%PDF- %PDF-
Direktori : /usr/share/vim/vim80/indent/ |
Current File : //usr/share/vim/vim80/indent/erlang.vim |
" Vim indent file " Language: Erlang (http://www.erlang.org) " Author: Csaba Hoch <csaba.hoch@gmail.com> " Contributors: Edwin Fine <efine145_nospam01 at usa dot net> " Pawel 'kTT' Salata <rockplayer.pl@gmail.com> " Ricardo Catalinas Jiménez <jimenezrick@gmail.com> " Last Update: 2013-Jul-21 " License: Vim license " URL: https://github.com/hcs42/vim-erlang " Note About Usage: " This indentation script works best with the Erlang syntax file created by " Kreąimir Marľić (Kresimir Marzic) and maintained by Csaba Hoch. " Notes About Implementation: " " - LTI = Line to indent. " - The index of the first line is 1, but the index of the first column is 0. " Initialization {{{1 " ============== " Only load this indent file when no other was loaded " Vim 7 or later is needed if exists("b:did_indent") || version < 700 finish else let b:did_indent = 1 endif setlocal indentexpr=ErlangIndent() setlocal indentkeys+=0=end,0=of,0=catch,0=after,0=when,0=),0=],0=},0=>> " Only define the functions once if exists("*ErlangIndent") finish endif let s:cpo_save = &cpo set cpo&vim " Logging library {{{1 " =============== " Purpose: " Logs the given string using the ErlangIndentLog function if it exists. " Parameters: " s: string function! s:Log(s) if exists("*ErlangIndentLog") call ErlangIndentLog(a:s) endif endfunction " Line tokenizer library {{{1 " ====================== " Indtokens are "indentation tokens". " Purpose: " Calculate the new virtual column after the given segment of a line. " Parameters: " line: string " first_index: integer -- the index of the first character of the segment " last_index: integer -- the index of the last character of the segment " vcol: integer -- the virtual column of the first character of the token " tabstop: integer -- the value of the 'tabstop' option to be used " Returns: " vcol: integer " Example: " " index: 0 12 34567 " " vcol: 0 45 89 " s:CalcVCol("\t'\tx', b", 1, 4, 4) -> 10 function! s:CalcVCol(line, first_index, last_index, vcol, tabstop) " We copy the relevent segment of the line, otherwise if the line were " e.g. `"\t", term` then the else branch below would consume the `", term` " part at once. let line = a:line[a:first_index : a:last_index] let i = 0 let last_index = a:last_index - a:first_index let vcol = a:vcol while 0 <= i && i <= last_index if line[i] ==# "\t" " Example (when tabstop == 4): " " vcol + tab -> next_vcol " 0 + tab -> 4 " 1 + tab -> 4 " 2 + tab -> 4 " 3 + tab -> 4 " 4 + tab -> 8 " " next_i - i == the number of tabs let next_i = matchend(line, '\t*', i + 1) let vcol = (vcol / a:tabstop + (next_i - i)) * a:tabstop call s:Log('new vcol after tab: '. vcol) else let next_i = matchend(line, '[^\t]*', i + 1) let vcol += next_i - i call s:Log('new vcol after other: '. vcol) endif let i = next_i endwhile return vcol endfunction " Purpose: " Go through the whole line and return the tokens in the line. " Parameters: " line: string -- the line to be examined " string_continuation: bool " atom_continuation: bool " Returns: " indtokens = [indtoken] " indtoken = [token, vcol, col] " token = string (examples: 'begin', '<variable>', '}') " vcol = integer (the virtual column of the first character of the token) " col = integer function! s:GetTokensFromLine(line, string_continuation, atom_continuation, \tabstop) let linelen = strlen(a:line) " The length of the line let i = 0 " The index of the current character in the line let vcol = 0 " The virtual column of the current character let indtokens = [] if a:string_continuation let i = matchend(a:line, '^\%([^"\\]\|\\.\)*"', 0) if i ==# -1 call s:Log(' Whole line is string continuation -> ignore') return [] else let vcol = s:CalcVCol(a:line, 0, i - 1, 0, a:tabstop) call add(indtokens, ['<string_end>', vcol, i]) endif elseif a:atom_continuation let i = matchend(a:line, "^\\%([^'\\\\]\\|\\\\.\\)*'", 0) if i ==# -1 call s:Log(' Whole line is quoted atom continuation -> ignore') return [] else let vcol = s:CalcVCol(a:line, 0, i - 1, 0, a:tabstop) call add(indtokens, ['<quoted_atom_end>', vcol, i]) endif endif while 0 <= i && i < linelen let next_vcol = '' " Spaces if a:line[i] ==# ' ' let next_i = matchend(a:line, ' *', i + 1) " Tabs elseif a:line[i] ==# "\t" let next_i = matchend(a:line, '\t*', i + 1) " See example in s:CalcVCol let next_vcol = (vcol / a:tabstop + (next_i - i)) * a:tabstop " Comment elseif a:line[i] ==# '%' let next_i = linelen " String token: "..." elseif a:line[i] ==# '"' let next_i = matchend(a:line, '\%([^"\\]\|\\.\)*"', i + 1) if next_i ==# -1 call add(indtokens, ['<string_start>', vcol, i]) else let next_vcol = s:CalcVCol(a:line, i, next_i - 1, vcol, a:tabstop) call add(indtokens, ['<string>', vcol, i]) endif " Quoted atom token: '...' elseif a:line[i] ==# "'" let next_i = matchend(a:line, "\\%([^'\\\\]\\|\\\\.\\)*'", i + 1) if next_i ==# -1 call add(indtokens, ['<quoted_atom_start>', vcol, i]) else let next_vcol = s:CalcVCol(a:line, i, next_i - 1, vcol, a:tabstop) call add(indtokens, ['<quoted_atom>', vcol, i]) endif " Keyword or atom or variable token or number elseif a:line[i] =~# '[a-zA-Z_@0-9]' let next_i = matchend(a:line, \'[[:alnum:]_@:]*\%(\s*#\s*[[:alnum:]_@:]*\)\=', \i + 1) call add(indtokens, [a:line[(i):(next_i - 1)], vcol, i]) " Character token: $<char> (as in: $a) elseif a:line[i] ==# '$' call add(indtokens, ['$.', vcol, i]) let next_i = i + 2 " Dot token: . elseif a:line[i] ==# '.' let next_i = i + 1 if i + 1 ==# linelen || a:line[i + 1] =~# '[[:blank:]%]' " End of clause token: . (as in: f() -> ok.) call add(indtokens, ['<end_of_clause>', vcol, i]) else " Possibilities: " - Dot token in float: . (as in: 3.14) " - Dot token in record: . (as in: #myrec.myfield) call add(indtokens, ['.', vcol, i]) endif " Equal sign elseif a:line[i] ==# '=' " This is handled separately so that "=<<" will be parsed as " ['=', '<<'] instead of ['=<', '<']. Although Erlang parses it " currently in the latter way, that may be fixed some day. call add(indtokens, [a:line[i], vcol, i]) let next_i = i + 1 " Three-character tokens elseif i + 1 < linelen && \ index(['=:=', '=/='], a:line[i : i + 1]) != -1 call add(indtokens, [a:line[i : i + 1], vcol, i]) let next_i = i + 2 " Two-character tokens elseif i + 1 < linelen && \ index(['->', '<<', '>>', '||', '==', '/=', '=<', '>=', '++', '--', \ '::'], \ a:line[i : i + 1]) != -1 call add(indtokens, [a:line[i : i + 1], vcol, i]) let next_i = i + 2 " Other character: , ; < > ( ) [ ] { } # + - * / : ? = ! | else call add(indtokens, [a:line[i], vcol, i]) let next_i = i + 1 endif if next_vcol ==# '' let vcol += next_i - i else let vcol = next_vcol endif let i = next_i endwhile return indtokens endfunction " TODO: doc, handle "not found" case function! s:GetIndtokenAtCol(indtokens, col) let i = 0 while i < len(a:indtokens) if a:indtokens[i][2] ==# a:col return [1, i] elseif a:indtokens[i][2] > a:col return [0, s:IndentError('No token at col ' . a:col . ', ' . \'indtokens = ' . string(a:indtokens), \'', '')] endif let i += 1 endwhile return [0, s:IndentError('No token at col ' . a:col . ', ' . \'indtokens = ' . string(a:indtokens), \'', '')] endfunction " Stack library {{{1 " ============= " Purpose: " Push a token onto the parser's stack. " Parameters: " stack: [token] " token: string function! s:Push(stack, token) call s:Log(' Stack Push: "' . a:token . '" into ' . string(a:stack)) call insert(a:stack, a:token) endfunction " Purpose: " Pop a token from the parser's stack. " Parameters: " stack: [token] " token: string " Returns: " token: string -- the removed element function! s:Pop(stack) let head = remove(a:stack, 0) call s:Log(' Stack Pop: "' . head . '" from ' . string(a:stack)) return head endfunction " Library for accessing and storing tokenized lines {{{1 " ================================================= " The Erlang token cache: an `lnum -> indtokens` dictionary that stores the " tokenized lines. let s:all_tokens = {} let s:file_name = '' let s:last_changedtick = -1 " Purpose: " Clear the Erlang token cache if we have a different file or the file has " been changed since the last indentation. function! s:ClearTokenCacheIfNeeded() let file_name = expand('%:p') if file_name != s:file_name || \ b:changedtick != s:last_changedtick let s:file_name = file_name let s:last_changedtick = b:changedtick let s:all_tokens = {} endif endfunction " Purpose: " Return the tokens of line `lnum`, if that line is not empty. If it is " empty, find the first non-empty line in the given `direction` and return " the tokens of that line. " Parameters: " lnum: integer " direction: 'up' | 'down' " Returns: " result: [] -- the result is an empty list if we hit the beginning or end " of the file " | [lnum, indtokens] " lnum: integer -- the index of the non-empty line that was found and " tokenized " indtokens: [indtoken] -- the tokens of line `lnum` function! s:TokenizeLine(lnum, direction) call s:Log('Tokenizing starts from line ' . a:lnum) if a:direction ==# 'up' let lnum = prevnonblank(a:lnum) else " a:direction ==# 'down' let lnum = nextnonblank(a:lnum) endif " We hit the beginning or end of the file if lnum ==# 0 let indtokens = [] call s:Log(' We hit the beginning or end of the file.') " The line has already been parsed elseif has_key(s:all_tokens, lnum) let indtokens = s:all_tokens[lnum] call s:Log('Cached line ' . lnum . ': ' . getline(lnum)) call s:Log(" Tokens in the line:\n - " . join(indtokens, "\n - ")) " The line should be parsed now else " Parse the line let line = getline(lnum) let string_continuation = s:IsLineStringContinuation(lnum) let atom_continuation = s:IsLineAtomContinuation(lnum) let indtokens = s:GetTokensFromLine(line, string_continuation, \atom_continuation, &tabstop) let s:all_tokens[lnum] = indtokens call s:Log('Tokenizing line ' . lnum . ': ' . line) call s:Log(" Tokens in the line:\n - " . join(indtokens, "\n - ")) endif return [lnum, indtokens] endfunction " Purpose: " As a helper function for PrevIndToken and NextIndToken, the FindIndToken " function finds the first line with at least one token in the given " direction. " Parameters: " lnum: integer " direction: 'up' | 'down' " Returns: " result: [] -- the result is an empty list if we hit the beginning or end " of the file " | indtoken function! s:FindIndToken(lnum, dir) let lnum = a:lnum while 1 let lnum += (a:dir ==# 'up' ? -1 : 1) let [lnum, indtokens] = s:TokenizeLine(lnum, a:dir) if lnum ==# 0 " We hit the beginning or end of the file return [] elseif !empty(indtokens) return indtokens[a:dir ==# 'up' ? -1 : 0] endif endwhile endfunction " Purpose: " Find the token that directly precedes the given token. " Parameters: " lnum: integer -- the line of the given token " i: the index of the given token within line `lnum` " Returns: " result = [] -- the result is an empty list if the given token is the first " token of the file " | indtoken function! s:PrevIndToken(lnum, i) call s:Log(' PrevIndToken called: lnum=' . a:lnum . ', i =' . a:i) " If the current line has a previous token, return that if a:i > 0 return s:all_tokens[a:lnum][a:i - 1] else return s:FindIndToken(a:lnum, 'up') endif endfunction " Purpose: " Find the token that directly succeeds the given token. " Parameters: " lnum: integer -- the line of the given token " i: the index of the given token within line `lnum` " Returns: " result = [] -- the result is an empty list if the given token is the last " token of the file " | indtoken function! s:NextIndToken(lnum, i) call s:Log(' NextIndToken called: lnum=' . a:lnum . ', i =' . a:i) " If the current line has a next token, return that if len(s:all_tokens[a:lnum]) > a:i + 1 return s:all_tokens[a:lnum][a:i + 1] else return s:FindIndToken(a:lnum, 'down') endif endfunction " ErlangCalcIndent helper functions {{{1 " ================================= " Purpose: " This function is called when the parser encounters a syntax error. " " If we encounter a syntax error, we return " g:erlang_unexpected_token_indent, which is -1 by default. This means that " the indentation of the LTI will not be changed. " Parameter: " msg: string " token: string " stack: [token] " Returns: " indent: integer function! s:IndentError(msg, token, stack) call s:Log('Indent error: ' . a:msg . ' -> return') call s:Log(' Token = ' . a:token . ', ' . \' stack = ' . string(a:stack)) return g:erlang_unexpected_token_indent endfunction " Purpose: " This function is called when the parser encounters an unexpected token, " and the parser will return the number given back by UnexpectedToken. " " If we encounter an unexpected token, we return " g:erlang_unexpected_token_indent, which is -1 by default. This means that " the indentation of the LTI will not be changed. " Parameter: " token: string " stack: [token] " Returns: " indent: integer function! s:UnexpectedToken(token, stack) call s:Log(' Unexpected token ' . a:token . ', stack = ' . \string(a:stack) . ' -> return') return g:erlang_unexpected_token_indent endfunction if !exists('g:erlang_unexpected_token_indent') let g:erlang_unexpected_token_indent = -1 endif " Purpose: " Return whether the given line starts with a string continuation. " Parameter: " lnum: integer " Returns: " result: bool " Example: " f() -> % IsLineStringContinuation = false " "This is a % IsLineStringContinuation = false " multiline % IsLineStringContinuation = true " string". % IsLineStringContinuation = true function! s:IsLineStringContinuation(lnum) if has('syntax_items') return synIDattr(synID(a:lnum, 1, 0), 'name') =~# '^erlangString' else return 0 endif endfunction " Purpose: " Return whether the given line starts with an atom continuation. " Parameter: " lnum: integer " Returns: " result: bool " Example: " 'function with % IsLineAtomContinuation = true, but should be false " weird name'() -> % IsLineAtomContinuation = true " ok. % IsLineAtomContinuation = false function! s:IsLineAtomContinuation(lnum) if has('syntax_items') return synIDattr(synID(a:lnum, 1, 0), 'name') =~# '^erlangQuotedAtom' else return 0 endif endfunction " Purpose: " Return whether the 'catch' token (which should be the `i`th token in line " `lnum`) is standalone or part of a try-catch block, based on the preceding " token. " Parameters: " lnum: integer " i: integer " Return: " is_standalone: bool function! s:IsCatchStandalone(lnum, i) call s:Log(' IsCatchStandalone called: lnum=' . a:lnum . ', i=' . a:i) let prev_indtoken = s:PrevIndToken(a:lnum, a:i) " If we hit the beginning of the file, it is not a catch in a try block if prev_indtoken == [] return 1 endif let prev_token = prev_indtoken[0] if prev_token =~# '[A-Z_@0-9]' let is_standalone = 0 elseif prev_token =~# '[a-z]' if index(['after', 'and', 'andalso', 'band', 'begin', 'bnot', 'bor', 'bsl', \ 'bsr', 'bxor', 'case', 'catch', 'div', 'not', 'or', 'orelse', \ 'rem', 'try', 'xor'], prev_token) != -1 " If catch is after these keywords, it is standalone let is_standalone = 1 else " If catch is after another keyword (e.g. 'end') or an atom, it is " part of try-catch. " " Keywords: " - may precede 'catch': end " - may not precede 'catch': fun if of receive when " - unused: cond let query let is_standalone = 0 endif elseif index([')', ']', '}', '<string>', '<string_end>', '<quoted_atom>', \ '<quoted_atom_end>', '$.'], prev_token) != -1 let is_standalone = 0 else " This 'else' branch includes the following tokens: " -> == /= =< < >= > =:= =/= + - * / ++ -- :: < > ; ( [ { ? = ! . | let is_standalone = 1 endif call s:Log(' "catch" preceded by "' . prev_token . '" -> catch ' . \(is_standalone ? 'is standalone' : 'belongs to try-catch')) return is_standalone endfunction " Purpose: " This function is called when a begin-type element ('begin', 'case', " '[', '<<', etc.) is found. It asks the caller to return if the stack " Parameters: " stack: [token] " token: string " curr_vcol: integer " stored_vcol: integer " sw: integer -- number of spaces to be used after the begin element as " indentation " Returns: " result: [should_return, indent] " should_return: bool -- if true, the caller should return `indent` to Vim " indent -- integer function! s:BeginElementFoundIfEmpty(stack, token, curr_vcol, stored_vcol, sw) if empty(a:stack) if a:stored_vcol ==# -1 call s:Log(' "' . a:token . '" directly preceeds LTI -> return') return [1, a:curr_vcol + a:sw] else call s:Log(' "' . a:token . \'" token (whose expression includes LTI) found -> return') return [1, a:stored_vcol] endif else return [0, 0] endif endfunction " Purpose: " This function is called when a begin-type element ('begin', 'case', '[', " '<<', etc.) is found, and in some cases when 'after' and 'when' is found. " It asks the caller to return if the stack is already empty. " Parameters: " stack: [token] " token: string " curr_vcol: integer " stored_vcol: integer " end_token: end token that belongs to the begin element found (e.g. if the " begin element is 'begin', the end token is 'end') " sw: integer -- number of spaces to be used after the begin element as " indentation " Returns: " result: [should_return, indent] " should_return: bool -- if true, the caller should return `indent` to Vim " indent -- integer function! s:BeginElementFound(stack, token, curr_vcol, stored_vcol, end_token, sw) " Return 'return' if the stack is empty let [ret, res] = s:BeginElementFoundIfEmpty(a:stack, a:token, a:curr_vcol, \a:stored_vcol, a:sw) if ret | return [ret, res] | endif if a:stack[0] ==# a:end_token call s:Log(' "' . a:token . '" pops "' . a:end_token . '"') call s:Pop(a:stack) if !empty(a:stack) && a:stack[0] ==# 'align_to_begin_element' call s:Pop(a:stack) if empty(a:stack) return [1, a:curr_vcol] else return [1, s:UnexpectedToken(a:token, a:stack)] endif else return [0, 0] endif else return [1, s:UnexpectedToken(a:token, a:stack)] endif endfunction " Purpose: " This function is called when we hit the beginning of a file or an " end-of-clause token -- i.e. when we found the beginning of the current " clause. " " If the stack contains an '->' or 'when', this means that we can return " now, since we were looking for the beginning of the clause. " Parameters: " stack: [token] " token: string " stored_vcol: integer " Returns: " result: [should_return, indent] " should_return: bool -- if true, the caller should return `indent` to Vim " indent -- integer function! s:BeginningOfClauseFound(stack, token, stored_vcol) if !empty(a:stack) && a:stack[0] ==# 'when' call s:Log(' BeginningOfClauseFound: "when" found in stack') call s:Pop(a:stack) if empty(a:stack) call s:Log(' Stack is ["when"], so LTI is in a guard -> return') return [1, a:stored_vcol + shiftwidth() + 2] else return [1, s:UnexpectedToken(a:token, a:stack)] endif elseif !empty(a:stack) && a:stack[0] ==# '->' call s:Log(' BeginningOfClauseFound: "->" found in stack') call s:Pop(a:stack) if empty(a:stack) call s:Log(' Stack is ["->"], so LTI is in function body -> return') return [1, a:stored_vcol + shiftwidth()] elseif a:stack[0] ==# ';' call s:Pop(a:stack) if empty(a:stack) call s:Log(' Stack is ["->", ";"], so LTI is in a function head ' . \'-> return') return [0, a:stored_vcol] else return [1, s:UnexpectedToken(a:token, a:stack)] endif else return [1, s:UnexpectedToken(a:token, a:stack)] endif else return [0, 0] endif endfunction let g:erlang_indent_searchpair_timeout = 2000 " TODO function! s:SearchPair(lnum, curr_col, start, middle, end) call cursor(a:lnum, a:curr_col + 1) let [lnum_new, col1_new] = \searchpairpos(a:start, a:middle, a:end, 'bW', \'synIDattr(synID(line("."), col("."), 0), "name") ' . \'=~? "string\\|quotedatom\\|todo\\|comment\\|' . \'erlangmodifier"', \0, g:erlang_indent_searchpair_timeout) return [lnum_new, col1_new - 1] endfunction function! s:SearchEndPair(lnum, curr_col) return s:SearchPair( \ a:lnum, a:curr_col, \ '\C\<\%(case\|try\|begin\|receive\|if\)\>\|' . \ '\<fun\>\%(\s\|\n\|%.*$\)*(', \ '', \ '\<end\>') endfunction " ErlangCalcIndent {{{1 " ================ " Purpose: " Calculate the indentation of the given line. " Parameters: " lnum: integer -- index of the line for which the indentation should be " calculated " stack: [token] -- initial stack " Return: " indent: integer -- if -1, that means "don't change the indentation"; " otherwise it means "indent the line with `indent` " number of spaces or equivalent tabs" function! s:ErlangCalcIndent(lnum, stack) let res = s:ErlangCalcIndent2(a:lnum, a:stack) call s:Log("ErlangCalcIndent returned: " . res) return res endfunction function! s:ErlangCalcIndent2(lnum, stack) let lnum = a:lnum let stored_vcol = -1 " Virtual column of the first character of the token that " we currently think we might align to. let mode = 'normal' let stack = a:stack let semicolon_abscol = '' " Walk through the lines of the buffer backwards (starting from the " previous line) until we can decide how to indent the current line. while 1 let [lnum, indtokens] = s:TokenizeLine(lnum, 'up') " Hit the start of the file if lnum ==# 0 let [ret, res] = s:BeginningOfClauseFound(stack, 'beginning_of_file', \stored_vcol) if ret | return res | endif return 0 endif let i = len(indtokens) - 1 let last_token_of_line = 1 while i >= 0 let [token, curr_vcol, curr_col] = indtokens[i] call s:Log(' Analyzing the following token: ' . string(indtokens[i])) if len(stack) > 256 " TODO: magic number return s:IndentError('Stack too long', token, stack) endif if token ==# '<end_of_clause>' let [ret, res] = s:BeginningOfClauseFound(stack, token, stored_vcol) if ret | return res | endif if stored_vcol ==# -1 call s:Log(' End of clause directly preceeds LTI -> return') return 0 else call s:Log(' End of clause (but not end of line) -> return') return stored_vcol endif elseif stack == ['prev_term_plus'] if token =~# '[a-zA-Z_@]' || \ token ==# '<string>' || token ==# '<string_start>' || \ token ==# '<quoted_atom>' || token ==# '<quoted_atom_start>' call s:Log(' previous token found: curr_vcol + plus = ' . \curr_vcol . " + " . plus) return curr_vcol + plus endif elseif token ==# 'begin' let [ret, res] = s:BeginElementFound(stack, token, curr_vcol, \stored_vcol, 'end', shiftwidth()) if ret | return res | endif " case EXPR of BRANCHES end " try EXPR catch BRANCHES end " try EXPR after BODY end " try EXPR catch BRANCHES after BODY end " try EXPR of BRANCHES catch BRANCHES end " try EXPR of BRANCHES after BODY end " try EXPR of BRANCHES catch BRANCHES after BODY end " receive BRANCHES end " receive BRANCHES after BRANCHES end " This branch is not Emacs-compatible elseif (index(['of', 'receive', 'after', 'if'], token) != -1 || \ (token ==# 'catch' && !s:IsCatchStandalone(lnum, i))) && \ !last_token_of_line && \ (empty(stack) || stack ==# ['when'] || stack ==# ['->'] || \ stack ==# ['->', ';']) " If we are after of/receive, but these are not the last " tokens of the line, we want to indent like this: " " % stack == [] " receive stored_vcol, " LTI " " % stack == ['->', ';'] " receive stored_vcol -> " B; " LTI " " % stack == ['->'] " receive stored_vcol -> " LTI " " % stack == ['when'] " receive stored_vcol when " LTI " stack = [] => LTI is a condition " stack = ['->'] => LTI is a branch " stack = ['->', ';'] => LTI is a condition " stack = ['when'] => LTI is a guard if empty(stack) || stack == ['->', ';'] call s:Log(' LTI is in a condition after ' . \'"of/receive/after/if/catch" -> return') return stored_vcol elseif stack == ['->'] call s:Log(' LTI is in a branch after ' . \'"of/receive/after/if/catch" -> return') return stored_vcol + shiftwidth() elseif stack == ['when'] call s:Log(' LTI is in a guard after ' . \'"of/receive/after/if/catch" -> return') return stored_vcol + shiftwidth() else return s:UnexpectedToken(token, stack) endif elseif index(['case', 'if', 'try', 'receive'], token) != -1 " stack = [] => LTI is a condition " stack = ['->'] => LTI is a branch " stack = ['->', ';'] => LTI is a condition " stack = ['when'] => LTI is in a guard if empty(stack) " pass elseif (token ==# 'case' && stack[0] ==# 'of') || \ (token ==# 'if') || \ (token ==# 'try' && (stack[0] ==# 'of' || \ stack[0] ==# 'catch' || \ stack[0] ==# 'after')) || \ (token ==# 'receive') " From the indentation point of view, the keyword " (of/catch/after/end) before the LTI is what counts, so " when we reached these tokens, and the stack already had " a catch/after/end, we didn't modify it. " " This way when we reach case/try/receive (i.e. now), " there is at most one of/catch/after/end token in the " stack. if token ==# 'case' || token ==# 'try' || \ (token ==# 'receive' && stack[0] ==# 'after') call s:Pop(stack) endif if empty(stack) call s:Log(' LTI is in a condition; matching ' . \'"case/if/try/receive" found') let stored_vcol = curr_vcol + shiftwidth() elseif stack[0] ==# 'align_to_begin_element' call s:Pop(stack) let stored_vcol = curr_vcol elseif len(stack) > 1 && stack[0] ==# '->' && stack[1] ==# ';' call s:Log(' LTI is in a condition; matching ' . \'"case/if/try/receive" found') call s:Pop(stack) call s:Pop(stack) let stored_vcol = curr_vcol + shiftwidth() elseif stack[0] ==# '->' call s:Log(' LTI is in a branch; matching ' . \'"case/if/try/receive" found') call s:Pop(stack) let stored_vcol = curr_vcol + 2 * shiftwidth() elseif stack[0] ==# 'when' call s:Log(' LTI is in a guard; matching ' . \'"case/if/try/receive" found') call s:Pop(stack) let stored_vcol = curr_vcol + 2 * shiftwidth() + 2 endif endif let [ret, res] = s:BeginElementFound(stack, token, curr_vcol, \stored_vcol, 'end', shiftwidth()) if ret | return res | endif elseif token ==# 'fun' let next_indtoken = s:NextIndToken(lnum, i) call s:Log(' Next indtoken = ' . string(next_indtoken)) if !empty(next_indtoken) && next_indtoken[0] ==# '(' " We have an anonymous function definition " (e.g. "fun () -> ok end") " stack = [] => LTI is a condition " stack = ['->'] => LTI is a branch " stack = ['->', ';'] => LTI is a condition " stack = ['when'] => LTI is in a guard if empty(stack) call s:Log(' LTI is in a condition; matching "fun" found') let stored_vcol = curr_vcol + shiftwidth() elseif len(stack) > 1 && stack[0] ==# '->' && stack[1] ==# ';' call s:Log(' LTI is in a condition; matching "fun" found') call s:Pop(stack) call s:Pop(stack) elseif stack[0] ==# '->' call s:Log(' LTI is in a branch; matching "fun" found') call s:Pop(stack) let stored_vcol = curr_vcol + 2 * shiftwidth() elseif stack[0] ==# 'when' call s:Log(' LTI is in a guard; matching "fun" found') call s:Pop(stack) let stored_vcol = curr_vcol + 2 * shiftwidth() + 2 endif let [ret, res] = s:BeginElementFound(stack, token, curr_vcol, \stored_vcol, 'end', shiftwidth()) if ret | return res | endif else " Pass: we have a function reference (e.g. "fun f/0") endif elseif token ==# '[' " Emacs compatibility let [ret, res] = s:BeginElementFound(stack, token, curr_vcol, \stored_vcol, ']', 1) if ret | return res | endif elseif token ==# '<<' " Emacs compatibility let [ret, res] = s:BeginElementFound(stack, token, curr_vcol, \stored_vcol, '>>', 2) if ret | return res | endif elseif token ==# '(' || token ==# '{' let end_token = (token ==# '(' ? ')' : \token ==# '{' ? '}' : 'error') if empty(stack) " We found the opening paren whose block contains the LTI. let mode = 'inside' elseif stack[0] ==# end_token call s:Log(' "' . token . '" pops "' . end_token . '"') call s:Pop(stack) if !empty(stack) && stack[0] ==# 'align_to_begin_element' " We found the opening paren whose closing paren " starts LTI let mode = 'align_to_begin_element' else " We found the opening pair for a closing paren that " was already in the stack. let mode = 'outside' endif else return s:UnexpectedToken(token, stack) endif if mode ==# 'inside' || mode ==# 'align_to_begin_element' if last_token_of_line && i != 0 " Examples: {{{ " " mode == 'inside': " " my_func( " LTI " " [Variable, { " LTI " " mode == 'align_to_begin_element': " " my_func( " Params " ) % LTI " " [Variable, { " Terms " } % LTI " }}} let stack = ['prev_term_plus'] let plus = (mode ==# 'inside' ? 2 : 1) call s:Log(' "' . token . \'" token found at end of line -> find previous token') elseif mode ==# 'align_to_begin_element' " Examples: {{{ " " mode == 'align_to_begin_element' && !last_token_of_line " " my_func(stored_vcol " ) % LTI " " [Variable, {stored_vcol " } % LTI " " mode == 'align_to_begin_element' && i == 0 " " ( " stored_vcol " ) % LTI " " { " stored_vcol " } % LTI " }}} call s:Log(' "' . token . '" token (whose closing token ' . \'starts LTI) found -> return') return curr_vcol elseif stored_vcol ==# -1 " Examples: {{{ " " mode == 'inside' && stored_vcol == -1 && !last_token_of_line " " my_func( " LTI " [Variable, { " LTI " " mode == 'inside' && stored_vcol == -1 && i == 0 " " ( " LTI " " { " LTI " }}} call s:Log(' "' . token . \'" token (which directly precedes LTI) found -> return') return curr_vcol + 1 else " Examples: {{{ " " mode == 'inside' && stored_vcol != -1 && !last_token_of_line " " my_func(stored_vcol, " LTI " " [Variable, {stored_vcol, " LTI " " mode == 'inside' && stored_vcol != -1 && i == 0 " " (stored_vcol, " LTI " " {stored_vcol, " LTI " }}} call s:Log(' "' . token . \'" token (whose block contains LTI) found -> return') return stored_vcol endif endif elseif index(['end', ')', ']', '}', '>>'], token) != -1 " If we can be sure that there is synchronization in the Erlang " syntax, we use searchpair to make the script quicker. Otherwise we " just push the token onto the stack and keep parsing. " No synchronization -> no searchpair optimization if !exists('b:erlang_syntax_synced') call s:Push(stack, token) " We don't have searchpair optimization for '>>' elseif token ==# '>>' call s:Push(stack, token) elseif token ==# 'end' let [lnum_new, col_new] = s:SearchEndPair(lnum, curr_col) if lnum_new ==# 0 return s:IndentError('Matching token for "end" not found', \token, stack) else if lnum_new != lnum call s:Log(' Tokenize for "end" <<<<') let [lnum, indtokens] = s:TokenizeLine(lnum_new, 'up') call s:Log(' >>>> Tokenize for "end"') endif let [success, i] = s:GetIndtokenAtCol(indtokens, col_new) if !success | return i | endif let [token, curr_vcol, curr_col] = indtokens[i] call s:Log(' Match for "end" in line ' . lnum_new . ': ' . \string(indtokens[i])) endif else " token is one of the following: ')', ']', '}' call s:Push(stack, token) " We have to escape '[', because this string will be interpreted as a " regexp let open_paren = (token ==# ')' ? '(' : \token ==# ']' ? '\[' : \ '{') let [lnum_new, col_new] = s:SearchPair(lnum, curr_col, \open_paren, '', token) if lnum_new ==# 0 return s:IndentError('Matching token not found', \token, stack) else if lnum_new != lnum call s:Log(' Tokenize the opening paren <<<<') let [lnum, indtokens] = s:TokenizeLine(lnum_new, 'up') call s:Log(' >>>>') endif let [success, i] = s:GetIndtokenAtCol(indtokens, col_new) if !success | return i | endif let [token, curr_vcol, curr_col] = indtokens[i] call s:Log(' Match in line ' . lnum_new . ': ' . \string(indtokens[i])) " Go back to the beginning of the loop and handle the opening paren continue endif endif elseif token ==# ';' if empty(stack) call s:Push(stack, ';') elseif index([';', '->', 'when', 'end', 'after', 'catch'], \stack[0]) != -1 " Pass: " " - If the stack top is another ';', then one ';' is " enough. " - If the stack top is an '->' or a 'when', then we " should keep that, because they signify the type of the " LTI (branch, condition or guard). " - From the indentation point of view, the keyword " (of/catch/after/end) before the LTI is what counts, so " if the stack already has a catch/after/end, we don't " modify it. This way when we reach case/try/receive, " there will be at most one of/catch/after/end token in " the stack. else return s:UnexpectedToken(token, stack) endif elseif token ==# '->' if empty(stack) && !last_token_of_line call s:Log(' LTI is in expression after arrow -> return') return stored_vcol elseif empty(stack) || stack[0] ==# ';' || stack[0] ==# 'end' " stack = [';'] -> LTI is either a branch or in a guard " stack = ['->'] -> LTI is a condition " stack = ['->', ';'] -> LTI is a branch call s:Push(stack, '->') elseif index(['->', 'when', 'end', 'after', 'catch'], stack[0]) != -1 " Pass: " " - If the stack top is another '->', then one '->' is " enough. " - If the stack top is a 'when', then we should keep " that, because this signifies that LTI is a in a guard. " - From the indentation point of view, the keyword " (of/catch/after/end) before the LTI is what counts, so " if the stack already has a catch/after/end, we don't " modify it. This way when we reach case/try/receive, " there will be at most one of/catch/after/end token in " the stack. else return s:UnexpectedToken(token, stack) endif elseif token ==# 'when' " Pop all ';' from the top of the stack while !empty(stack) && stack[0] ==# ';' call s:Pop(stack) endwhile if empty(stack) if semicolon_abscol != '' let stored_vcol = semicolon_abscol endif if !last_token_of_line " Example: " when A, " LTI let [ret, res] = s:BeginElementFoundIfEmpty(stack, token, curr_vcol, \stored_vcol, shiftwidth()) if ret | return res | endif else " Example: " when " LTI call s:Push(stack, token) endif elseif index(['->', 'when', 'end', 'after', 'catch'], stack[0]) != -1 " Pass: " - If the stack top is another 'when', then one 'when' is " enough. " - If the stack top is an '->' or a 'when', then we " should keep that, because they signify the type of the " LTI (branch, condition or guard). " - From the indentation point of view, the keyword " (of/catch/after/end) before the LTI is what counts, so " if the stack already has a catch/after/end, we don't " modify it. This way when we reach case/try/receive, " there will be at most one of/catch/after/end token in " the stack. else return s:UnexpectedToken(token, stack) endif elseif token ==# 'of' || token ==# 'after' || \ (token ==# 'catch' && !s:IsCatchStandalone(lnum, i)) if token ==# 'after' " If LTI is between an 'after' and the corresponding " 'end', then let's return let [ret, res] = s:BeginElementFoundIfEmpty(stack, token, curr_vcol, \stored_vcol, shiftwidth()) if ret | return res | endif endif if empty(stack) || stack[0] ==# '->' || stack[0] ==# 'when' call s:Push(stack, token) elseif stack[0] ==# 'catch' || stack[0] ==# 'after' || stack[0] ==# 'end' " Pass: From the indentation point of view, the keyword " (of/catch/after/end) before the LTI is what counts, so " if the stack already has a catch/after/end, we don't " modify it. This way when we reach case/try/receive, " there will be at most one of/catch/after/end token in " the stack. else return s:UnexpectedToken(token, stack) endif elseif token ==# '||' && empty(stack) && !last_token_of_line call s:Log(' LTI is in expression after "||" -> return') return stored_vcol else call s:Log(' Misc token, stack unchanged = ' . string(stack)) endif if empty(stack) || stack[0] ==# '->' || stack[0] ==# 'when' let stored_vcol = curr_vcol let semicolon_abscol = '' call s:Log(' Misc token when the stack is empty or has "->" ' . \'-> setting stored_vcol to ' . stored_vcol) elseif stack[0] ==# ';' let semicolon_abscol = curr_vcol call s:Log(' Setting semicolon-stored_vcol to ' . stored_vcol) endif let i -= 1 call s:Log(' Token processed. stored_vcol=' . stored_vcol) let last_token_of_line = 0 endwhile " iteration on tokens in a line call s:Log(' Line analyzed. stored_vcol=' . stored_vcol) if empty(stack) && stored_vcol != -1 && \ (!empty(indtokens) && indtokens[0][0] != '<string_end>' && \ indtokens[0][0] != '<quoted_atom_end>') call s:Log(' Empty stack at the beginning of the line -> return') return stored_vcol endif let lnum -= 1 endwhile " iteration on lines endfunction " ErlangIndent function {{{1 " ===================== function! ErlangIndent() call s:ClearTokenCacheIfNeeded() let currline = getline(v:lnum) call s:Log('Indenting line ' . v:lnum . ': ' . currline) if s:IsLineStringContinuation(v:lnum) || s:IsLineAtomContinuation(v:lnum) call s:Log('String or atom continuation found -> ' . \'leaving indentation unchanged') return -1 endif let ml = matchlist(currline, \'^\(\s*\)\(\%(end\|of\|catch\|after\)\>\|[)\]}]\|>>\)') " If the line has a special beginning, but not a standalone catch if !empty(ml) && !(ml[2] ==# 'catch' && s:IsCatchStandalone(v:lnum, 0)) let curr_col = len(ml[1]) " If we can be sure that there is synchronization in the Erlang " syntax, we use searchpair to make the script quicker. if ml[2] ==# 'end' && exists('b:erlang_syntax_synced') let [lnum, col] = s:SearchEndPair(v:lnum, curr_col) if lnum ==# 0 return s:IndentError('Matching token for "end" not found', \'end', []) else call s:Log(' Tokenize for "end" <<<<') let [lnum, indtokens] = s:TokenizeLine(lnum, 'up') call s:Log(' >>>> Tokenize for "end"') let [success, i] = s:GetIndtokenAtCol(indtokens, col) if !success | return i | endif let [token, curr_vcol, curr_col] = indtokens[i] call s:Log(' Match for "end" in line ' . lnum . ': ' . \string(indtokens[i])) return curr_vcol endif else call s:Log(" Line type = 'end'") let new_col = s:ErlangCalcIndent(v:lnum - 1, \[ml[2], 'align_to_begin_element']) endif else call s:Log(" Line type = 'normal'") let new_col = s:ErlangCalcIndent(v:lnum - 1, []) if currline =~# '^\s*when\>' let new_col += 2 endif endif if new_col < -1 call s:Log('WARNING: returning new_col == ' . new_col) return g:erlang_unexpected_token_indent endif return new_col endfunction " Cleanup {{{1 " ======= let &cpo = s:cpo_save unlet s:cpo_save " vim: sw=2 et fdm=marker