Download nvim-linux-x86_64/share/nvim/runtime/autoload/xmlformat.vim from skyzos/first-humanoid-dataset: direct link, hf CLI and curl.
- Browser
- Download file 6.19 kB
-
https://huggingface.co/datasets/skyzos/first-humanoid-dataset/resolve/main/nvim-linux-x86_64/share/nvim/runtime/autoload/xmlformat.vim
- Command line
-
hf download hf://datasets/skyzos/first-humanoid-dataset/nvim-linux-x86_64/share/nvim/runtime/autoload/xmlformat.vim
-
curl -L -o xmlformat.vim https://huggingface.co/datasets/skyzos/first-humanoid-dataset/resolve/main/nvim-linux-x86_64/share/nvim/runtime/autoload/xmlformat.vim
6.19 kB
| " Vim plugin for formatting XML | |
| " Last Change: 2020 Jan 06 | |
| " Version: 0.3 | |
| " Author: Christian Brabandt <cb@256bit.org> | |
| " Repository: https://github.com/chrisbra/vim-xml-ftplugin | |
| " License: VIM License | |
| " Documentation: see :h xmlformat.txt (TODO!) | |
| " --------------------------------------------------------------------- | |
| " Load Once: {{{1 | |
| if exists("g:loaded_xmlformat") || &cp | |
| finish | |
| endif | |
| let g:loaded_xmlformat = 1 | |
| let s:keepcpo = &cpo | |
| set cpo&vim | |
| " Main function: Format the input {{{1 | |
| func! xmlformat#Format() abort | |
| " only allow reformatting through the gq command | |
| " (e.g. Vim is in normal mode) | |
| if mode() != 'n' | |
| " do not fall back to internal formatting | |
| return 0 | |
| endif | |
| let count_orig = v:count | |
| let sw = shiftwidth() | |
| let prev = prevnonblank(v:lnum-1) | |
| let s:indent = indent(prev)/sw | |
| let result = [] | |
| let lastitem = prev ? getline(prev) : '' | |
| let is_xml_decl = 0 | |
| " go through every line, but don't join all content together and join it | |
| " back. We might lose empty lines | |
| let list = getline(v:lnum, (v:lnum + count_orig - 1)) | |
| let current = 0 | |
| for line in list | |
| " Keep empty input lines? | |
| if empty(line) | |
| call add(result, '') | |
| continue | |
| elseif line !~# '<[/]\?[^>]*>' | |
| let nextmatch = match(list, '<[/]\?[^>]*>', current) | |
| if nextmatch > -1 | |
| let line .= ' '. join(list[(current + 1):(nextmatch-1)], " ") | |
| call remove(list, current+1, nextmatch-1) | |
| endif | |
| endif | |
| " split on `>`, but don't split on very first opening < | |
| " this means, items can be like ['<tag>', 'tag content</tag>'] | |
| for item in split(line, '.\@<=[>]\zs') | |
| if s:EndTag(item) | |
| call s:DecreaseIndent() | |
| call add(result, s:Indent(item)) | |
| elseif s:EmptyTag(lastitem) | |
| call add(result, s:Indent(item)) | |
| elseif s:StartTag(lastitem) && s:IsTag(item) | |
| let s:indent += 1 | |
| call add(result, s:Indent(item)) | |
| else | |
| if !s:IsTag(item) | |
| " Simply split on '<', if there is one, | |
| " but reformat according to &textwidth | |
| let t=split(item, '.<\@=\zs') | |
| " if the content fits well within a single line, add it there | |
| " so that the output looks like this: | |
| " | |
| " <foobar>1</foobar> | |
| if s:TagContent(lastitem) is# s:TagContent(t[1]) && strlen(result[-1]) + strlen(item) <= s:Textwidth() | |
| let result[-1] .= item | |
| let lastitem = t[1] | |
| continue | |
| endif | |
| " t should only contain 2 items, but just be safe here | |
| if s:IsTag(lastitem) | |
| let s:indent+=1 | |
| endif | |
| let result+=s:FormatContent([t[0]]) | |
| if s:EndTag(t[1]) | |
| call s:DecreaseIndent() | |
| endif | |
| "for y in t[1:] | |
| let result+=s:FormatContent(t[1:]) | |
| "endfor | |
| else | |
| call add(result, s:Indent(item)) | |
| endif | |
| endif | |
| let lastitem = item | |
| endfor | |
| let current += 1 | |
| endfor | |
| if !empty(result) | |
| let lastprevline = getline(v:lnum + count_orig) | |
| let delete_lastline = v:lnum + count_orig - 1 == line('$') | |
| exe v:lnum. ",". (v:lnum + count_orig - 1). 'd' | |
| call append(v:lnum - 1, result) | |
| " Might need to remove the last line, if it became empty because of the | |
| " append() call | |
| let last = v:lnum + len(result) | |
| " do not use empty(), it returns true for `empty(0)` | |
| if getline(last) is '' && lastprevline is '' && delete_lastline | |
| exe last. 'd' | |
| endif | |
| endif | |
| " do not run internal formatter! | |
| return 0 | |
| endfunc | |
| " Check if given tag is XML Declaration header {{{1 | |
| func! s:IsXMLDecl(tag) abort | |
| return a:tag =~? '^\s*<?xml\s\?\%(version="[^"]*"\)\?\s\?\%(encoding="[^"]*"\)\? ?>\s*$' | |
| endfunc | |
| " Return tag indented by current level {{{1 | |
| func! s:Indent(item) abort | |
| return repeat(' ', shiftwidth()*s:indent). s:Trim(a:item) | |
| endfu | |
| " Return item trimmed from leading whitespace {{{1 | |
| func! s:Trim(item) abort | |
| if exists('*trim') | |
| return trim(a:item) | |
| else | |
| return matchstr(a:item, '\S\+.*') | |
| endif | |
| endfunc | |
| " Check if tag is a new opening tag <tag> {{{1 | |
| func! s:StartTag(tag) abort | |
| let is_comment = s:IsComment(a:tag) | |
| return a:tag =~? '^\s*<[^/?]' && !is_comment | |
| endfunc | |
| " Check if tag is a Comment start {{{1 | |
| func! s:IsComment(tag) abort | |
| return a:tag =~? '<!--' | |
| endfunc | |
| " Remove one level of indentation {{{1 | |
| func! s:DecreaseIndent() abort | |
| let s:indent = (s:indent > 0 ? s:indent - 1 : 0) | |
| endfunc | |
| " Check if tag is a closing tag </tag> {{{1 | |
| func! s:EndTag(tag) abort | |
| return a:tag =~? '^\s*</' | |
| endfunc | |
| " Check that the tag is actually a tag and not {{{1 | |
| " something like "foobar</foobar>" | |
| func! s:IsTag(tag) abort | |
| return s:Trim(a:tag)[0] == '<' | |
| endfunc | |
| " Check if tag is empty <tag/> {{{1 | |
| func! s:EmptyTag(tag) abort | |
| return a:tag =~ '/>\s*$' | |
| endfunc | |
| func! s:TagContent(tag) abort "{{{1 | |
| " Return content of a tag | |
| return substitute(a:tag, '^\s*<[/]\?\([^>]*\)>\s*$', '\1', '') | |
| endfunc | |
| func! s:Textwidth() abort "{{{1 | |
| " return textwidth (or 80 if not set) | |
| return &textwidth == 0 ? 80 : &textwidth | |
| endfunc | |
| " Format input line according to textwidth {{{1 | |
| func! s:FormatContent(list) abort | |
| let result=[] | |
| let limit = s:Textwidth() | |
| let column=0 | |
| let idx = -1 | |
| let add_indent = 0 | |
| let cnt = 0 | |
| for item in a:list | |
| for word in split(item, '\s\+\S\+\zs') | |
| if match(word, '^\s\+$') > -1 | |
| " skip empty words | |
| continue | |
| endif | |
| let column += strdisplaywidth(word, column) | |
| if match(word, "^\\s*\n\\+\\s*$") > -1 | |
| call add(result, '') | |
| let idx += 1 | |
| let column = 0 | |
| let add_indent = 1 | |
| elseif column > limit || cnt == 0 | |
| let add = s:Indent(s:Trim(word)) | |
| call add(result, add) | |
| let column = strdisplaywidth(add) | |
| let idx += 1 | |
| else | |
| if add_indent | |
| let result[idx] = s:Indent(s:Trim(word)) | |
| else | |
| let result[idx] .= ' '. s:Trim(word) | |
| endif | |
| let add_indent = 0 | |
| endif | |
| let cnt += 1 | |
| endfor | |
| endfor | |
| return result | |
| endfunc | |
| " Restoration And Modelines: {{{1 | |
| let &cpo= s:keepcpo | |
| unlet s:keepcpo | |
| " Modeline {{{1 | |
| " vim: fdm=marker fdl=0 ts=2 et sw=0 sts=-1 | |