summaryrefslogtreecommitdiffstats
path: root/runtime/autoload/dist/vimindent.vim
diff options
context:
space:
mode:
Diffstat (limited to 'runtime/autoload/dist/vimindent.vim')
-rw-r--r--runtime/autoload/dist/vimindent.vim1200
1 files changed, 1200 insertions, 0 deletions
diff --git a/runtime/autoload/dist/vimindent.vim b/runtime/autoload/dist/vimindent.vim
new file mode 100644
index 0000000000..572fc7c6cb
--- /dev/null
+++ b/runtime/autoload/dist/vimindent.vim
@@ -0,0 +1,1200 @@
+vim9script
+
+# Language: Vim script
+# Maintainer: github user lacygoill
+# Last Change: 2022 Sep 24
+
+# Config {{{1
+
+const TIMEOUT: number = get(g:, 'vim_indent', {})
+ ->get('searchpair_timeout', 100)
+
+def IndentMoreInBracketBlock(): number # {{{2
+ if get(g:, 'vim_indent', {})
+ ->get('more_in_bracket_block', false)
+ return shiftwidth()
+ else
+ return 0
+ endif
+enddef
+
+def IndentMoreLineContinuation(): number # {{{2
+ var n: any = get(g:, 'vim_indent', {})
+ # We inspect `g:vim_indent_cont` to stay backward compatible.
+ ->get('line_continuation', get(g:, 'vim_indent_cont', shiftwidth() * 3))
+
+ if n->typename() == 'string'
+ return n->eval()
+ else
+ return n
+ endif
+enddef
+# }}}2
+
+# Init {{{1
+var patterns: list<string>
+# Tokens {{{2
+# BAR_SEPARATION {{{3
+
+const BAR_SEPARATION: string = '[^|\\]\@1<=|'
+
+# OPENING_BRACKET {{{3
+
+const OPENING_BRACKET: string = '[[{(]'
+
+# CLOSING_BRACKET {{{3
+
+const CLOSING_BRACKET: string = '[]})]'
+
+# NON_BRACKET {{{3
+
+const NON_BRACKET: string = '[^[\]{}()]'
+
+# LIST_OR_DICT_CLOSING_BRACKET {{{3
+
+const LIST_OR_DICT_CLOSING_BRACKET: string = '[]}]'
+
+# LIST_OR_DICT_OPENING_BRACKET {{{3
+
+const LIST_OR_DICT_OPENING_BRACKET: string = '[[{]'
+
+# CHARACTER_UNDER_CURSOR {{{3
+
+const CHARACTER_UNDER_CURSOR: string = '\%.c.'
+
+# INLINE_COMMENT {{{3
+
+# TODO: It is not required for an inline comment to be surrounded by whitespace.
+# But it might help against false positives.
+# To be more reliable, we should inspect the syntax, and only require whitespace
+# before the `#` comment leader. But that might be too costly (because of
+# `synstack()`).
+const INLINE_COMMENT: string = '\s[#"]\%(\s\|[{}]\{3}\)'
+
+# INLINE_VIM9_COMMENT {{{3
+
+const INLINE_VIM9_COMMENT: string = '\s#'
+
+# COMMENT {{{3
+
+# TODO: Technically, `"\s` is wrong.
+#
+# First, whitespace is not required.
+# Second, in Vim9, a string might appear at the start of the line.
+# To be sure, we should also inspect the syntax.
+# We can't use `INLINE_COMMENT` here. {{{
+#
+# const COMMENT: string = $'^\s*{INLINE_COMMENT}'
+# ^------------^
+# ✘
+#
+# Because `INLINE_COMMENT` asserts the presence of a whitespace before the
+# comment leader. This assertion is not satisfied for a comment starting at the
+# start of the line.
+#}}}
+const COMMENT: string = '^\s*\%(#\|"\\\=\s\).*$'
+
+# DICT_KEY {{{3
+
+const DICT_KEY: string = '^\s*\%('
+ .. '\%(\w\|-\)\+'
+ .. '\|'
+ .. '"[^"]*"'
+ .. '\|'
+ .. "'[^']*'"
+ .. '\|'
+ .. '\[[^]]\+\]'
+ .. '\)'
+ .. ':\%(\s\|$\)'
+
+# END_OF_COMMAND {{{3
+
+const END_OF_COMMAND: string = $'\s*\%($\|||\@!\|{INLINE_COMMENT}\)'
+
+# END_OF_LINE {{{3
+
+const END_OF_LINE: string = $'\s*\%($\|{INLINE_COMMENT}\)'
+
+# END_OF_VIM9_LINE {{{3
+
+const END_OF_VIM9_LINE: string = $'\s*\%($\|{INLINE_VIM9_COMMENT}\)'
+
+# OPERATOR {{{3
+
+const OPERATOR: string = '\%(^\|\s\)\%([-+*/%]\|\.\.\|||\|&&\|??\|?\|<<\|>>\|\%([=!]=\|[<>]=\=\|[=!]\~\|is\|isnot\)[?#]\=\)\%(\s\|$\)\@=\%(\s*[|<]\)\@!'
+ # assignment operators
+ .. '\|' .. '\s\%([-+*/%]\|\.\.\)\==\%(\s\|$\)\@='
+ # support `:` when used inside conditional operator `?:`
+ .. '\|' .. '\%(\s\|^\):\%(\s\|$\)'
+
+# HEREDOC_OPERATOR {{{3
+
+const HEREDOC_OPERATOR: string = '\s=<<\s\@=\%(\s\+\%(trim\|eval\)\)\{,2}'
+
+# PATTERN_DELIMITER {{{3
+
+# A better regex would be:
+#
+# [^-+*/%.:# \t[:alnum:]\"|]\@=.\|->\@!\%(=\s\)\@!\|[+*/%]\%(=\s\)\@!
+#
+# But sometimes, it can be too costly and cause `E363` to be given.
+const PATTERN_DELIMITER: string = '[-+*/%]\%(=\s\)\@!'
+
+# QUOTE {{{3
+
+const QUOTE: string = '["'']'
+# }}}2
+# Syntaxes {{{2
+# ASSIGNS_HEREDOC {{{3
+
+const ASSIGNS_HEREDOC: string = $'^\%({COMMENT}\)\@!.*\%({HEREDOC_OPERATOR}\)\s\+\zs[A-Z]\+{END_OF_LINE}'
+
+# CD_COMMAND {{{3
+
+const CD_COMMAND: string = $'[lt]\=cd!\=\s\+-{END_OF_COMMAND}'
+
+# HIGHER_ORDER_COMMAND {{{3
+
+patterns =<< trim eval END
+ argdo\>!\=
+ bufdo\>!\=
+ cdo\>!\=
+ folddoc\%[losed]\>
+ foldd\%[oopen]\>
+ ldo\=\>!\=
+ tabdo\=\>
+ windo\>
+ au\%[tocmd]\>.*
+ com\%[mand]\>.*
+ g\%[lobal]!\={PATTERN_DELIMITER}.*
+ v\%[global]!\={PATTERN_DELIMITER}.*
+END
+const HIGHER_ORDER_COMMAND: string = $'\%(^\|{BAR_SEPARATION}\)\s*\<\%(' .. patterns->join('\|') .. '\):\@!'
+
+# MAPPING_COMMAND {{{3
+
+const MAPPING_COMMAND: string = '\%(\<sil\%[ent]!\=\s\+\)\=[nvxsoilct]\=\%(nore\|un\)map!\=\s'
+
+# NORMAL_COMMAND {{{3
+
+const NORMAL_COMMAND: string = '\<norm\%[al]!\=\s*\S\+$'
+
+# PLUS_MINUS_COMMAND {{{3
+
+# In legacy, the `:+` and `:-` commands are not required to be preceded by a colon.
+# As a result, when `+` or `-` is alone on a line, there is ambiguity.
+# It might be an operator or a command.
+# To not break the indentation in legacy scripts, we might need to consider such
+# lines as commands.
+const PLUS_MINUS_COMMAND: string = '^\s*[+-]\s*$'
+
+# ENDS_BLOCK {{{3
+
+const ENDS_BLOCK: string = '^\s*\%('
+ .. 'en\%[dif]'
+ .. '\|' .. 'endfor\='
+ .. '\|' .. 'endw\%[hile]'
+ .. '\|' .. 'endt\%[ry]'
+ .. '\|' .. 'enddef'
+ .. '\|' .. 'endf\%[unction]'
+ .. '\|' .. 'aug\%[roup]\s\+[eE][nN][dD]'
+ .. '\|' .. CLOSING_BRACKET
+ .. $'\){END_OF_COMMAND}'
+
+# ENDS_BLOCK_OR_CLAUSE {{{3
+
+patterns =<< trim END
+ en\%[dif]
+ el\%[se]
+ endfor\=
+ endw\%[hile]
+ endt\%[ry]
+ fina\|finally\=
+ enddef
+ endf\%[unction]
+ aug\%[roup]\s\+[eE][nN][dD]
+END
+
+const ENDS_BLOCK_OR_CLAUSE: string = '^\s*\%(' .. patterns->join('\|') .. $'\){END_OF_COMMAND}'
+ .. $'\|^\s*cat\%[ch]\%(\s\+\({PATTERN_DELIMITER}\).*\1\)\={END_OF_COMMAND}'
+ .. $'\|^\s*elseif\=\s\+\%({OPERATOR}\)\@!'
+
+# STARTS_CURLY_BLOCK {{{3
+
+# TODO: `{` alone on a line is not necessarily the start of a block.
+# It could be a dictionary if the previous line ends with a binary/ternary
+# operator. This can cause an issue whenever we use `STARTS_CURLY_BLOCK` or
+# `LINE_CONTINUATION_AT_EOL`.
+const STARTS_CURLY_BLOCK: string = '\%('
+ .. '^\s*{'
+ .. '\|' .. '^.*\zs\s=>\s\+{'
+ .. '\|' .. $'^\%(\s*\|.*{BAR_SEPARATION}\s*\)\%(com\%[mand]\|au\%[tocmd]\).*\zs\s{{'
+ .. '\)' .. END_OF_COMMAND
+
+# STARTS_NAMED_BLOCK {{{3
+
+# All of these will be used at the start of a line (or after a bar).
+# NOTE: Don't replace `\%x28` with `(`.{{{
+#
+# Otherwise, the paren would be unbalanced which might cause syntax highlighting
+# issues much later in the code of the current script (sometimes, the syntax
+# highlighting plugin fails to correctly recognize a heredoc which is far away
+# and/or not displayed because inside a fold).
+# }}}
+patterns =<< trim END
+ if
+ el\%[se]
+ elseif\=
+ for
+ wh\%[ile]
+ try
+ cat\%[ch]
+ fina\|finally\=
+ fu\%[nction]\%x28\@!
+ \%(export\s\+\)\=def
+ aug\%[roup]\%(\s\+[eE][nN][dD]\)\@!\s\+\S\+
+END
+const STARTS_NAMED_BLOCK: string = '^\s*\%(sil\%[ent]\s\+\)\=\%(' .. patterns->join('\|') .. '\)\>:\@!'
+
+# STARTS_FUNCTION {{{3
+
+const STARTS_FUNCTION: string = '^\s*\%(export\s\+\)\=def\>:\@!'
+
+# ENDS_FUNCTION {{{3
+
+const ENDS_FUNCTION: string = $'^\s*enddef\>:\@!{END_OF_COMMAND}'
+
+# START_MIDDLE_END {{{3
+
+const START_MIDDLE_END: dict<list<string>> = {
+ if: ['if', 'el\%[se]\|elseif\=', 'en\%[dif]'],
+ else: ['if', 'el\%[se]\|elseif\=', 'en\%[dif]'],
+ elseif: ['if', 'el\%[se]\|elseif\=', 'en\%[dif]'],
+ endif: ['if', 'el\%[se]\|elseif\=', 'en\%[dif]'],
+ for: ['for', '', 'endfor\='],
+ endfor: ['for', '', 'endfor\='],
+ while: ['wh\%[ile]', '', 'endw\%[hile]'],
+ endwhile: ['wh\%[ile]', '', 'endw\%[hile]'],
+ try: ['try', 'cat\%[ch]\|fina\|finally\=', 'endt\%[ry]'],
+ catch: ['try', 'cat\%[ch]\|fina\|finally\=', 'endt\%[ry]'],
+ finally: ['try', 'cat\%[ch]\|fina\|finally\=', 'endt\%[ry]'],
+ endtry: ['try', 'cat\%[ch]\|fina\|finally\=', 'endt\%[ry]'],
+ def: ['\%(export\s\+\)\=def', '', 'enddef'],
+ enddef: ['\%(export\s\+\)\=def', '', 'enddef'],
+ function: ['fu\%[nction]', '', 'endf\%[unction]'],
+ endfunction: ['fu\%[nction]', '', 'endf\%[unction]'],
+ augroup: ['aug\%[roup]\%(\s\+[eE][nN][dD]\)\@!\s\+\S\+', '', 'aug\%[roup]\s\+[eE][nN][dD]'],
+}->map((_, kwds: list<string>) =>
+ kwds->map((_, kwd: string) => kwd == ''
+ ? ''
+ : $'\%(^\|{BAR_SEPARATION}\|\<sil\%[ent]\|{HIGHER_ORDER_COMMAND}\)\s*'
+ .. $'\%({printf('\C\<\%%(%s\)\>:\@!\%%(\s*%s\)\@!', kwd, OPERATOR)}\)'))
+# }}}2
+# EOL {{{2
+# OPENING_BRACKET_AT_EOL {{{3
+
+const OPENING_BRACKET_AT_EOL: string = $'{OPENING_BRACKET}{END_OF_VIM9_LINE}'
+
+# COMMA_AT_EOL {{{3
+
+const COMMA_AT_EOL: string = $',{END_OF_VIM9_LINE}'
+
+# COMMA_OR_DICT_KEY_AT_EOL {{{3
+
+const COMMA_OR_DICT_KEY_AT_EOL: string = $'\%(,\|{DICT_KEY}\){END_OF_VIM9_LINE}'
+
+# LAMBDA_ARROW_AT_EOL {{{3
+
+const LAMBDA_ARROW_AT_EOL: string = $'\s=>{END_OF_VIM9_LINE}'
+
+# LINE_CONTINUATION_AT_EOL {{{3
+
+const LINE_CONTINUATION_AT_EOL: string = '\%('
+ .. ','
+ .. '\|' .. OPERATOR
+ .. '\|' .. '\s=>'
+ .. '\|' .. '[^=]\zs[[(]'
+ .. '\|' .. DICT_KEY
+ # `{` is ambiguous.
+ # It can be the start of a dictionary or a block.
+ # We only want to match the former.
+ .. '\|' .. $'^\%({STARTS_CURLY_BLOCK}\)\@!.*\zs{{'
+ .. '\)\s*\%(\s#.*\)\=$'
+# }}}2
+# SOL {{{2
+# BACKSLASH_AT_SOL {{{3
+
+const BACKSLASH_AT_SOL: string = '^\s*\%(\\\|[#"]\\ \)'
+
+# CLOSING_BRACKET_AT_SOL {{{3
+
+const CLOSING_BRACKET_AT_SOL: string = $'^\s*{CLOSING_BRACKET}'
+
+# LINE_CONTINUATION_AT_SOL {{{3
+
+const LINE_CONTINUATION_AT_SOL: string = '^\s*\%('
+ .. '\\'
+ .. '\|' .. '[#"]\\ '
+ .. '\|' .. OPERATOR
+ .. '\|' .. '->\s*\h'
+ .. '\|' .. '\.\h' # dict member
+ .. '\|' .. '|'
+ # TODO: `}` at the start of a line is not necessarily a line continuation.
+ # Could be the end of a block.
+ .. '\|' .. CLOSING_BRACKET
+ .. '\)'
+
+# RANGE_AT_SOL {{{3
+
+const RANGE_AT_SOL: string = '^\s*:\S'
+# }}}1
+# Interface {{{1
+export def Expr(lnum: number): number # {{{2
+ # line which is indented
+ var line_A: dict<any> = {text: getline(lnum), lnum: lnum}
+ # line above, on which we'll base the indent of line A
+ var line_B: dict<any>
+
+ if line_A->AtStartOf('HereDoc')
+ line_A->CacheHeredoc()
+ elseif line_A.lnum->IsInside('HereDoc')
+ return line_A.text->HereDocIndent()
+ elseif line_A.lnum->IsRightBelow('HereDoc')
+ var ind: number = b:vimindent.startindent
+ unlet! b:vimindent
+ return ind
+ endif
+
+ # Don't move this block after the function header one.
+ # Otherwise, we might clear the cache too early if the line following the
+ # header is a comment.
+ if line_A.text =~ COMMENT
+ return CommentIndent()
+ endif
+
+ line_B = PrevCodeLine(line_A.lnum)
+ if line_A.text =~ BACKSLASH_AT_SOL
+ if line_B.text =~ BACKSLASH_AT_SOL
+ return Indent(line_B.lnum)
+ else
+ return Indent(line_B.lnum) + IndentMoreLineContinuation()
+ endif
+ endif
+
+ if line_A->AtStartOf('FuncHeader')
+ line_A.lnum->CacheFuncHeader()
+ elseif line_A.lnum->IsInside('FuncHeader')
+ return b:vimindent.startindent + 2 * shiftwidth()
+ elseif line_A.lnum->IsRightBelow('FuncHeader')
+ var startindent: number = b:vimindent.startindent
+ unlet! b:vimindent
+ if line_A.text =~ ENDS_FUNCTION
+ return startindent
+ else
+ return startindent + shiftwidth()
+ endif
+ endif
+
+ var past_bracket_block: dict<any>
+ if exists('b:vimindent')
+ && b:vimindent->has_key('is_BracketBlock')
+ past_bracket_block = RemovePastBracketBlock(line_A)
+ endif
+ if line_A->AtStartOf('BracketBlock')
+ line_A->CacheBracketBlock()
+ endif
+ if line_A.lnum->IsInside('BracketBlock')
+ && !b:vimindent.block_stack[0].is_curly_block
+ for block: dict<any> in b:vimindent.block_stack
+ # Can't call `BracketBlockIndent()` before we're indenting a line *after* the start of the block.{{{
+ #
+ # That's because it might need the correct indentation of the start
+ # of the block. But if we're still *on* the start, we haven't yet
+ # computed that indentation.
+ #}}}
+ if line_A.lnum > block.startlnum
+ && !block.is_curly_block
+ return BracketBlockIndent(line_A, block)
+ endif
+ endfor
+ endif
+ if line_A.text->ContinuesBelowBracketBlock(line_B, past_bracket_block)
+ && line_A.text !~ CLOSING_BRACKET_AT_SOL
+ return past_bracket_block.startindent
+ endif
+
+ # Problem: If we press `==` on the line right below the start of a multiline
+ # lambda (split after its arrow `=>`), the indent is not correct.
+ # Solution: Indent relative to the line above.
+ if line_B->EndsWithLambdaArrow()
+ return Indent(line_B.lnum) + shiftwidth() + IndentMoreInBracketBlock()
+ endif
+
+ # Don't move this block before the heredoc one.{{{
+ #
+ # A heredoc might be assigned on the very first line.
+ # And if it is, we need to cache some info.
+ #}}}
+ # Don't move it before the function header and bracket block ones either.{{{
+ #
+ # You could, because these blocks of code deal with construct which can only
+ # appear in a Vim9 script. And in a Vim9 script, the first line is
+ # `vim9script`. Or maybe some legacy code/comment (see `:help vim9-mix`).
+ # But you can't find a Vim9 function header or Vim9 bracket block on the
+ # first line.
+ #
+ # Anyway, even if you could, don't. First, it would be inconsistent.
+ # Second, it could give unexpected results while we're trying to fix some
+ # failing test.
+ #}}}
+ if line_A.lnum == 1
+ return 0
+ endif
+
+ # Don't do that:
+ # if line_A.text !~ '\S'
+ # return -1
+ # endif
+ # It would prevent a line from being automatically indented when using the
+ # normal command `o`.
+ # TODO: Can we write a test for this?
+
+ if line_B.text =~ STARTS_CURLY_BLOCK
+ return Indent(line_B.lnum) + shiftwidth() + IndentMoreInBracketBlock()
+
+ elseif line_A.text =~ CLOSING_BRACKET_AT_SOL
+ var start: number = MatchingOpenBracket(line_A)
+ if start <= 0
+ return -1
+ endif
+ return Indent(start) + IndentMoreInBracketBlock()
+
+ elseif line_A.text =~ ENDS_BLOCK_OR_CLAUSE
+ && !line_B->EndsWithLineContinuation()
+ var kwd: string = BlockStartKeyword(line_A.text)
+ if !START_MIDDLE_END->has_key(kwd)
+ return -1
+ endif
+
+ # If the cursor is after the match for the end pattern, we won't find
+ # the start of the block. Let's make sure that doesn't happen.
+ cursor(line_A.lnum, 1)
+
+ var [start: string, middle: string, end: string] = START_MIDDLE_END[kwd]
+ var block_start = SearchPairStart(start, middle, end)
+ if block_start > 0
+ return Indent(block_start)
+ else
+ return -1
+ endif
+ endif
+
+ var base_ind: number
+ if line_A->IsFirstLineOfCommand(line_B)
+ line_A.isfirst = true
+ line_B = line_B->FirstLinePreviousCommand()
+ base_ind = Indent(line_B.lnum)
+
+ if line_B->EndsWithCurlyBlock()
+ && !line_A->IsInThisBlock(line_B.lnum)
+ return base_ind
+ endif
+
+ else
+ line_A.isfirst = false
+ base_ind = Indent(line_B.lnum)
+
+ var line_C: dict<any> = PrevCodeLine(line_B.lnum)
+ if !line_B->IsFirstLineOfCommand(line_C) || line_C.lnum <= 0
+ return base_ind
+ endif
+ endif
+
+ var ind: number = base_ind + Offset(line_A, line_B)
+ return [ind, 0]->max()
+enddef
+
+def g:GetVimIndent(): number # {{{2
+ # for backward compatibility
+ return Expr(v:lnum)
+enddef
+# }}}1
+# Core {{{1
+def Offset( # {{{2
+ # we indent this line ...
+ line_A: dict<any>,
+ # ... relatively to this line
+ line_B: dict<any>,
+ ): number
+
+ # increase indentation inside a block
+ if line_B.text =~ STARTS_NAMED_BLOCK || line_B->EndsWithCurlyBlock()
+ # But don't indent if the line starting the block also closes it.
+ if line_B->AlsoClosesBlock()
+ return 0
+ # Indent twice for a line continuation in the block header itself, so that
+ # we can easily distinguish the end of the block header from the start of
+ # the block body.
+ elseif line_B->EndsWithLineContinuation()
+ && !line_A.isfirst
+ || line_A.text =~ LINE_CONTINUATION_AT_SOL
+ && line_A.text !~ PLUS_MINUS_COMMAND
+ || line_A.text->Is_IN_KeywordForLoop(line_B.text)
+ return 2 * shiftwidth()
+ else
+ return shiftwidth()
+ endif
+
+ # increase indentation of a line if it's the continuation of a command which
+ # started on a previous line
+ elseif !line_A.isfirst
+ && (line_B->EndsWithLineContinuation()
+ || line_A.text =~ LINE_CONTINUATION_AT_SOL)
+ return shiftwidth()
+ endif
+
+ return 0
+enddef
+
+def HereDocIndent(line_A: string): number # {{{2
+ # at the end of a heredoc
+ if line_A =~ $'^\s*{b:vimindent.endmarker}$'
+ # `END` must be at the very start of the line if the heredoc is not trimmed
+ if !b:vimindent.is_trimmed
+ # We can't invalidate the cache just yet.
+ # The indent of `END` is meaningless; it's always 0. The next line
+ # will need to be indented relative to the start of the heredoc. It
+ # must know where it starts; it needs the cache.
+ return 0
+ else
+ var ind: number = b:vimindent.startindent
+ # invalidate the cache so that it's not used for the next heredoc
+ unlet! b:vimindent
+ return ind
+ endif
+ endif
+
+ # In a non-trimmed heredoc, all of leading whitespace is semantic.
+ # Leave it alone.
+ if !b:vimindent.is_trimmed
+ # But do save the indent of the assignment line.
+ if !b:vimindent->has_key('startindent')
+ b:vimindent.startindent = b:vimindent.startlnum->Indent()
+ endif
+ return -1
+ endif
+
+ # In a trimmed heredoc, *some* of the leading whitespace is semantic.
+ # We want to preserve it, so we can't just indent relative to the assignment
+ # line. That's because we're dealing with data, not with code.
+ # Instead, we need to compute by how much the indent of the assignment line
+ # was increased or decreased. Then, we need to apply that same change to
+ # every line inside the body.
+ var offset: number
+ if !b:vimindent->has_key('offset')
+ var old_startindent: number = b:vimindent.startindent
+ var new_startindent: number = b:vimindent.startlnum->Indent()
+ offset = new_startindent - old_startindent
+
+ # If all the non-empty lines in the body have a higher indentation relative
+ # to the assignment, there is no need to indent them more.
+ # But if at least one of them does have the same indentation level (or a
+ # lower one), then we want to indent it further (and the whole block with it).
+ # This way, we can clearly distinguish the heredoc block from the rest of
+ # the code.
+ var end: number = search($'^\s*{b:vimindent.endmarker}$', 'nW')
+ var should_indent_more: bool = range(v:lnum, end - 1)
+ ->indexof((_, lnum: number): bool => Indent(lnum) <= old_startindent && getline(lnum) != '') >= 0
+ if should_indent_more
+ offset += shiftwidth()
+ endif
+
+ b:vimindent.offset = offset
+ b:vimindent.startindent = new_startindent
+ endif
+
+ return [0, Indent(v:lnum) + b:vimindent.offset]->max()
+enddef
+
+def CommentIndent(): number # {{{2
+ var line_B: dict<any>
+ line_B.lnum = prevnonblank(v:lnum - 1)
+ line_B.text = getline(line_B.lnum)
+ if line_B.text =~ COMMENT
+ return Indent(line_B.lnum)
+ endif
+
+ var next: number = NextCodeLine()
+ if next == 0
+ return 0
+ endif
+ var vimindent_save: dict<any> = get(b:, 'vimindent', {})->deepcopy()
+ var ind: number = next->Expr()
+ # The previous `Expr()` might have set or deleted `b:vimindent`.
+ # This could cause issues (e.g. when indenting 2 commented lines above a
+ # heredoc). Let's make sure the state of the variable is not altered.
+ if vimindent_save->empty()
+ unlet! b:vimindent
+ else
+ b:vimindent = vimindent_save
+ endif
+ if getline(next) =~ ENDS_BLOCK
+ return ind + shiftwidth()
+ else
+ return ind
+ endif
+enddef
+
+def BracketBlockIndent(line_A: dict<any>, block: dict<any>): number # {{{2
+ if !block->has_key('startindent')
+ block.startindent = block.startlnum->Indent()
+ endif
+
+ var ind: number = block.startindent
+
+ if line_A.text =~ CLOSING_BRACKET_AT_SOL
+ if b:vimindent.is_on_named_block_line
+ ind += 2 * shiftwidth()
+ endif
+ return ind + IndentMoreInBracketBlock()
+ endif
+
+ var startline: dict<any> = {
+ text: block.startline,
+ lnum: block.startlnum
+ }
+ if startline->EndsWithComma()
+ || startline->EndsWithLambdaArrow()
+ || (startline->EndsWithOpeningBracket()
+ # TODO: Is that reliable?
+ && block.startline !~
+ $'^\s*{NON_BRACKET}\+{LIST_OR_DICT_CLOSING_BRACKET},\s\+{LIST_OR_DICT_OPENING_BRACKET}')
+ ind += shiftwidth() + IndentMoreInBracketBlock()
+ endif
+
+ if b:vimindent.is_on_named_block_line
+ ind += shiftwidth()
+ endif
+
+ if block.is_dict
+ && line_A.text !~ DICT_KEY
+ ind += shiftwidth()
+ endif
+
+ return ind
+enddef
+
+def CacheHeredoc(line_A: dict<any>) # {{{2
+ var endmarker: string = line_A.text->matchstr(ASSIGNS_HEREDOC)
+ var endlnum: number = search($'^\s*{endmarker}$', 'nW')
+ var is_trimmed: bool = line_A.text =~ $'.*\s\%(trim\%(\s\+eval\)\=\)\s\+[A-Z]\+{END_OF_LINE}'
+ b:vimindent = {
+ is_HereDoc: true,
+ startlnum: line_A.lnum,
+ endlnum: endlnum,
+ endmarker: endmarker,
+ is_trimmed: is_trimmed,
+ }
+ if is_trimmed
+ b:vimindent.startindent = Indent(line_A.lnum)
+ endif
+ RegisterCacheInvalidation()
+enddef
+
+def CacheFuncHeader(startlnum: number) # {{{2
+ var pos: list<number> = getcurpos()
+ cursor(startlnum, 1)
+ if search('(', 'W', startlnum) <= 0
+ return
+ endif
+ var endlnum: number = SearchPair('(', '', ')', 'nW')
+ setpos('.', pos)
+ if endlnum == startlnum
+ return
+ endif
+
+ b:vimindent = {
+ is_FuncHeader: true,
+ startindent: startlnum->Indent(),
+ endlnum: endlnum,
+ }
+ RegisterCacheInvalidation()
+enddef
+
+def CacheBracketBlock(line_A: dict<any>) # {{{2
+ var pos: list<number> = getcurpos()
+ var opening: string = line_A.text->matchstr(CHARACTER_UNDER_CURSOR)
+ var closing: string = {'[': ']', '{': '}', '(': ')'}[opening]
+ var endlnum: number = SearchPair(opening, '', closing, 'nW')
+ setpos('.', pos)
+ if endlnum <= line_A.lnum
+ return
+ endif
+
+ if !exists('b:vimindent')
+ b:vimindent = {
+ is_BracketBlock: true,
+ is_on_named_block_line: line_A.text =~ STARTS_NAMED_BLOCK,
+ block_stack: [],
+ }
+ endif
+
+ var is_dict: bool
+ var is_curly_block: bool
+ if opening == '{'
+ if line_A.text =~ STARTS_CURLY_BLOCK
+ [is_dict, is_curly_block] = [false, true]
+ else
+ [is_dict, is_curly_block] = [true, false]
+ endif
+ endif
+ b:vimindent.block_stack->insert({
+ is_dict: is_dict,
+ is_curly_block: is_curly_block,
+ startline: line_A.text,
+ startlnum: line_A.lnum,
+ endlnum: endlnum,
+ })
+
+ RegisterCacheInvalidation()
+enddef
+
+def RegisterCacheInvalidation() # {{{2
+ # invalidate the cache so that it's not used for the next `=` normal command
+ autocmd_add([{
+ cmd: 'unlet! b:vimindent',
+ event: 'ModeChanged',
+ group: '__VimIndent__',
+ once: true,
+ pattern: '*:n',
+ replace: true,
+ }])
+enddef
+
+def RemovePastBracketBlock(line_A: dict<any>): dict<any> # {{{2
+ var stack: list<dict<any>> = b:vimindent.block_stack
+
+ var removed: dict<any>
+ if line_A.lnum > stack[0].endlnum
+ removed = stack[0]
+ endif
+
+ stack->filter((_, block: dict<any>): bool => line_A.lnum <= block.endlnum)
+ if stack->empty()
+ unlet! b:vimindent
+ endif
+ return removed
+enddef
+# }}}1
+# Util {{{1
+# Get {{{2
+def Indent(lnum: number): number # {{{3
+ if lnum <= 0
+ # Don't return `-1`. It could cause `Expr()` to return a non-multiple of `'shiftwidth'`.{{{
+ #
+ # It would be OK if we were always returning `Indent()` directly. But
+ # we don't. Most of the time, we include it in some computation
+ # like `Indent(...) + shiftwidth()`. If `'shiftwidth'` is `4`, and
+ # `Indent()` returns `-1`, `Expr()` will end up returning `3`.
+ #}}}
+ return 0
+ endif
+ return indent(lnum)
+enddef
+
+def BlockStartKeyword(line: string): string # {{{3
+ var kwd: string = line->matchstr('\l\+')
+ return fullcommand(kwd, false)
+enddef
+
+def MatchingOpenBracket(line: dict<any>): number # {{{3
+ var end: string = line.text->matchstr(CLOSING_BRACKET)
+ var start: string = {']': '[', '}': '{', ')': '('}[end]
+ cursor(line.lnum, 1)
+ return SearchPairStart(start, '', end)
+enddef
+
+def FirstLinePreviousCommand(line: dict<any>): dict<any> # {{{3
+ var line_B: dict<any> = line
+
+ while line_B.lnum > 1
+ var code_line_above: dict<any> = PrevCodeLine(line_B.lnum)
+
+ if line_B.text =~ CLOSING_BRACKET_AT_SOL
+ var n: number = MatchingOpenBracket(line_B)
+
+ if n <= 0
+ break
+ endif
+
+ line_B.lnum = n
+ line_B.text = getline(line_B.lnum)
+ continue
+
+ elseif line_B->IsFirstLineOfCommand(code_line_above)
+ break
+ endif
+
+ line_B = code_line_above
+ endwhile
+
+ return line_B
+enddef
+
+def PrevCodeLine(lnum: number): dict<any> # {{{3
+ var line: string = getline(lnum)
+ if line =~ '^\s*[A-Z]\+$'
+ var endmarker: string = line->matchstr('[A-Z]\+')
+ var pos: list<number> = getcurpos()
+ cursor(lnum, 1)
+ var n: number = search(ASSIGNS_HEREDOC, 'bnW')
+ setpos('.', pos)
+ if n > 0
+ line = getline(n)
+ if line =~ $'{HEREDOC_OPERATOR}\s\+{endmarker}'
+ return {lnum: n, text: line}
+ endif
+ endif
+ endif
+
+ var n: number = prevnonblank(lnum - 1)
+ line = getline(n)
+ while line =~ COMMENT && n > 1
+ n = prevnonblank(n - 1)
+ line = getline(n)
+ endwhile
+ # If we get back to the first line, we return 1 no matter what; even if it's a
+ # commented line. That should not cause an issue though. We just want to
+ # avoid a commented line above which there is a line of code which is more
+ # relevant. There is nothing above the first line.
+ return {lnum: n, text: line}
+enddef
+
+def NextCodeLine(): number # {{{3
+ var last: number = line('$')
+ if v:lnum == last
+ return 0
+ endif
+
+ var lnum: number = v:lnum + 1
+ while lnum <= last
+ var line: string = getline(lnum)
+ if line != '' && line !~ COMMENT
+ return lnum
+ endif
+ ++lnum
+ endwhile
+ return 0
+enddef
+
+def SearchPair( # {{{3
+ start: string,
+ middle: string,
+ end: string,
+ flags: string,
+ stopline = 0,
+ ): number
+
+ var s: string = start
+ var e: string = end
+ if start == '[' || start == ']'
+ s = s->escape('[]')
+ endif
+ if end == '[' || end == ']'
+ e = e->escape('[]')
+ endif
+ return searchpair(s, middle, e, flags, (): bool => InCommentOrString(), stopline, TIMEOUT)
+enddef
+
+def SearchPairStart( # {{{3
+ start: string,
+ middle: string,
+ end: string,
+ ): number
+ return SearchPair(start, middle, end, 'bnW')
+enddef
+
+def SearchPairEnd( # {{{3
+ start: string,
+ middle: string,
+ end: string,
+ stopline = 0,
+ ): number
+ return SearchPair(start, middle, end, 'nW', stopline)
+enddef
+# }}}2
+# Test {{{2
+def AtStartOf(line_A: dict<any>, syntax: string): bool # {{{3
+ if syntax == 'BracketBlock'
+ return AtStartOfBracketBlock(line_A)
+ endif
+
+ var pat: string = {
+ HereDoc: ASSIGNS_HEREDOC,
+ FuncHeader: STARTS_FUNCTION
+ }[syntax]
+ return line_A.text =~ pat
+ && (!exists('b:vimindent') || !b:vimindent->has_key('is_HereDoc'))
+enddef
+
+def AtStartOfBracketBlock(line_A: dict<any>): bool # {{{3
+ # We ignore bracket blocks while we're indenting a function header
+ # because it makes the logic simpler. It might mean that we don't
+ # indent correctly a multiline bracket block inside a function header,
+ # but that's a corner case for which it doesn't seem worth making the
+ # code more complex.
+ if exists('b:vimindent')
+ && !b:vimindent->has_key('is_BracketBlock')
+ return false
+ endif
+
+ var pos: list<number> = getcurpos()
+ cursor(line_A.lnum, [line_A.lnum, '$']->col())
+
+ if SearchPair(OPENING_BRACKET, '', CLOSING_BRACKET, 'bcW', line_A.lnum) <= 0
+ setpos('.', pos)
+ return false
+ endif
+ # Don't restore the cursor position.
+ # It needs to be on a bracket for `CacheBracketBlock()` to work as intended.
+
+ return line_A->EndsWithOpeningBracket()
+ || line_A->EndsWithCommaOrDictKey()
+ || line_A->EndsWithLambdaArrow()
+enddef
+
+def ContinuesBelowBracketBlock( # {{{3
+ line_A: string,
+ line_B: dict<any>,
+ block: dict<any>
+ ): bool
+
+ return !block->empty()
+ && (line_A =~ LINE_CONTINUATION_AT_SOL
+ || line_B->EndsWithLineContinuation())
+enddef
+
+def IsInside(lnum: number, syntax: string): bool # {{{3
+ if !exists('b:vimindent')
+ || !b:vimindent->has_key($'is_{syntax}')
+ return false
+ endif
+
+ if syntax == 'BracketBlock'
+ if !b:vimindent->has_key('block_stack')
+ || b:vimindent.block_stack->empty()
+ return false
+ endif
+ return lnum <= b:vimindent.block_stack[0].endlnum
+ endif
+
+ return lnum <= b:vimindent.endlnum
+enddef
+
+def IsRightBelow(lnum: number, syntax: string): bool # {{{3
+ return exists('b:vimindent')
+ && b:vimindent->has_key($'is_{syntax}')
+ && lnum > b:vimindent.endlnum
+enddef
+
+def IsInThisBlock(line_A: dict<any>, lnum: number): bool # {{{3
+ var pos: list<number> = getcurpos()
+ cursor(lnum, [lnum, '$']->col())
+ var end: number = SearchPairEnd('{', '', '}')
+ setpos('.', pos)
+
+ return line_A.lnum <= end
+enddef
+
+def IsFirstLineOfCommand(line_1: dict<any>, line_2: dict<any>): bool # {{{3
+ if line_1.text->Is_IN_KeywordForLoop(line_2.text)
+ return false
+ endif
+
+ if line_1.text =~ RANGE_AT_SOL
+ || line_1.text =~ PLUS_MINUS_COMMAND
+ return true
+ endif
+
+ if line_2.text =~ DICT_KEY
+ && !line_1->IsInThisBlock(line_2.lnum)
+ return true
+ endif
+
+ var line_1_is_good: bool = line_1.text !~ COMMENT
+ && line_1.text !~ DICT_KEY
+ && line_1.text !~ LINE_CONTINUATION_AT_SOL
+
+ var line_2_is_good: bool = !line_2->EndsWithLineContinuation()
+
+ return line_1_is_good && line_2_is_good
+enddef
+
+def Is_IN_KeywordForLoop(line_1: string, line_2: string): bool # {{{3
+ return line_2 =~ '^\s*for\s'
+ && line_1 =~ '^\s*in\s'
+enddef
+
+def InCommentOrString(): bool # {{{3
+ for synID: number in synstack('.', col('.'))
+ if synIDattr(synID, 'name') =~ '\ccomment\|string\|heredoc'
+ return true
+ endif
+ endfor
+
+ return false
+enddef
+
+def AlsoClosesBlock(line_B: dict<any>): bool # {{{3
+ # We know that `line_B` opens a block.
+ # Let's see if it also closes that block.
+ var kwd: string = BlockStartKeyword(line_B.text)
+ if !START_MIDDLE_END->has_key(kwd)
+ return false
+ endif
+
+ var [start: string, middle: string, end: string] = START_MIDDLE_END[kwd]
+ var pos: list<number> = getcurpos()
+ cursor(line_B.lnum, 1)
+ var block_end: number = SearchPairEnd(start, middle, end, line_B.lnum)
+ setpos('.', pos)
+
+ return block_end > 0
+enddef
+
+def EndsWithComma(line: dict<any>): bool # {{{3
+ return NonCommentedMatch(line, COMMA_AT_EOL)
+enddef
+
+def EndsWithCommaOrDictKey(line_A: dict<any>): bool # {{{3
+ return NonCommentedMatch(line_A, COMMA_OR_DICT_KEY_AT_EOL)
+enddef
+
+def EndsWithCurlyBlock(line_B: dict<any>): bool # {{{3
+ return NonCommentedMatch(line_B, STARTS_CURLY_BLOCK)
+enddef
+
+def EndsWithLambdaArrow(line_A: dict<any>): bool # {{{3
+ return NonCommentedMatch(line_A, LAMBDA_ARROW_AT_EOL)
+enddef
+
+def EndsWithLineContinuation(line_B: dict<any>): bool # {{{3
+ return NonCommentedMatch(line_B, LINE_CONTINUATION_AT_EOL)
+enddef
+
+def EndsWithOpeningBracket(line: dict<any>): bool # {{{3
+ return NonCommentedMatch(line, OPENING_BRACKET_AT_EOL)
+enddef
+
+def NonCommentedMatch(line: dict<any>, pat: string): bool # {{{3
+ # Could happen if there is no code above us, and we're not on the 1st line.
+ # In that case, `PrevCodeLine()` returns `{lnum: 0, line: ''}`.
+ if line.lnum == 0
+ return