizer.prototype.stateInTagName = function (c) { | |
| if (c === "/" || c === ">" || whitespace(c)) { | |
| this.emitToken("onopentagname"); | |
| this._state = 8 /* BeforeAttributeName */; | |
| this._index--; | |
| } | |
| }; | |
| Tokenizer.prototype.stateBeforeClosingTagName = function (c) { | |
| if (whitespace(c)) { | |
| // Ignore | |
| } | |
| else if (c === ">") { | |
| this._state = 1 /* Text */; | |
| } | |
| else if (this.special !== 1 /* None */) { | |
| if (this.special !== 4 /* Title */ && (c === "s" || c === "S")) { | |
| this._state = 33 /* BeforeSpecialSEnd */; | |
| } | |
| else if (this.special === 4 /* Title */ && | |
| (c === "t" || c === "T")) { | |
| this._state = 53 /* BeforeSpecialTEnd */; | |
| } | |
| else { | |
| this._state = 1 /* Text */; | |
| this._index--; | |
| } | |
| } | |
| else if (!this.isTagStartChar(c)) { | |
| this._state = 20 /* InSpecialComment */; | |
| this.sectionStart = this._index; | |
| } | |
| else { | |
| this._state = 6 /* InClosingTagName */; | |
| this.sectionStart = this._index; | |
| } | |
| }; | |
| Tokenizer.prototype.stateInClosingTagName = function (c) { | |
| if (c === ">" || whitespace(c)) { | |
| this.emitToken("onclosetag"); | |
| this._state = 7 /* AfterClosingTagName */; | |
| this._index--; | |
| } | |
| }; | |
| Tokenizer.prototype.stateAfterClosingTagName = function (c) { | |
| // Skip everything until ">" | |
| if (c === ">") { | |
| this._state = 1 /* Text */; | |
| this.sectionStart = this._index + 1; | |
| } | |
| }; | |
| Tokenizer.prototype.stateBeforeAttributeName = function (c) { | |
| if (c === ">") { | |
| this.cbs.onopentagend(); | |
| this._state = 1 /* Text */; | |
| this.sectionStart = this._index + 1; | |
| } | |
| else if (c === "/") { | |
| this._state = 4 /* InSelfClosingTag */; | |
| } | |
| else if (!whitespace(c)) { | |
| this._state = 9 /* InAttributeName */; | |
| this.sectionStart = this._index; | |
| } | |
| }; | |
| Tokenizer.prototype.stateInSelfClosingTag = function (c) { | |
| if (c === ">") { | |
| this.cbs.onselfclosingtag(); | |
| this._state = 1 /* Text */; | |
| this.sectionStart = this._index + 1; | |
| this.special = 1 /* None */; // Reset special state, in case of self-closing special tags | |
| } | |
| else if (!whitespace(c)) { | |
| this._state = 8 /* BeforeAttributeName */; | |
| this._index--; | |
| } | |
| }; | |
| Tokenizer.prototype.stateInAttributeName = function (c) { | |
| if (c === "=" || c === "/" || c === ">" || whitespace(c)) { | |
| this.cbs.onattribname(this.getSection()); | |
| this.sectionStart = -1; | |
| this._state = 10 /* AfterAttributeName */; | |
| this._index--; | |
| } | |
| }; | |
| Tokenizer.prototype.stateAfterAttributeName = function (c) { | |
| if (c === "=") { | |
| this._state = 11 /* BeforeAttributeValue */; | |
| } | |
| else if (c === "/" || c === ">") { | |
| this.cbs.onattribend(undefined); | |
| this._state = 8 /* BeforeAttributeName */; | |
| this._index--; | |
| } | |
| else if (!whitespace(c)) { | |
| this.cbs.onattribend(undefined); | |
| this._state = 9 /* InAttributeName */; | |
| this.sectionStart = this._index; | |
| } | |
| }; | |
| Tokenizer.prototype.stateBeforeAttributeValue = function (c) { | |
| if (c === '"') { | |
| this._state = 12 /* InAttributeValueDq */; | |
| this.sectionStart = this._index + 1; | |
| } | |
| else if (c === "'") { | |
| this._state = 13 /* InAttributeValueSq */; | |
| this.sectionStart = this._index + 1; | |
| } | |
| else if (!whitespace(c)) { | |
| this._state = 14 /* InAttributeValueNq */; | |
| this.sectionStart = this._index; | |
| this._index--; // Reconsume token | |
| } | |
| }; | |
| Tokenizer.prototype.handleInAttributeValue = function (c, quote) { | |
| if (c === quote) { | |
| this.emitToken("onattribdata"); | |
| this.cbs.onattribend(quote); | |
| this._state = 8 /* BeforeAttributeName */; | |
| } | |
| else if (this.decodeEntities && c === "&") { | |
| this.emitToken("onattribdata"); | |
| this.baseState = this._state; | |
| this._state = 62 /* BeforeEntity */; | |
| this.sectionStart = this._index; | |
| } | |
| }; | |
| Tokenizer.prototype.stateInAttributeValueDoubleQuotes = function (c) { | |
| this.handleInAttributeValue(c, '"'); | |
| }; | |
| Tokenizer.prototype.stateInAttributeValueSingleQuotes = function (c) { | |
| this.handleInAttributeValue(c, "'"); | |
| }; | |
| Tokenizer.prototype.stateInAttributeValueNoQuotes = function (c) { | |
| if (whitespace(c) || c === ">") { | |
| this.emitToken("onattribdata"); | |
| this.cbs.onattribend(null); | |
| this._state = 8 /* BeforeAttributeName */; | |
| this._index--; | |
| } | |
| else if (this.decodeEntities && c === "&") { | |
| this.emitToken("onattribdata"); | |
| this.baseState = this._state; | |
| this._state = 62 /* BeforeEntity */; | |
| this.sectionStart = this._index; | |
| } | |
| }; | |
| Tokenizer.prototype.stateBeforeDeclaration = function (c) { | |
| this._state = | |
| c === "[" | |
| ? 23 /* BeforeCdata1 */ | |
| : c === "-" | |
| ? 18 /* BeforeComment */ | |
| : 16 /* InDeclaration */; | |
| }; | |
| Tokenizer.prototype.stateInDeclaration = function (c) { | |
| if (c === ">") { | |
| this.cbs.ondeclaration(this.getSection()); | |
| this._state = 1 /* Text */; | |
| this.sectionStart = this._index + 1; | |
| } | |
| }; | |
| Tokenizer.prototype.stateInProcessingInstruction = function (c) { | |
| if (c === ">") { | |
| this.cbs.onprocessinginstruction(this.getSection()); | |
| this._state = 1 /* Text */; | |
| this.sectionStart = this._index + 1; | |
| } | |
| }; | |
| Tokenizer.prototype.stateBeforeComment = function (c) { | |
| if (c === "-") { | |
| this._state = 19 /* InComment */; | |
| this.sectionStart = this._index + 1; | |
| } | |
| else { | |
| this._state = 16 /* InDeclaration */; | |
| } | |
| }; | |
| Tokenizer.prototype.stateInComment = function (c) { | |
| if (c === "-") | |
| this._state = 21 /* AfterComment1 */; | |
| }; | |
| Tokenizer.prototype.stateInSpecialComment = function (c) { | |
| if (c === ">") { | |
| this.cbs.oncomment(this.buffer.substring(this.sectionStart, this._index)); | |
| this._state = 1 /* Text */; | |
| this.sectionStart = this._index + 1; | |
| } | |
| }; | |
| Tokenizer.prototype.stateAfterComment1 = function (c) { | |
| if (c === "-") { | |
| this._state = 22 /* AfterComment2 */; | |
| } | |
| else { | |
| this._state = 19 /* InComment */; | |
| } | |
| }; | |
| Tokenizer.prototype.stateAfterComment2 = function (c) { | |
| if (c === ">") { | |
| // Remove 2 trailing chars | |
| this.cbs.oncomment(this.buffer.substring(this.sectionStart, this._index - 2)); | |
| this._state = 1 /* Text */; | |
| this.sectionStart = this._index + 1; | |
| } | |
| else if (c !== "-") { | |
| this._state = 19 /* InComment */; | |
| } | |
| // Else: stay in AFTER_COMMENT_2 (`--->`) | |
| }; | |
| Tokenizer.prototype.stateBeforeCdata6 = function (c) { | |
| if (c === "[") { | |
| this._state = 29 /* InCdata */; | |
| this.sectionStart = this._index + 1; | |
| } | |
| else { | |
| this._state = 16 /* InDeclaration */; | |
| this._index--; | |
| } | |
| }; | |
| Tokenizer.prototype.stateInCdata = function (c) { | |
| if (c === "]") | |
| this._state = 30 /* AfterCdata1 */; | |
| }; | |
| Tokenizer.prototype.stateAfterCdata1 = function (c) { | |
| if (c === "]") | |
| this._state = 31 /* AfterCdata2 */; | |
| else | |
| this._state = 29 /* InCdata */; | |
| }; | |
| Tokenizer.prototype.stateAfterCdata2 = function (c) { | |
| if (c === ">") { | |
| // Remove 2 trailing chars | |
| this.cbs.oncdata(this.buffer.substring(this.sectionStart, this._index - 2)); | |
| this._state = 1 /* Text */; | |
| this.sectionStart = this._index + 1; | |
| } | |
| else if (c !== "]") { | |
| this._state = 29 /* InCdata */; | |
| } | |
| // Else: stay in AFTER_CDATA_2 (`]]]>`) | |
| }; | |
| Tokenizer.prototype.stateBeforeSpecialS = function (c) { | |
| if (c === "c" || c === "C") { | |
| this._state = 34 /* BeforeScript1 */; | |
| } | |
| else if (c === "t" || c === "T") { | |
| this._state = 44 /* BeforeStyle1 */; | |
| } | |
| else { | |
| this._state = 3 /* InTagName */; | |
| this._index--; // Consume the token again | |
| } | |
| }; | |
| Tokenizer.prototype.stateBeforeSpecialSEnd = function (c) { | |
| if (this.special === 2 /* Script */ && (c === "c" || c === "C")) { | |
| this._state = 39 /* AfterScript1 */; | |
| } | |
| else if (this.special === 3 /* Style */ && (c === "t" || c === "T")) { | |
| this._state = 48 /* AfterStyle1 */; | |
| } | |
| else | |
| this._state = 1 /* Text */; | |
| }; | |
| Tokenizer.prototype.stateBeforeSpecialLast = function (c, special) { | |
| if (c === "/" || c === ">" || whitespace(c)) { | |
| this.special = special; | |
| } | |
| this._state = 3 /* InTagName */; | |
| this._index--; // Consume the token again | |
| }; | |
| Tokenizer.prototype.stateAfterSpecialLast = function (c, sectionStartOffset) { | |
| if (c === ">" || whitespace(c)) { | |
| this.special = 1 /* None */; | |
| this._state = 6 /* InClosingTagName */; | |
| this.sectionStart = this._index - sectionStartOffset; | |
| this._index--; // Reconsume the token | |
| } | |
| else | |
| this._state = 1 /* Text */; | |
| }; | |
| // For entities terminated with a semicolon | |
| Tokenizer.prototype.parseFixedEntity = function (map) { | |
| if (map === void 0) { map = this.xmlMode ? xml_json_1.default : entities_json_1.default; } | |
| // Offset = 1 | |
| if (this.sectionStart + 1 < this._index) { | |
| var entity = this.buffer.substring(this.sectionStart + 1, this._index); | |
| if (Object.prototype.hasOwnProperty.call(map, entity)) { | |
| this.emitPartial(map[entity]); | |
| this.sectionStart = this._index + 1; | |
| } | |
| } | |
| }; | |
| // Parses legacy entities (without trailing semicolon) | |
| Tokenizer.prototype.parseLegacyEntity = function () { | |
| var start = this.sectionStart + 1; | |
| // The max length of legacy entities is 6 | |
| var limit = Math.min(this._index - start, 6); | |
| while (limit >= 2) { | |
| // The min length of legacy entities is 2 | |
| var entity = this.buffer.substr(start, limit); | |
| if (Object.prototype.hasOwnProperty.call(legacy_json_1.default, entity)) { | |
| this.emitPartial(legacy_json_1.default[entity]); | |
| this.sectionStart += limit + 1; | |
| return; | |
| } | |
| limit--; | |
| } | |
| }; | |
| Tokenizer.prototype.stateInNamedEntity = function (c) { | |
| if (c === ";") { | |
| this.parseFixedEntity(); | |
| // Retry as legacy entity if entity wasn't parsed | |
| if (this.baseState === 1 /* Text */ && | |
| this.sectionStart + 1 < this._index && | |
| !this.xmlMode) { | |
| this.parseLegacyEntity(); | |
| } | |
| this._state = this.baseState; | |
| } | |
| else if ((c < "0" || c > "9") && !isASCIIAlpha(c)) { | |
| if (this.xmlMode || this.sectionStart + 1 === this._index) { | |
| // Ignore | |
| } | |
| else if (this.baseState !== 1 /* Text */) { | |
| if (c !== "=") { | |
| // Parse as legacy entity, without allowing additional characters. | |
| this.parseFixedEntity(legacy_json_1.default); | |
| } | |
| } | |
| else { | |
| this.parseLegacyEntity(); | |
| } | |
| this._state = this.baseState; | |
| this._index--; | |
| } | |
| }; | |
| Tokenizer.prototype.decodeNumericEntity = function (offset, base, strict) { | |
| var sectionStart = this.sectionStart + offset; | |
| if (sectionStart !== this._index) { | |
| // Parse entity | |
| var entity = this.buffer.substring(sectionStart, this._index); | |
| var parsed = parseInt(entity, base); | |
| this.emitPartial(decode_codepoint_1.default(parsed)); | |
| this.sectionStart = strict ? this._index + 1 : this._index; | |
| } | |
| this._state = this.baseState; | |
| }; | |
| Tokenizer.prototype.stateInNumericEntity = function (c) { | |
| if (c === ";") { | |
| this.decodeNumericEntity(2, 10, true); | |
| } | |
| else if (c < "0" || c > "9") { | |
| if (!this.xmlMode) { | |
| this.decodeNumericEntity(2, 10, false); | |
| } | |
| else { | |
| this._state = this.baseState; | |
| } | |
| this._index--; | |
| } | |
| }; | |
| Tokenizer.prototype.stateInHexEntity = function (c) { | |
| if (c === ";") { | |
| this.decodeNumericEntity(3, 16, true); | |
| } | |
| else if ((c < "a" || c > "f") && | |
| (c < "A" || c > "F") && | |
| (c < "0" || c > "9")) { | |
| if (!this.xmlMode) { | |
| this.decodeNumericEntity(3, 16, false); | |
| } | |
| else { | |
| this._state = this.baseState; | |
| } | |
| this._index--; | |
| } | |
| }; | |
| Tokenizer.prototype.cleanup = function () { | |
| if (this.sectionStart < 0) { | |
| this.buffer = ""; | |
| this.bufferOffset += this._index; | |
| this._index = 0; | |
| } | |
| else if (this.running) { | |
| if (this._state === 1 /* Text */) { | |
| if (this.sectionStart !== this._index) { | |
| this.cbs.ontext(this.buffer.substr(this.sectionStart)); | |
| } | |
| this.buffer = ""; | |
| this.bufferOffset += this._index; | |
| this._index = 0; | |
| } | |
| else if (this.sectionStart === this._index) { | |
| // The section just started | |
| this.buffer = ""; | |
| this.bufferOffset += this._index; | |
| this._index = 0; | |
| } | |
| else { | |
| // Remove everything unnecessary | |
| this.buffer = this.buffer.substr(this.sectionStart); | |
| this._index -= this.sectionStart; | |
| this.bufferOffset += this.sectionStart; | |
| } | |
| this.sectionStart = 0; | |
| } | |
| }; | |
| /** | |
| * Iterates through the buffer, calling the function corresponding to the current state. | |
| * | |
| * States that are more likely to be hit are higher up, as a performance improvement. | |
| */ | |
| Tokenizer.prototype.parse = function () { | |
| while (this._index < this.buffer.length && this.running) { | |
| var c = this.buffer.charAt(this._index); | |
| if (this._state === 1 /* Text */) { | |
| this.stateText(c); | |
| } | |
| else if (this._state === 12 /* InAttributeValueDq */) { | |
| this.stateInAttributeValueDoubleQuotes(c); | |
| } | |
| else if (this._state === 9 /* InAttributeName */) { | |
| this.stateInAttributeName(c); | |
| } | |
| else if (this._state === 19 /* InComment */) { | |
| this.stateInComment(c); | |
| } | |
| else if (this._state === 20 /* InSpecialComment */) { | |
| this.stateInSpecialComment(c); | |
| } | |
| else if (this._state === 8 /* BeforeAttributeName */) { | |
| this.stateBeforeAttributeName(c); | |
| } | |
| else if (this._state === 3 /* InTagName */) { | |
| this.stateInTagName(c); | |
| } | |
| else if (this._state === 6 /* InClosingTagName */) { | |
| this.stateInClosingTagName(c); | |
| } | |
| else if (this._state === 2 /* BeforeTagName */) { | |
| this.stateBeforeTagName(c); | |
| } | |
| else if (this._state === 10 /* AfterAttributeName */) { | |
| this.stateAfterAttributeName(c); | |
| } | |
| else if (this._state === 13 /* InAttributeValueSq */) { | |
| this.stateInAttributeValueSingleQuotes(c); | |
| } | |
| else if (this._state === 11 /* BeforeAttributeValue */) { | |
| this.stateBeforeAttributeValue(c); | |
| } | |
| else if (this._state === 5 /* BeforeClosingTagName */) { | |
| this.stateBeforeClosingTagName(c); | |
| } | |
| else if (this._state === 7 /* AfterClosingTagName */) { | |
| this.stateAfterClosingTagName(c); | |
| } | |
| else if (this._state === 32 /* BeforeSpecialS */) { | |
| this.stateBeforeSpecialS(c); | |
| } | |
| else if (this._state === 21 /* AfterComment1 */) { | |
| this.stateAfterComment1(c); | |
| } | |
| else if (this._state === 14 /* InAttributeValueNq */) { | |
| this.stateInAttributeValueNoQuotes(c); | |
| } | |
| else if (this._state === 4 /* InSelfClosingTag */) { | |
| this.stateInSelfClosingTag(c); | |
| } | |
| else if (this._state === 16 /* InDeclaration */) { | |
| this.stateInDeclaration(c); | |
| } | |
| else if (this._state === 15 /* BeforeDeclaration */) { | |
| this.stateBeforeDeclaration(c); | |
| } | |
| else if (this._state === 22 /* AfterComment2 */) { | |
| this.stateAfterComment2(c); | |
| } | |
| else if (this._state === 18 /* BeforeComment */) { | |
| this.stateBeforeComment(c); | |
| } | |
| else if (this._state === 33 /* BeforeSpecialSEnd */) { | |
| this.stateBeforeSpecialSEnd(c); | |
| } | |
| else if (this._state === 53 /* BeforeSpecialTEnd */) { | |
| stateAfterSpecialTEnd(this, c); | |
| } | |
| else if (this._state === 39 /* AfterScript1 */) { | |
| stateAfterScript1(this, c); | |
| } | |
| else if (this._state === 40 /* AfterScript2 */) { | |
| stateAfterScript2(this, c); | |
| } | |
| else if (this._state === 41 /* AfterScript3 */) { | |
| stateAfterScript3(this, c); | |
| } | |
| else if (this._state === 34 /* BeforeScript1 */) { | |
| stateBeforeScript1(this, c); | |
| } | |
| else if (this._state === 35 /* BeforeScript2 */) { | |
| stateBeforeScript2(this, c); | |
| } | |
| else if (this._state === 36 /* BeforeScript3 */) { | |
| stateBeforeScript3(this, c); | |
| } | |
| else if (this._state === 37 /* BeforeScript4 */) { | |
| stateBeforeScript4(this, c); | |
| } | |
| else if (this._state === 38 /* BeforeScript5 */) { | |
| this.stateBeforeSpecialLast(c, 2 /* Script */); | |
| } | |
| else if (this._state === 42 /* AfterScript4 */) { | |
| stateAfterScript4(this, c); | |
| } | |
| else if (this._state === 43 /* AfterScript5 */) { | |
| this.stateAfterSpecialLast(c, 6); | |
| } | |
| else if (this._state === 44 /* BeforeStyle1 */) { | |
| stateBeforeStyle1(this, c); | |
| } | |
| else if (this._state === 29 /* InCdata */) { | |
| this.stateInCdata(c); | |
| } | |
| else if (this._state === 45 /* BeforeStyle2 */) { | |
| stateBeforeStyle2(this, c); | |
| } | |
| else if (this._state === 46 /* BeforeStyle3 */) { | |
| stateBeforeStyle3(this, c); | |
| } | |
| else if (this._state === 47 /* BeforeStyle4 */) { | |
| this.stateBeforeSpecialLast(c, 3 /* Style */); | |
| } | |
| else if (this._state === 48 /* AfterStyle1 */) { | |
| stateAfterStyle1(this, c); | |
| } | |
| else if (this._state === 49 /* AfterStyle2 */) { | |
| stateAfterStyle2(this, c); | |
| } | |
| else if (this._state === 50 /* AfterStyle3 */) { | |
| stateAfterStyle3(this, c); | |
| } | |
| else if (this._state === 51 /* AfterStyle4 */) { | |
| this.stateAfterSpecialLast(c, 5); | |
| } | |
| else if (this._state === 52 /* BeforeSpecialT */) { | |
| stateBeforeSpecialT(this, c); | |
| } | |
| else if (this._state === 54 /* BeforeTitle1 */) { | |
| stateBeforeTitle1(this, c); | |
| } | |
| else if (this._state === 55 /* BeforeTitle2 */) { | |
| stateBeforeTitle2(this, c); | |
| } | |
| else if (this._state === 56 /* BeforeTitle3 */) { | |
| stateBeforeTitle3(this, c); | |
| } | |
| else if (this._state === 57 /* BeforeTitle4 */) { | |
| this.stateBeforeSpecialLast(c, 4 /* Title */); | |
| } | |
| else if (this._state === 58 /* AfterTitle1 */) { | |
| stateAfterTitle1(this, c); | |
| } | |
| else if (this._state === 59 /* AfterTitle2 */) { | |
| stateAfterTitle2(this, c); | |
| } | |
| else if (this._state === 60 /* AfterTitle3 */) { | |
| stateAfterTitle3(this, c); | |
| } | |
| else if (this._state === 61 /* AfterTitle4 */) { | |
| this.stateAfterSpecialLast(c, 5); | |
| } | |
| else if (this._state === 17 /* InProcessingInstruction */) { | |
| this.stateInProcessingInstruction(c); | |
| } | |
| else if (this._state === 64 /* InNamedEntity */) { | |
| this.stateInNamedEntity(c); | |
| } | |
| else if (this._state === 23 /* BeforeCdata1 */) { | |
| stateBeforeCdata1(this, c); | |
| } | |
| else if (this._state === 62 /* BeforeEntity */) { | |
| stateBeforeEntity(this, c); | |
| } | |
| else if (this._state === 24 /* BeforeCdata2 */) { | |
| stateBeforeCdata2(this, c); | |
| } | |
| else if (this._state === 25 /* BeforeCdata3 */) { | |
| stateBeforeCdata3(this, c); | |
| } | |
| else if (this._state === 30 /* AfterCdata1 */) { | |
| this.stateAfterCdata1(c); | |
| } | |
| else if (this._state === 31 /* AfterCdata2 */) { | |
| this.stateAfterCdata2(c); | |
| } | |
| else if (this._state === 26 /* BeforeCdata4 */) { | |
| stateBeforeCdata4(this, c); | |
| } | |
| else if (this._state === 27 /* BeforeCdata5 */) { | |
| stateBeforeCdata5(this, c); | |
| } | |
| else if (this._state === 28 /* BeforeCdata6 */) { | |
| this.stateBeforeCdata6(c); | |
| } | |
| else if (this._state === 66 /* InHexEntity */) { | |
| this.stateInHexEntity(c); | |
| } | |
| else if (this._state === 65 /* InNumericEntity */) { | |
| this.stateInNumericEntity(c); | |
| // eslint-disable-next-line @typescript-eslint/no-unnecessary-condition | |
| } | |
| else if (this._state === 63 /* BeforeNumericEntity */) { | |
| stateBeforeNumericEntity(this, c); | |
| } | |
| else { | |
| this.cbs.onerror(Error("unknown _state"), this._state); | |
| } | |
| this._index++; | |
| } | |
| this.cleanup(); | |
| }; | |
| Tokenizer.prototype.finish = function () { | |
| // If there is remaining data, emit it in a reasonable way | |
| if (this.sectionStart < this._index) { | |
| this.handleTrailingData(); | |
| } | |
| this.cbs.onend(); | |
| }; | |
| Tokenizer.prototype.handleTrailingData = function () { | |
| var data = this.buffer.substr(this.sectionStart); | |
| if (this._state === 29 /* InCdata */ || | |
| this._state === 30 /* AfterCdata1 */ || | |
| this._state === 31 /* AfterCdata2 */) { | |
| this.cbs.oncdata(data); | |
| } | |
| else if (this._state === 19 /* InComment */ || | |
| this._state === 21 /* AfterComment1 */ || | |
| this._state === 22 /* AfterComment2 */) { | |
| this.cbs.oncomment(data); | |
| } | |
| else if (this._state === 64 /* InNamedEntity */ && !this.xmlMode) { | |
| this.parseLegacyEntity(); | |
| if (this.sectionStart < this._index) { | |
| this._state = this.baseState; | |
| this.handleTrailingData(); | |
| } | |
| } | |
| else if (this._state === 65 /* InNumericEntity */ && !this.xmlMode) { | |
| this.decodeNumericEntity(2, 10, false); | |
| if (this.sectionStart < this._index) { | |
| this._state = this.baseState; | |
| this.handleTrailingData(); | |
| } | |
| } | |
| else if (this._state === 66 /* InHexEntity */ && !this.xmlMode) { | |
| this.decodeNumericEntity(3, 16, false); | |
| if (this.sectionStart < this._index) { | |
| this._state = this.baseState; | |
| this.handleTrailingData(); | |
| } | |
| } | |
| else if (this._state !== 3 /* InTagName */ && | |
| this._state !== 8 /* BeforeAttributeName */ && | |
| this._state !== 11 /* BeforeAttributeValue */ && | |
| this._state !== 10 /* AfterAttributeName */ && | |
| this._state !== 9 /* InAttributeName */ && | |
| this._state !== 13 /* InAttributeValueSq */ && | |
| this._state !== 12 /* InAttributeValueDq */ && | |
| this._state !== 14 /* InAttributeValueNq */ && | |
| this._state !== 6 /* InClosingTagName */) { | |
| this.cbs.ontext(data); | |
| } | |
| /* | |
| * Else, ignore remaining data | |
| * TODO add a way to remove current tag | |
| */ | |
| }; | |
| Tokenizer.prototype.getSection = function () { | |
| return this.buffer.substring(this.sectionStart, this._index); | |
| }; | |
| Tokenizer.prototype.emitToken = function (name) { | |
| this.cbs[name](this.getSection()); | |
| this.sectionStart = -1; | |
| }; | |
| Tokenizer.prototype.emitPartial = function (value) { | |
| if (this.baseState !== 1 /* Text */) { | |
| this.cbs.onattribdata(value); // TODO implement the new event | |
| } | |
| else { | |
| this.cbs.ontext(value); | |
| } | |
| }; | |
| return Tokenizer; | |
| }()); | |
| exports.default = Tokenizer; | |