neovim/runtime/indent/dtd.vim
Justin M. Keyes a98a6996c2 re-integrate runtime/ vim-patch:0 #938
Vim runtime files based on 7.4.384 / hg changeset 7090d7f160f7

Excluding:
  Amiga icons (*.info, icons/)
  doc/hangulin.txt
  tutor/
  spell/
  lang/ (only used for menu translations)
  macros/maze/, macros/hanoi/, macros/life/, macros/urm/
      These were used to test vi compatibility.
  termcap
      "Demonstration of a termcap file (for the Amiga and Archimedes)"

Helped-by: Rich Wareham <rjw57@cam.ac.uk>
Helped-by: John <john.schmidt.h@gmail.com>
Helped-by: Yann <yann@yann-salaun.com>
Helped-by: Christophe Badoit <c.badoit@lesiteimmo.com>
Helped-by: drasill <github@tof2k.com>
Helped-by: Tae Sandoval Murgan <taecilla@gmail.com>
Helped-by: Lowe Thiderman <lowe.thiderman@gmail.com>
2014-07-29 02:12:31 +00:00

326 lines
11 KiB
VimL
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

" Vim indent file
" Language: DTD (Document Type Definition for XML)
" Maintainer: Nikolai Weibull <now@bitwi.se>
" Latest Revision: 2011-07-08
let s:cpo_save = &cpo
set cpo&vim
setlocal indentexpr=GetDTDIndent()
setlocal indentkeys=!^F,o,O,>
setlocal nosmartindent
if exists("*GetDTDIndent")
finish
endif
" TODO: Needs to be adjusted to stop at [, <, and ].
let s:token_pattern = '^[^[:space:]]\+'
function s:lex1(input, start, ...)
let pattern = a:0 > 0 ? a:1 : s:token_pattern
let start = matchend(a:input, '^\_s*', a:start)
if start == -1
return ["", a:start]
endif
let end = matchend(a:input, pattern, start)
if end == -1
return ["", a:start]
endif
let token = strpart(a:input, start, end - start)
return [token, end]
endfunction
function s:lex(input, start, ...)
let pattern = a:0 > 0 ? a:1 : s:token_pattern
let info = s:lex1(a:input, a:start, pattern)
while info[0] == '--'
let info = s:lex1(a:input, info[1], pattern)
while info[0] != "" && info[0] != '--'
let info = s:lex1(a:input, info[1], pattern)
endwhile
if info[0] == ""
return info
endif
let info = s:lex1(a:input, info[1], pattern)
endwhile
return info
endfunction
function s:indent_to_innermost_parentheses(line, end)
let token = '('
let end = a:end
let parentheses = [end - 1]
while token != ""
let [token, end] = s:lex(a:line, end, '^\%([(),|]\|[A-Za-z0-9_-]\+\|#P\=CDATA\|%[A-Za-z0-9_-]\+;\)[?*+]\=')
if token[0] == '('
call add(parentheses, end - 1)
elseif token[0] == ')'
if len(parentheses) == 1
return [-1, end]
endif
call remove(parentheses, -1)
endif
endwhile
return [parentheses[-1] - strridx(a:line, "\n", parentheses[-1]), end]
endfunction
" TODO: Line and end could be script global (think OO members).
function GetDTDIndent()
if v:lnum == 1
return 0
endif
" Begin by searching back for a <! that isnt inside a comment.
" From here, depending on what follows immediately after, parse to
" where were at to determine what to do.
if search('<!', 'bceW') == 0
return indent(v:lnum - 1)
endif
let lnum = line('.')
let col = col('.')
let indent = indent('.')
let line = lnum == v:lnum ? getline(lnum) : join(getline(lnum, v:lnum - 1), "\n")
let [declaration, end] = s:lex1(line, col)
if declaration == ""
return indent + &sw
elseif declaration == '--'
" Were looking at a comment. Now, simply determine if the comment is
" terminated or not. If it isnt, let Vim take care of that using
" 'comments' and 'autoindent'. Otherwise, indent to the first lines level.
while declaration != ""
let [declaration, end] = s:lex(line, end)
if declaration == "-->"
return indent
endif
endwhile
return -1
elseif declaration == 'ELEMENT'
" Check for element name. If none exists, indent one level.
let [name, end] = s:lex(line, end)
if name == ""
return indent + &sw
endif
" Check for token following element name. This can be a specification of
" whether the start or end tag may be omitted. If nothing is found, indent
" one level.
let [token, end] = s:lex(line, end, '^\%([-O(]\|ANY\|EMPTY\)')
let n = 0
while token =~ '[-O]' && n < 2
let [token, end] = s:lex(line, end, '^\%([-O(]\|ANY\|EMPTY\)')
let n += 1
endwhile
if token == ""
return indent + &sw
endif
" Next comes the content model. If the token weve found isnt a
" parenthesis it must be either ANY, EMPTY or some random junk. Either
" way, were done indenting this element, so set it to that of the first
" line so that the terminating “>” winds up having the same indention.
if token != '('
return indent
endif
" Now go through the content model. We need to keep track of the nesting
" of parentheses. As soon as we hit 0 were done. If that happens we must
" have a complete content model. Thus set indention to be the same as that
" of the first line so that the terminating “>” winds up having the same
" indention. Otherwise, well indent to the innermost parentheses not yet
" matched.
let [indent_of_innermost, end] = s:indent_to_innermost_parentheses(line, end)
if indent_of_innermost != -1
return indent_of_innermost
endif
" Finally, look for any additions and/or exceptions to the content model.
" This is defined by a “+” or “-” followed by another content model
" declaration.
" TODO: Can the “-” be separated by whitespace from the “(”?
let seen = { '+(': 0, '-(': 0 }
while 1
let [additions_exceptions, end] = s:lex(line, end, '^[+-](')
if additions_exceptions != '+(' && additions_exceptions != '-('
let [token, end] = s:lex(line, end)
if token == '>'
return indent
endif
" TODO: Should use s:lex here on getline(v:lnum) and check for >.
return getline(v:lnum) =~ '^\s*>' || count(values(seen), 0) == 0 ? indent : (indent + &sw)
endif
" If weve seen an addition or exception already and this is of the same
" kind, the user is writing a broken DTD. Time to bail.
if seen[additions_exceptions]
return indent
endif
let seen[additions_exceptions] = 1
let [indent_of_innermost, end] = s:indent_to_innermost_parentheses(line, end)
if indent_of_innermost != -1
return indent_of_innermost
endif
endwhile
elseif declaration == 'ATTLIST'
" Check for element name. If none exists, indent one level.
let [name, end] = s:lex(line, end)
if name == ""
return indent + &sw
endif
" Check for any number of attributes.
while 1
" Check for attribute name. If none exists, indent one level, unless the
" current line is a lone “>”, in which case we indent to the same level
" as the first line. Otherwise, if the attribute name is “>”, we have
" actually hit the end of the attribute list, in which case we indent to
" the same level as the first line.
let [name, end] = s:lex(line, end)
if name == ""
" TODO: Should use s:lex here on getline(v:lnum) and check for >.
return getline(v:lnum) =~ '^\s*>' ? indent : (indent + &sw)
elseif name == ">"
return indent
endif
" Check for attribute value declaration. If none exists, indent two
" levels. Otherwise, if its an enumerated value, check for nested
" parentheses and indent to the innermost one if we dont reach the end
" of the listc. Otherwise, just continue with looking for the default
" attribute value.
" TODO: Do validation of keywords
" (CDATA|NMTOKEN|NMTOKENS|ID|IDREF|IDREFS|ENTITY|ENTITIES)?
let [value, end] = s:lex(line, end, '^\%((\|[^[:space:]]\+\)')
if value == ""
return indent + &sw * 2
elseif value == 'NOTATION'
" If this is a enumerated value based on notations, read another token
" for the actual value. If it doesnt exist, indent three levels.
" TODO: If validating according to above, value must be equal to '('.
let [value, end] = s:lex(line, end, '^\%((\|[^[:space:]]\+\)')
if value == ""
return indent + &sw * 3
endif
endif
if value == '('
let [indent_of_innermost, end] = s:indent_to_innermost_parentheses(line, end)
if indent_of_innermost != -1
return indent_of_innermost
endif
endif
" Finally look for the attributes default value. If non exists, indent
" two levels.
let [default, end] = s:lex(line, end, '^\%("\_[^"]*"\|#\(REQUIRED\|IMPLIED\|FIXED\)\)')
if default == ""
return indent + &sw * 2
elseif default == '#FIXED'
" We need to look for the fixed value. If non exists, indent three
" levels.
let [default, end] = s:lex(line, end, '^"\_[^"]*"')
if default == ""
return indent + &sw * 3
endif
endif
endwhile
elseif declaration == 'ENTITY'
" Check for entity name. If none exists, indent one level. Otherwise, if
" the name actually turns out to be a percent sign, “%”, this is a
" parameter entity. Read another token to determine the entity name and,
" again, if none exists, indent one level.
let [name, end] = s:lex(line, end)
if name == ""
return indent + &sw
elseif name == '%'
let [name, end] = s:lex(line, end)
if name == ""
return indent + &sw
endif
endif
" Now check for the entity value. If none exists, indent one level. If it
" does exist, indent to same level as first line, as were now done with
" this entity.
"
" The entity value can be a string in single or double quotes (no escapes
" to worry about, as entities are used instead). However, it can also be
" that this is an external unparsed entity. In that case we have to look
" further for (possibly) a public ID and an URI followed by the NDATA
" keyword and the actual notation name. For the public ID and URI, indent
" two levels, if they dont exist. If the NDATA keyword doesnt exist,
" indent one level. Otherwise, if the actual notation name doesnt exist,
" indent two level. If it does, indent to same level as first line, as
" were now done with this entity.
let [value, end] = s:lex(line, end)
if value == ""
return indent + &sw
elseif value == 'SYSTEM' || value == 'PUBLIC'
let [quoted_string, end] = s:lex(line, end, '\%("[^"]\+"\|''[^'']\+''\)')
if quoted_string == ""
return indent + &sw * 2
endif
if value == 'PUBLIC'
let [quoted_string, end] = s:lex(line, end, '\%("[^"]\+"\|''[^'']\+''\)')
if quoted_string == ""
return indent + &sw * 2
endif
endif
let [ndata, end] = s:lex(line, end)
if ndata == ""
return indent + &sw
endif
let [name, end] = s:lex(line, end)
return name == "" ? (indent + &sw * 2) : indent
else
return indent
endif
elseif declaration == 'NOTATION'
" Check for notation name. If none exists, indent one level.
let [name, end] = s:lex(line, end)
if name == ""
return indent + &sw
endif
" Now check for the external ID. If none exists, indent one level.
let [id, end] = s:lex(line, end)
if id == ""
return indent + &sw
elseif id == 'SYSTEM' || id == 'PUBLIC'
let [quoted_string, end] = s:lex(line, end, '\%("[^"]\+"\|''[^'']\+''\)')
if quoted_string == ""
return indent + &sw * 2
endif
if id == 'PUBLIC'
let [quoted_string, end] = s:lex(line, end, '\%("[^"]\+"\|''[^'']\+''\|>\)')
if quoted_string == ""
" TODO: Should use s:lex here on getline(v:lnum) and check for >.
return getline(v:lnum) =~ '^\s*>' ? indent : (indent + &sw * 2)
elseif quoted_string == '>'
return indent
endif
endif
endif
return indent
endif
" TODO: Processing directives could be indented I suppose. But perhaps its
" just as well to let the user decide how to indent them (perhaps extending
" this function to include proper support for whatever processing directive
" language they want to use).
" Conditional sections are simply passed along to let Vim decide what to do
" (and hence the user).
return -1
endfunction
let &cpo = s:cpo_save
unlet s:cpo_save