mirror of
https://github.com/overleaf/overleaf.git
synced 2024-11-07 20:31:06 -05:00
7f37ba737c
* Update Copybara options in preparation for open-sourcing the source editor * Move files * Update paths * Remove source-editor module and checks for its existence * Explicitly mention CM6 license in files that contain code adapted from CM6 GitOrigin-RevId: 89b7cc2b409db01ad103198ccbd1b126ab56349b
770 lines
18 KiB
JavaScript
770 lines
18 KiB
JavaScript
/* Hand-written tokenizer for LaTeX. */
|
|
|
|
import { ExternalTokenizer, ContextTracker } from '@lezer/lr'
|
|
|
|
import {
|
|
LiteralArgContent,
|
|
SpaceDelimitedLiteralArgContent,
|
|
VerbContent,
|
|
VerbatimContent,
|
|
LstInlineContent,
|
|
Begin,
|
|
End,
|
|
KnownEnvironment,
|
|
MathDelimiter,
|
|
Csname,
|
|
TrailingWhitespaceOnly,
|
|
TrailingContent,
|
|
RefCtrlSeq,
|
|
RefStarrableCtrlSeq,
|
|
CiteCtrlSeq,
|
|
CiteStarrableCtrlSeq,
|
|
LabelCtrlSeq,
|
|
MathTextCtrlSeq,
|
|
HboxCtrlSeq,
|
|
TitleCtrlSeq,
|
|
AuthorCtrlSeq,
|
|
DocumentClassCtrlSeq,
|
|
UsePackageCtrlSeq,
|
|
HrefCtrlSeq,
|
|
VerbCtrlSeq,
|
|
LstInlineCtrlSeq,
|
|
IncludeGraphicsCtrlSeq,
|
|
CaptionCtrlSeq,
|
|
DefCtrlSeq,
|
|
LeftCtrlSeq,
|
|
RightCtrlSeq,
|
|
NewCommandCtrlSeq,
|
|
RenewCommandCtrlSeq,
|
|
NewEnvironmentCtrlSeq,
|
|
RenewEnvironmentCtrlSeq,
|
|
DocumentEnvName,
|
|
TabularEnvName,
|
|
EquationEnvName,
|
|
EquationArrayEnvName,
|
|
VerbatimEnvName,
|
|
TikzPictureEnvName,
|
|
FigureEnvName,
|
|
OpenParenCtrlSym,
|
|
CloseParenCtrlSym,
|
|
OpenBracketCtrlSym,
|
|
CloseBracketCtrlSym,
|
|
// Sectioning commands
|
|
BookCtrlSeq,
|
|
PartCtrlSeq,
|
|
ChapterCtrlSeq,
|
|
SectionCtrlSeq,
|
|
SubSectionCtrlSeq,
|
|
SubSubSectionCtrlSeq,
|
|
ParagraphCtrlSeq,
|
|
SubParagraphCtrlSeq,
|
|
InputCtrlSeq,
|
|
IncludeCtrlSeq,
|
|
ItemCtrlSeq,
|
|
BibliographyCtrlSeq,
|
|
BibliographyStyleCtrlSeq,
|
|
CenteringCtrlSeq,
|
|
ListEnvName,
|
|
} from './latex.terms.mjs'
|
|
|
|
function nameChar(ch) {
|
|
// we accept A-Z a-z 0-9 * + @ in environment names
|
|
return (
|
|
(ch >= 65 && ch <= 90) ||
|
|
(ch >= 97 && ch <= 122) ||
|
|
(ch >= 48 && ch <= 57) ||
|
|
ch === 42 ||
|
|
ch === 43 ||
|
|
ch === 64
|
|
)
|
|
}
|
|
|
|
// match [a-zA-Z]
|
|
function alphaChar(ch) {
|
|
return (ch >= 65 && ch <= 90) || (ch >= 97 && ch <= 122)
|
|
}
|
|
|
|
let cachedName = null
|
|
let cachedInput = null
|
|
let cachedPos = 0
|
|
function envNameAfter(input, offset) {
|
|
const pos = input.pos + offset
|
|
if (cachedInput === input && cachedPos === pos) {
|
|
return cachedName
|
|
}
|
|
if (input.peek(offset) !== '{'.charCodeAt(0)) return
|
|
offset++
|
|
let name = ''
|
|
for (;;) {
|
|
const next = input.peek(offset)
|
|
if (!nameChar(next)) break
|
|
name += String.fromCharCode(next)
|
|
offset++
|
|
}
|
|
cachedInput = input
|
|
cachedPos = pos
|
|
return (cachedName = name || null)
|
|
}
|
|
|
|
function ElementContext(name, parent) {
|
|
this.name = name
|
|
this.parent = parent
|
|
this.hash = parent ? parent.hash : 0
|
|
for (let i = 0; i < name.length; i++)
|
|
this.hash +=
|
|
(this.hash << 4) + name.charCodeAt(i) + (name.charCodeAt(i) << 8)
|
|
}
|
|
|
|
export const elementContext = new ContextTracker({
|
|
start: null,
|
|
shift(context, term, stack, input) {
|
|
return term === Begin
|
|
? new ElementContext(envNameAfter(input, '\\begin'.length) || '', context)
|
|
: context
|
|
},
|
|
reduce(context, term) {
|
|
return term === KnownEnvironment && context ? context.parent : context
|
|
},
|
|
reuse(context, node, _stack, input) {
|
|
const type = node.type.id
|
|
return type === Begin
|
|
? new ElementContext(envNameAfter(input, 0) || '', context)
|
|
: context
|
|
},
|
|
hash(context) {
|
|
return context ? context.hash : 0
|
|
},
|
|
strict: false,
|
|
})
|
|
|
|
// tokenizer for \verb|...| commands
|
|
export const verbTokenizer = new ExternalTokenizer(
|
|
(input, stack) => {
|
|
if (input.next === '*'.charCodeAt(0)) input.advance()
|
|
const delimiter = input.next
|
|
if (delimiter === -1) return // hit end of file
|
|
if (/\s|\*/.test(String.fromCharCode(delimiter))) return // invalid delimiter
|
|
input.advance()
|
|
for (;;) {
|
|
const next = input.next
|
|
if (next === -1 || next === CHAR_NEWLINE) return
|
|
input.advance()
|
|
if (next === delimiter) break
|
|
}
|
|
return input.acceptToken(VerbContent)
|
|
},
|
|
{ contextual: false }
|
|
)
|
|
|
|
// tokenizer for \lstinline|...| commands
|
|
export const lstinlineTokenizer = new ExternalTokenizer(
|
|
(input, stack) => {
|
|
let delimiter = input.next
|
|
if (delimiter === -1) return // hit end of file
|
|
if (/\s/.test(String.fromCharCode(delimiter))) {
|
|
return // invalid delimiter
|
|
}
|
|
if (delimiter === CHAR_OPEN_BRACE) {
|
|
delimiter = CHAR_CLOSE_BRACE
|
|
}
|
|
input.advance()
|
|
let content = ''
|
|
for (;;) {
|
|
let next = input.next
|
|
if (next === -1 || next === CHAR_NEWLINE) return
|
|
content += String.fromCharCode(next)
|
|
input.advance()
|
|
if (next === delimiter) break
|
|
}
|
|
return input.acceptToken(LstInlineContent)
|
|
},
|
|
{ contextual: false }
|
|
)
|
|
|
|
const matchForward = (input, expected, offset = 0) => {
|
|
for (let i = 0; i < expected.length; i++) {
|
|
if (String.fromCharCode(input.peek(offset + i)) !== expected[i]) {
|
|
return false
|
|
}
|
|
}
|
|
return true
|
|
}
|
|
|
|
// tokenizer for \begin{verbatim}...\end{verbatim} environments
|
|
export const verbatimTokenizer = new ExternalTokenizer(
|
|
(input, stack) => {
|
|
const delimiter = '\\end{' + stack.context.name + '}'
|
|
let offset = 0
|
|
let end = -1
|
|
for (;;) {
|
|
const next = input.peek(offset)
|
|
if (next === -1) {
|
|
end = offset - 1
|
|
break
|
|
}
|
|
if (matchForward(input, delimiter, offset)) {
|
|
// Found the end marker
|
|
end = offset - 1
|
|
break
|
|
}
|
|
offset++
|
|
}
|
|
return input.acceptToken(VerbatimContent, end + 1)
|
|
},
|
|
{ contextual: false }
|
|
)
|
|
|
|
// tokenizer for \href{...} and similar commands
|
|
export const literalArgTokenizer = new ExternalTokenizer(
|
|
(input, stack) => {
|
|
const delimiter = '}'
|
|
let content = ''
|
|
let offset = 0
|
|
let end = -1
|
|
for (;;) {
|
|
const next = input.peek(offset)
|
|
if (next === -1) {
|
|
end = offset - 1
|
|
break
|
|
}
|
|
content += String.fromCharCode(next)
|
|
if (content.slice(-delimiter.length) === delimiter) {
|
|
// found the '}'
|
|
end = offset - delimiter.length
|
|
break
|
|
}
|
|
offset++
|
|
}
|
|
return input.acceptToken(LiteralArgContent, end + 1)
|
|
},
|
|
{ contextual: false }
|
|
)
|
|
|
|
// tokenizer for literal content delimited by whitespace, such as in `\input foo.tex`
|
|
export const spaceDelimitedLiteralArgTokenizer = new ExternalTokenizer(
|
|
(input, stack) => {
|
|
let content = ''
|
|
let offset = 0
|
|
let end = -1
|
|
for (;;) {
|
|
const next = input.peek(offset)
|
|
if (next === -1) {
|
|
end = offset - 1
|
|
break
|
|
}
|
|
content += String.fromCharCode(next)
|
|
if (content.slice(-1) === ' ' || content.slice(-1) === '\n') {
|
|
// found the whitespace
|
|
end = offset - 1
|
|
break
|
|
}
|
|
offset++
|
|
}
|
|
return input.acceptToken(SpaceDelimitedLiteralArgContent, end + 1)
|
|
},
|
|
{ contextual: false }
|
|
)
|
|
|
|
// helper function to look up charCodes
|
|
function _char(s) {
|
|
return s.charCodeAt(0)
|
|
}
|
|
|
|
// Allowed delimiters, from the LaTeX manual, table 3.10
|
|
// ( ) [ ] / | \{ \} \| and additional names below
|
|
// The empty delimiter . is also allowed
|
|
|
|
const CHAR_SLASH = _char('/')
|
|
const CHAR_PIPE = _char('|')
|
|
const CHAR_OPEN_PAREN = _char('(')
|
|
const CHAR_CLOSE_PAREN = _char(')')
|
|
const CHAR_OPEN_BRACKET = _char('[')
|
|
const CHAR_CLOSE_BRACKET = _char(']')
|
|
const CHAR_FULL_STOP = _char('.')
|
|
const CHAR_BACKSLASH = _char('\\')
|
|
const CHAR_OPEN_BRACE = _char('{')
|
|
const CHAR_CLOSE_BRACE = _char('}')
|
|
|
|
const ALLOWED_DELIMITER_NAMES = [
|
|
'lfloor',
|
|
'rfloor',
|
|
'lceil',
|
|
'rceil',
|
|
'langle',
|
|
'rangle',
|
|
'backslash',
|
|
'uparrow',
|
|
'downarrow',
|
|
'Uparrow',
|
|
'Downarrow',
|
|
'updownarrow',
|
|
'Updownarrow',
|
|
'lvert',
|
|
'rvert',
|
|
'lVert',
|
|
'rVert',
|
|
]
|
|
|
|
// Given a list of allowed command names, return those with leading characters that are the same as the matchString
|
|
function findPartialMatches(list, matchString) {
|
|
const size = matchString.length
|
|
return list.filter(
|
|
entry => entry.length >= size && entry.substring(0, size) === matchString
|
|
)
|
|
}
|
|
|
|
// tokenizer for \leftX ... \rightX delimiter tokens
|
|
export const mathDelimiterTokenizer = new ExternalTokenizer(
|
|
(input, stack) => {
|
|
let content = ''
|
|
let offset = 0
|
|
let end = -1
|
|
// look at the first character, we only accept the following /|()[].
|
|
let next = input.peek(offset)
|
|
if (next === -1) {
|
|
return
|
|
}
|
|
if (
|
|
next === CHAR_SLASH ||
|
|
next === CHAR_PIPE ||
|
|
next === CHAR_OPEN_PAREN ||
|
|
next === CHAR_CLOSE_PAREN ||
|
|
next === CHAR_OPEN_BRACKET ||
|
|
next === CHAR_CLOSE_BRACKET ||
|
|
next === CHAR_FULL_STOP
|
|
) {
|
|
return input.acceptToken(MathDelimiter, 1)
|
|
}
|
|
// reject anything else not starting with a backslash,
|
|
// we only accept control symbols or control sequences
|
|
if (next !== CHAR_BACKSLASH) {
|
|
return
|
|
}
|
|
// look at the second character, we only accept \{ and \} and \| as control symbols
|
|
offset++
|
|
next = input.peek(offset)
|
|
if (next === -1) {
|
|
return
|
|
}
|
|
if (
|
|
next === CHAR_OPEN_BRACE ||
|
|
next === CHAR_CLOSE_BRACE ||
|
|
next === CHAR_PIPE
|
|
) {
|
|
return input.acceptToken(MathDelimiter, 2)
|
|
}
|
|
// We haven't matched any symbols, so now try matching command names.
|
|
// Is this character a potential match to the remaining allowed delimiter names?
|
|
content = String.fromCharCode(next)
|
|
let candidates = findPartialMatches(ALLOWED_DELIMITER_NAMES, content)
|
|
if (!candidates.length) return
|
|
// we have some candidates, look at subsequent characters
|
|
offset++
|
|
for (;;) {
|
|
const next = input.peek(offset)
|
|
// stop when we reach the end of file or a non-alphabetic character
|
|
if (next === -1 || !nameChar(next)) {
|
|
end = offset - 1
|
|
break
|
|
}
|
|
content += String.fromCharCode(next)
|
|
// find how many candidates remain with the new input
|
|
candidates = findPartialMatches(candidates, content)
|
|
if (!candidates.length) return // no matches remaining
|
|
end = offset
|
|
offset++
|
|
}
|
|
if (!candidates.includes(content)) return // not a valid delimiter
|
|
// accept the content as a valid delimiter
|
|
return input.acceptToken(MathDelimiter, end + 1)
|
|
},
|
|
{ contextual: false }
|
|
)
|
|
|
|
const CHAR_AT_SYMBOL = _char('@')
|
|
|
|
export const csnameTokenizer = new ExternalTokenizer((input, stack) => {
|
|
let offset = 0
|
|
let end = -1
|
|
// look at the first character, we are looking for acceptable control sequence names
|
|
// including @ signs, \\[a-zA-Z@]+
|
|
const next = input.peek(offset)
|
|
if (next === -1) {
|
|
return
|
|
}
|
|
// reject anything not starting with a backslash,
|
|
// we only accept control sequences
|
|
if (next !== CHAR_BACKSLASH) {
|
|
return
|
|
}
|
|
offset++
|
|
for (;;) {
|
|
const next = input.peek(offset)
|
|
// stop when we reach the end of file or a non-csname character
|
|
if (next === -1 || !(alphaChar(next) || next === CHAR_AT_SYMBOL)) {
|
|
end = offset - 1
|
|
break
|
|
}
|
|
end = offset
|
|
offset++
|
|
}
|
|
if (end === -1) return
|
|
// accept the content as a valid control sequence
|
|
return input.acceptToken(Csname, end + 1)
|
|
})
|
|
|
|
const CHAR_SPACE = _char(' ')
|
|
const CHAR_NEWLINE = _char('\n')
|
|
const END_DOCUMENT_MARK = '\\end{document}'.split('').reverse()
|
|
|
|
export const trailingContentTokenizer = new ExternalTokenizer(
|
|
(input, stack) => {
|
|
if (input.next === -1) return // no trailing content
|
|
// Look back for end-document mark, bail out if any characters do not match
|
|
for (let i = 1; i < END_DOCUMENT_MARK.length + 1; i++) {
|
|
if (String.fromCharCode(input.peek(-i)) !== END_DOCUMENT_MARK[i - 1]) {
|
|
return
|
|
}
|
|
}
|
|
while (input.next === CHAR_SPACE || input.next === CHAR_NEWLINE) {
|
|
const next = input.advance()
|
|
if (next === -1) return input.acceptToken(TrailingWhitespaceOnly) // trailing whitespace only
|
|
}
|
|
// accept the all content up to the end of the document
|
|
while (input.advance() !== -1) {
|
|
//
|
|
}
|
|
return input.acceptToken(TrailingContent)
|
|
}
|
|
)
|
|
|
|
const refCommands = new Set([
|
|
'\\fullref',
|
|
'\\Vref',
|
|
'\\autopageref',
|
|
'\\autoref',
|
|
'\\eqref',
|
|
'\\labelcpageref',
|
|
'\\labelcref',
|
|
'\\lcnamecref',
|
|
'\\lcnamecrefs',
|
|
'\\namecref',
|
|
'\\nameCref',
|
|
'\\namecrefs',
|
|
'\\nameCrefs',
|
|
'\\thnameref',
|
|
'\\thref',
|
|
'\\titleref',
|
|
'\\vrefrange',
|
|
'\\Crefrange',
|
|
'\\Crefrang',
|
|
])
|
|
|
|
const refStarrableCommands = new Set([
|
|
'\\vpageref',
|
|
'\\vref',
|
|
'\\zcpageref',
|
|
'\\zcref',
|
|
'\\zfullref',
|
|
'\\zref',
|
|
'\\zvpageref',
|
|
'\\zvref',
|
|
'\\cref',
|
|
'\\Cref',
|
|
'\\pageref',
|
|
'\\ref',
|
|
'\\Ref',
|
|
'\\zpageref',
|
|
'\\ztitleref',
|
|
'\\vpagerefrange',
|
|
'\\zvpagerefrange',
|
|
'\\zvrefrange',
|
|
'\\crefrange',
|
|
])
|
|
|
|
const citeCommands = new Set([
|
|
'\\autocites',
|
|
'\\Autocites',
|
|
'\\Cite',
|
|
'\\citeA',
|
|
'\\citealp',
|
|
'\\Citealp',
|
|
'\\citealt',
|
|
'\\Citealt',
|
|
'\\citeauthorNP',
|
|
'\\citeauthorp',
|
|
'\\Citeauthorp',
|
|
'\\citeauthort',
|
|
'\\Citeauthort',
|
|
'\\citeNP',
|
|
'\\citenum',
|
|
'\\cites',
|
|
'\\Cites',
|
|
'\\citeurl',
|
|
'\\citeyearpar',
|
|
'\\defcitealias',
|
|
'\\fnotecite',
|
|
'\\footcite',
|
|
'\\footcitetext',
|
|
'\\footfullcite',
|
|
'\\footnotecites',
|
|
'\\Footnotecites',
|
|
'\\fullcite',
|
|
'\\fullciteA',
|
|
'\\fullciteauthor',
|
|
'\\fullciteauthorNP',
|
|
'\\maskcite',
|
|
'\\maskciteA',
|
|
'\\maskcitealp',
|
|
'\\maskCitealp',
|
|
'\\maskcitealt',
|
|
'\\maskCitealt',
|
|
'\\maskciteauthor',
|
|
'\\maskciteauthorNP',
|
|
'\\maskciteauthorp',
|
|
'\\maskCiteauthorp',
|
|
'\\maskciteauthort',
|
|
'\\maskCiteauthort',
|
|
'\\maskciteNP',
|
|
'\\maskcitenum',
|
|
'\\maskcitep',
|
|
'\\maskCitep',
|
|
'\\maskcitepalias',
|
|
'\\maskcitet',
|
|
'\\maskCitet',
|
|
'\\maskcitetalias',
|
|
'\\maskciteyear',
|
|
'\\maskciteyearNP',
|
|
'\\maskciteyearpar',
|
|
'\\maskfullcite',
|
|
'\\maskfullciteA',
|
|
'\\maskfullciteauthor',
|
|
'\\maskfullciteauthorNP',
|
|
'\\masknocite',
|
|
'\\maskshortcite',
|
|
'\\maskshortciteA',
|
|
'\\maskshortciteauthor',
|
|
'\\maskshortciteauthorNP',
|
|
'\\maskshortciteNP',
|
|
'\\mautocite',
|
|
'\\Mautocite',
|
|
'\\mcite',
|
|
'\\Mcite',
|
|
'\\mfootcite',
|
|
'\\mfootcitetext',
|
|
'\\mparencite',
|
|
'\\Mparencite',
|
|
'\\msupercite',
|
|
'\\mtextcite',
|
|
'\\Mtextcite',
|
|
'\\nocite',
|
|
'\\nocitemeta',
|
|
'\\notecite',
|
|
'\\Parencite',
|
|
'\\parencites',
|
|
'\\Parencites',
|
|
'\\pnotecite',
|
|
'\\shortcite',
|
|
'\\shortciteA',
|
|
'\\shortciteauthor',
|
|
'\\shortciteauthorNP',
|
|
'\\shortciteNP',
|
|
'\\smartcite',
|
|
'\\Smartcite',
|
|
'\\smartcites',
|
|
'\\Smartcites',
|
|
'\\supercite',
|
|
'\\supercites',
|
|
'\\textcite',
|
|
'\\Textcite',
|
|
'\\textcites',
|
|
'\\Textcites',
|
|
])
|
|
|
|
const citeStarredCommands = new Set([
|
|
'\\cite',
|
|
'\\citeauthor',
|
|
'\\Citeauthor',
|
|
'\\citedate',
|
|
'\\citep',
|
|
'\\Citep',
|
|
'\\citetitle',
|
|
'\\citeyear',
|
|
'\\parencite',
|
|
'\\citet',
|
|
'\\autocite',
|
|
'\\Autocite',
|
|
])
|
|
|
|
const labelCommands = new Set(['\\label', '\\thlabel', '\\zlabel'])
|
|
|
|
const mathTextCommands = new Set(['\\text', '\\tag', '\\textrm', '\\intertext'])
|
|
|
|
const otherKnowncommands = {
|
|
'\\hbox': HboxCtrlSeq,
|
|
'\\title': TitleCtrlSeq,
|
|
'\\author': AuthorCtrlSeq,
|
|
'\\documentclass': DocumentClassCtrlSeq,
|
|
'\\usepackage': UsePackageCtrlSeq,
|
|
'\\href': HrefCtrlSeq,
|
|
'\\verb': VerbCtrlSeq,
|
|
'\\lstinline': LstInlineCtrlSeq,
|
|
'\\includegraphics': IncludeGraphicsCtrlSeq,
|
|
'\\caption': CaptionCtrlSeq,
|
|
'\\def': DefCtrlSeq,
|
|
'\\left': LeftCtrlSeq,
|
|
'\\right': RightCtrlSeq,
|
|
'\\newcommand': NewCommandCtrlSeq,
|
|
'\\renewcommand': RenewCommandCtrlSeq,
|
|
'\\newenvironment': NewEnvironmentCtrlSeq,
|
|
'\\renewenvironment': RenewEnvironmentCtrlSeq,
|
|
'\\book': BookCtrlSeq,
|
|
'\\part': PartCtrlSeq,
|
|
'\\addpart': PartCtrlSeq,
|
|
'\\chapter': ChapterCtrlSeq,
|
|
'\\addchap': ChapterCtrlSeq,
|
|
'\\section': SectionCtrlSeq,
|
|
'\\addseq': SectionCtrlSeq,
|
|
'\\subsection': SubSectionCtrlSeq,
|
|
'\\subsubsection': SubSubSectionCtrlSeq,
|
|
'\\paragraph': ParagraphCtrlSeq,
|
|
'\\subparagraph': SubParagraphCtrlSeq,
|
|
'\\input': InputCtrlSeq,
|
|
'\\include': IncludeCtrlSeq,
|
|
'\\item': ItemCtrlSeq,
|
|
'\\centering': CenteringCtrlSeq,
|
|
'\\bibliography': BibliographyCtrlSeq,
|
|
'\\bibliographystyle': BibliographyStyleCtrlSeq,
|
|
}
|
|
// specializer for control sequences
|
|
// return new tokens for specific control sequences
|
|
export const specializeCtrlSeq = (name, terms) => {
|
|
if (name === '\\begin') return Begin
|
|
if (name === '\\end') return End
|
|
if (refCommands.has(name)) {
|
|
return RefCtrlSeq
|
|
}
|
|
if (refStarrableCommands.has(name)) {
|
|
return RefStarrableCtrlSeq
|
|
}
|
|
if (citeCommands.has(name)) {
|
|
return CiteCtrlSeq
|
|
}
|
|
if (citeStarredCommands.has(name)) {
|
|
return CiteStarrableCtrlSeq
|
|
}
|
|
if (labelCommands.has(name)) {
|
|
return LabelCtrlSeq
|
|
}
|
|
if (mathTextCommands.has(name)) {
|
|
return MathTextCtrlSeq
|
|
}
|
|
return otherKnowncommands[name] || -1
|
|
}
|
|
|
|
const tabularEnvNames = new Set([
|
|
'tabular',
|
|
'xltabular',
|
|
'tabularx',
|
|
'longtable',
|
|
])
|
|
|
|
const equationEnvNames = new Set([
|
|
'equation',
|
|
'equation*',
|
|
'displaymath',
|
|
'displaymath*',
|
|
'math',
|
|
'math*',
|
|
'multline',
|
|
'multline*',
|
|
'matrix',
|
|
'tikzcd',
|
|
])
|
|
|
|
const equationArrayEnvNames = new Set([
|
|
'array',
|
|
'eqnarray',
|
|
'eqnarray*',
|
|
'align',
|
|
'align*',
|
|
'alignat',
|
|
'alignat*',
|
|
'flalign',
|
|
'flalign*',
|
|
'gather',
|
|
'gather*',
|
|
'pmatrix',
|
|
'pmatrix*',
|
|
'bmatrix',
|
|
'bmatrix*',
|
|
'Bmatrix',
|
|
'Bmatrix*',
|
|
'vmatrix',
|
|
'vmatrix*',
|
|
'Vmatrix',
|
|
'Vmatrix*',
|
|
'smallmatrix',
|
|
'smallmatrix*',
|
|
'split',
|
|
'split*',
|
|
'gathered',
|
|
'gathered*',
|
|
'aligned',
|
|
'aligned*',
|
|
'alignedat',
|
|
'alignedat*',
|
|
'cases',
|
|
'cases*',
|
|
'dcases',
|
|
'dcases*',
|
|
'IEEEeqnarray',
|
|
'IEEEeqnarray*',
|
|
])
|
|
|
|
const verbatimEnvNames = new Set([
|
|
'verbatim',
|
|
'boxedverbatim',
|
|
'lstlisting',
|
|
'minted',
|
|
'Verbatim',
|
|
'lstlisting',
|
|
'codeexample',
|
|
'comment',
|
|
])
|
|
|
|
const otherKnownEnvNames = {
|
|
document: DocumentEnvName,
|
|
tikzpicture: TikzPictureEnvName,
|
|
figure: FigureEnvName,
|
|
subfigure: FigureEnvName,
|
|
enumerate: ListEnvName,
|
|
itemize: ListEnvName,
|
|
}
|
|
|
|
export const specializeEnvName = (name, terms) => {
|
|
if (tabularEnvNames.has(name)) {
|
|
return TabularEnvName
|
|
}
|
|
if (equationEnvNames.has(name)) {
|
|
return EquationEnvName
|
|
}
|
|
if (equationArrayEnvNames.has(name)) {
|
|
return EquationArrayEnvName
|
|
}
|
|
if (verbatimEnvNames.has(name)) {
|
|
return VerbatimEnvName
|
|
}
|
|
return otherKnownEnvNames[name] || -1
|
|
}
|
|
|
|
const otherKnownCtrlSyms = {
|
|
'\\(': OpenParenCtrlSym,
|
|
'\\)': CloseParenCtrlSym,
|
|
'\\[': OpenBracketCtrlSym,
|
|
'\\]': CloseBracketCtrlSym,
|
|
}
|
|
|
|
export const specializeCtrlSym = (name, terms) => {
|
|
return otherKnownCtrlSyms[name] || -1
|
|
}
|