mirror of
https://github.com/preservim/nerdcommenter.git
synced 2025-02-27 14:57:41 +08:00

I can't find any record of this being a legitimate filetype. It isn't in vim or nvim's filetype detection system and nothing comes up in search. If it's legit somebody can complain and we'll add it back, but it seems likely this was left over from a fat-fingered mistake during the initial development of the plugin. Fixes #363
3075 lines
114 KiB
VimL
3075 lines
114 KiB
VimL
" ============================================================================
|
|
" File: NERD_commenter.vim
|
|
" Description: vim global plugin that provides easy code commenting
|
|
" Author: Martin Grenfell <martin.grenfell at gmail dot com>
|
|
" Maintainer: Caleb Maclennan <caleb@alerque.com>
|
|
" Version: 2.5.1
|
|
" Last Change: Tue Nov 7 10:45:53 +03 2017
|
|
" License: This program is free software. It comes without any warranty,
|
|
" to the extent permitted by applicable law. You can redistribute
|
|
" it and/or modify it under the terms of the Do What The Fuck You
|
|
" Want To Public License, Version 2, as published by Sam Hocevar.
|
|
" See http://sam.zoy.org/wtfpl/COPYING for more details.
|
|
"
|
|
" ============================================================================
|
|
|
|
" Section: script init stuff {{{1
|
|
if exists("loaded_nerd_comments")
|
|
finish
|
|
endif
|
|
if v:version < 700
|
|
echoerr "NERDCommenter: this plugin requires vim >= 7. DOWNLOAD IT! You'll thank me later!"
|
|
finish
|
|
endif
|
|
let loaded_nerd_comments = 1
|
|
|
|
" Function: s:InitVariable() function {{{2
|
|
" This function is used to initialise a given variable to a given value. The
|
|
" variable is only initialised if it does not exist prior
|
|
"
|
|
" Args:
|
|
" -var: the name of the var to be initialised
|
|
" -value: the value to initialise var to
|
|
"
|
|
" Returns:
|
|
" 1 if the var is set, 0 otherwise
|
|
function s:InitVariable(var, value)
|
|
if !exists(a:var)
|
|
execute 'let ' . a:var . ' = ' . "'" . a:value . "'"
|
|
return 1
|
|
endif
|
|
return 0
|
|
endfunction
|
|
|
|
" Section: space string init{{{2
|
|
" When putting spaces after the left delimiter and before the right we use
|
|
" s:spaceStr for the space char. This way we can make it add anything after
|
|
" the left and before the right by modifying this variable
|
|
let s:spaceStr = ' '
|
|
let s:lenSpaceStr = strlen(s:spaceStr)
|
|
|
|
" Section: variable initialization {{{2
|
|
call s:InitVariable("g:NERDAllowAnyVisualDelims", 1)
|
|
call s:InitVariable("g:NERDBlockComIgnoreEmpty", 0)
|
|
call s:InitVariable("g:NERDCommentWholeLinesInVMode", 0)
|
|
call s:InitVariable("g:NERDCommentEmptyLines", 0)
|
|
call s:InitVariable("g:NERDCompactSexyComs", 0)
|
|
call s:InitVariable("g:NERDCreateDefaultMappings", 1)
|
|
call s:InitVariable("g:NERDDefaultNesting", 1)
|
|
call s:InitVariable("g:NERDMenuMode", 3)
|
|
call s:InitVariable("g:NERDLPlace", "[>")
|
|
call s:InitVariable("g:NERDUsePlaceHolders", 1)
|
|
call s:InitVariable("g:NERDRemoveAltComs", 1)
|
|
call s:InitVariable("g:NERDRemoveExtraSpaces", 0)
|
|
call s:InitVariable("g:NERDRPlace", "<]")
|
|
call s:InitVariable("g:NERDSpaceDelims", 0)
|
|
call s:InitVariable("g:NERDDefaultAlign", "none")
|
|
call s:InitVariable("g:NERDTrimTrailingWhitespace", 0)
|
|
call s:InitVariable("g:NERDToggleCheckAllLines", 0)
|
|
|
|
let s:NERDFileNameEscape="[]#*$%'\" ?`!&();<>\\"
|
|
|
|
let s:delimiterMap = {
|
|
\ 'aap': { 'left': '#' },
|
|
\ 'abc': { 'left': '%' },
|
|
\ 'acedb': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'actionscript': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'ada': { 'left': '--', 'leftAlt': '-- ' },
|
|
\ 'ahdl': { 'left': '--' },
|
|
\ 'ahk': { 'left': ';', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'amiga': { 'left': ';' },
|
|
\ 'aml': { 'left': '/*' },
|
|
\ 'ampl': { 'left': '#' },
|
|
\ 'ansible': { 'left': '#' },
|
|
\ 'apache': { 'left': '#' },
|
|
\ 'apachestyle': { 'left': '#' },
|
|
\ 'apdl': { 'left': '!' },
|
|
\ 'applescript': { 'left': '--', 'leftAlt': '(*', 'rightAlt': '*)' },
|
|
\ 'armasm': { 'left': ';' },
|
|
\ 'asciidoc': { 'left': '//' },
|
|
\ 'asm': { 'left': ';', 'leftAlt': '#' },
|
|
\ 'asm68k': { 'left': ';' },
|
|
\ 'asn': { 'left': '--' },
|
|
\ 'asp': { 'left': '%', 'leftAlt': '%*', 'rightAlt': '*%' },
|
|
\ 'aspvbs': { 'left': '''', 'leftAlt': '<!--', 'rightAlt': '-->' },
|
|
\ 'asterisk': { 'left': ';' },
|
|
\ 'asy': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'atlas': { 'left': 'C', 'right': '$' },
|
|
\ 'augeas': { 'left': '(*', 'right': '*)' },
|
|
\ 'autohotkey': { 'left': ';', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'autoit': { 'left': ';' },
|
|
\ 'ave': { 'left': "'" },
|
|
\ 'awk': { 'left': '#' },
|
|
\ 'basic': { 'left': "'", 'leftAlt': 'REM ' },
|
|
\ 'bbx': { 'left': '%' },
|
|
\ 'bc': { 'left': '#' },
|
|
\ 'bib': { 'left': '//' },
|
|
\ 'bindzone': { 'left': ';' },
|
|
\ 'blade': { 'left': '{{--', 'right': '--}}' },
|
|
\ 'bst': { 'left': '%' },
|
|
\ 'btm': { 'left': '::' },
|
|
\ 'c': { 'left': '/*', 'right': '*/', 'leftAlt': '//' },
|
|
\ 'cabal': { 'left': '--' },
|
|
\ 'calibre': { 'left': '//' },
|
|
\ 'caos': { 'left': '*' },
|
|
\ 'catalog': { 'left': '--', 'right': '--' },
|
|
\ 'cf': { 'left': '<!---', 'right': '--->' },
|
|
\ 'cfg': { 'left': '#' },
|
|
\ 'cg': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'ch': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'cl': { 'left': '#' },
|
|
\ 'clean': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'clipper': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'clojure': { 'left': ';' },
|
|
\ 'cmake': { 'left': '#' },
|
|
\ 'coffee': { 'left': '#', 'leftAlt': '###', 'rightAlt': '###' },
|
|
\ 'conkyrc': { 'left': '#' },
|
|
\ 'context': { 'left': '%', 'leftAlt': '--' },
|
|
\ 'coq': { 'left': '(*', 'right': '*)' },
|
|
\ 'cpp': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'crontab': { 'left': '#' },
|
|
\ 'cs': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'csp': { 'left': '--' },
|
|
\ 'cterm': { 'left': '*' },
|
|
\ 'cucumber': { 'left': '#' },
|
|
\ 'cuda': { 'left': '/*', 'right': '*/', 'leftAlt': '//' },
|
|
\ 'cvs': { 'left': 'CVS:' },
|
|
\ 'cython': { 'left': '# ', 'leftAlt': '#' },
|
|
\ 'd': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'dakota': { 'left': '#' },
|
|
\ 'dcl': { 'left': '$!' },
|
|
\ 'debcontrol': { 'left': '#' },
|
|
\ 'debsources': { 'left': '#' },
|
|
\ 'def': { 'left': ';' },
|
|
\ 'desktop': { 'left': '#' },
|
|
\ 'dhcpd': { 'left': '#' },
|
|
\ 'diff': { 'left': '#' },
|
|
\ 'django': { 'left': '{% comment %}', 'right': '{% endcomment %}', 'leftAlt': '{#', 'rightAlt': '#}' },
|
|
\ 'dns': { 'left': ';' },
|
|
\ 'docbk': { 'left': '<!--', 'right': '-->' },
|
|
\ 'dockerfile': { 'left': '#' },
|
|
\ 'dosbatch': { 'left': 'REM ', 'leftAlt': '::' },
|
|
\ 'dosini': { 'left': ';' },
|
|
\ 'dot': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'dracula': { 'left': ';' },
|
|
\ 'dsl': { 'left': ';' },
|
|
\ 'dtml': { 'left': '<dtml-comment>', 'right': '</dtml-comment>' },
|
|
\ 'dylan': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'ebuild': { 'left': '#' },
|
|
\ 'ecd': { 'left': '#' },
|
|
\ 'eclass': { 'left': '#' },
|
|
\ 'eiffel': { 'left': '--' },
|
|
\ 'elf': { 'left': "'" },
|
|
\ 'elixir': { 'left': '#' },
|
|
\ 'elm': { 'left': '--', 'leftAlt': '{--', 'rightAlt': '--}' },
|
|
\ 'elmfilt': { 'left': '#' },
|
|
\ 'ember-script': { 'left': '#' },
|
|
\ 'emblem': { 'left': '/' },
|
|
\ 'erlang': { 'left': '%', 'leftAlt': '%%' },
|
|
\ 'eruby': { 'left': '<%#', 'right': '%>', 'leftAlt': '<!--', 'rightAlt': '-->' },
|
|
\ 'esmtprc': { 'left': '#' },
|
|
\ 'expect': { 'left': '#' },
|
|
\ 'exports': { 'left': '#' },
|
|
\ 'factor': { 'left': '! ', 'leftAlt': '!# ' },
|
|
\ 'fancy': { 'left': '#' },
|
|
\ 'faust': { 'left': '//' },
|
|
\ 'fgl': { 'left': '#' },
|
|
\ 'focexec': { 'left': '-*' },
|
|
\ 'form': { 'left': '*' },
|
|
\ 'fortran': { 'left': '!' },
|
|
\ 'foxpro': { 'left': '*' },
|
|
\ 'fsharp': { 'left': '(*', 'right': '*)', 'leftAlt': '//' },
|
|
\ 'fstab': { 'left': '#' },
|
|
\ 'fvwm': { 'left': '#' },
|
|
\ 'fx': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'gams': { 'left': '*' },
|
|
\ 'gdb': { 'left': '#' },
|
|
\ 'gdmo': { 'left': '--' },
|
|
\ 'geek': { 'left': 'GEEK_COMMENT:' },
|
|
\ 'genshi': { 'left': '<!--', 'right': '-->', 'leftAlt': '{#', 'rightAlt': '#}' },
|
|
\ 'gentoo-conf-d': { 'left': '#' },
|
|
\ 'gentoo-env-d': { 'left': '#' },
|
|
\ 'gentoo-init-d': { 'left': '#' },
|
|
\ 'gentoo-make-conf': { 'left': '#' },
|
|
\ 'gentoo-package-keywords': { 'left': '#' },
|
|
\ 'gentoo-package-mask': { 'left': '#' },
|
|
\ 'gentoo-package-use': { 'left': '#' },
|
|
\ 'gitcommit': { 'left': '#' },
|
|
\ 'gitconfig': { 'left': ';' },
|
|
\ 'gitignore': { 'left': '#' },
|
|
\ 'gitrebase': { 'left': '#' },
|
|
\ 'glsl': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'gnuplot': { 'left': '#' },
|
|
\ 'go': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'groovy': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'gsp': { 'left': '<%--', 'right': '--%>', 'leftAlt': '<!--', 'rightAlt': '-->' },
|
|
\ 'gtkrc': { 'left': '#' },
|
|
\ 'h': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'haml': { 'left': '-#', 'leftAlt': '/' },
|
|
\ 'handlebars': { 'left': '{{!-- ', 'right': ' --}}' },
|
|
\ 'haskell': { 'left': '--', 'nested': 0, 'leftAlt': '{-', 'rightAlt': '-}', 'nestedAlt': 1 },
|
|
\ 'haxe': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'hb': { 'left': '#' },
|
|
\ 'hbs': { 'left': '{{!-- ', 'right': ' --}}' },
|
|
\ 'hercules': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'hive': { 'left': '-- ' },
|
|
\ 'hog': { 'left': '#' },
|
|
\ 'hostsaccess': { 'left': '#' },
|
|
\ 'htmlcheetah': { 'left': '##' },
|
|
\ 'htmldjango': { 'left': '{% comment %}', 'right': '{% endcomment %}', 'leftAlt': '{#', 'rightAlt': '#}' },
|
|
\ 'htmlos': { 'left': '#', 'right': '/#' },
|
|
\ 'hxml': { 'left': '#' },
|
|
\ 'hyphy': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'ia64': { 'left': '#' },
|
|
\ 'icon': { 'left': '#' },
|
|
\ 'idl': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'idlang': { 'left': ';' },
|
|
\ 'idris': { 'leftAlt': '--', 'left': '{-', 'right': '-}' },
|
|
\ 'incar': { 'left': '!' },
|
|
\ 'inform': { 'left': '!' },
|
|
\ 'inittab': { 'left': '#' },
|
|
\ 'ishd': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'iss': { 'left': ';' },
|
|
\ 'ist': { 'left': '%' },
|
|
\ 'jade': { 'left': '//-', 'leftAlt': '//' },
|
|
\ 'java': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'javacc': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'javascript': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'javascript.jquery': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'jess': { 'left': ';' },
|
|
\ 'jgraph': { 'left': '(*', 'right': '*)' },
|
|
\ 'jinja': { 'left': '<!--', 'right': '-->', 'leftAlt': '{#', 'rightAlt': '#}' },
|
|
\ 'jproperties': { 'left': '#' },
|
|
\ 'jsp': { 'left': '<%--', 'right': '--%>' },
|
|
\ 'julia': { 'left': '# ', 'leftAlt': '#=', 'rightAlt': '=#' },
|
|
\ 'kivy': { 'left': '#' },
|
|
\ 'kix': { 'left': ';' },
|
|
\ 'kscript': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'lace': { 'left': '--' },
|
|
\ 'laravel': { 'left': '{{--', 'right': '--}}' },
|
|
\ 'ldif': { 'left': '#' },
|
|
\ 'less': { 'left': '/*', 'right': '*/' },
|
|
\ 'lhaskell': { 'left': '>{-', 'right': '-}', 'leftAlt': '>-- ' },
|
|
\ 'lilo': { 'left': '#' },
|
|
\ 'lilypond': { 'left': '%' },
|
|
\ 'liquid': { 'left': '{% comment %}', 'right': '{% endcomment %}' },
|
|
\ 'lisp': { 'left': ';', 'nested': 1, 'leftAlt': '#|', 'rightAlt': '|#', 'nestedAlt': 1 },
|
|
\ 'llvm': { 'left': ';' },
|
|
\ 'lotos': { 'left': '(*', 'right': '*)' },
|
|
\ 'lout': { 'left': '#' },
|
|
\ 'lpc': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'lprolog': { 'left': '%' },
|
|
\ 'lscript': { 'left': "'" },
|
|
\ 'lss': { 'left': '#' },
|
|
\ 'lua': { 'left': '--', 'leftAlt': '--[[', 'rightAlt': ']]' },
|
|
\ 'lynx': { 'left': '#' },
|
|
\ 'lytex': { 'left': '%' },
|
|
\ 'm4': { 'left': 'dnl ' },
|
|
\ 'mail': { 'left': '> ' },
|
|
\ 'mako': { 'left': '##' },
|
|
\ 'man': { 'left': '."' },
|
|
\ 'map': { 'left': '%' },
|
|
\ 'maple': { 'left': '#' },
|
|
\ 'markdown': { 'left': '<!--', 'right': '-->' },
|
|
\ 'masm': { 'left': ';' },
|
|
\ 'mason': { 'left': '<% #', 'right': '%>' },
|
|
\ 'master': { 'left': '$' },
|
|
\ 'matlab': { 'left': '%', 'leftAlt': '%{', 'rightAlt': '%}' },
|
|
\ 'mel': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'meson': { 'left': '#' },
|
|
\ 'mib': { 'left': '--' },
|
|
\ 'minizinc': { 'left': '% ', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'mips': { 'left': '#' },
|
|
\ 'mirah': {'left': '#' },
|
|
\ 'mkd': { 'left': '<!---', 'right': '-->' },
|
|
\ 'mma': { 'left': '(*', 'right': '*)' },
|
|
\ 'model': { 'left': '$', 'right': '$' },
|
|
\ 'modula2': { 'left': '(*', 'right': '*)' },
|
|
\ 'modula3': { 'left': '(*', 'right': '*)' },
|
|
\ 'molpro': { 'left': '!' },
|
|
\ 'monk': { 'left': ';' },
|
|
\ 'mush': { 'left': '#' },
|
|
\ 'mustache': { 'left': '{{!', 'right': '}}' },
|
|
\ 'nagios': { 'left': ';' },
|
|
\ 'named': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'nasm': { 'left': ';' },
|
|
\ 'nastran': { 'left': '$' },
|
|
\ 'natural': { 'left': '/*' },
|
|
\ 'ncf': { 'left': ';' },
|
|
\ 'newlisp': { 'left': ';' },
|
|
\ 'nginx': { 'left': '#' },
|
|
\ 'nimrod': { 'left': '#' },
|
|
\ 'nroff': { 'left': '\"' },
|
|
\ 'nsis': { 'left': '#' },
|
|
\ 'ntp': { 'left': '#' },
|
|
\ 'objc': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'objcpp': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'objj': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'ocaml': { 'left': '(*', 'right': '*)', 'nested': 1 },
|
|
\ 'occam': { 'left': '--' },
|
|
\ 'octave': { 'left': '%', 'leftAlt': '#' },
|
|
\ 'omlet': { 'left': '(*', 'right': '*)' },
|
|
\ 'omnimark': { 'left': ';' },
|
|
\ 'ooc': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'openroad': { 'left': '//' },
|
|
\ 'opl': { 'left': "REM" },
|
|
\ 'ora': { 'left': '#' },
|
|
\ 'ox': { 'left': '//' },
|
|
\ 'paludis-use-conf': { 'left': '#' },
|
|
\ 'pandoc': { 'left': '<!--', 'right': '-->' },
|
|
\ 'pascal': { 'left': '{', 'right': '}', 'leftAlt': '(*', 'rightAlt': '*)' },
|
|
\ 'patran': { 'left': '$', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'pcap': { 'left': '#' },
|
|
\ 'pccts': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'pdf': { 'left': '%' },
|
|
\ 'perl': { 'left': '#' },
|
|
\ 'pfmain': { 'left': '//' },
|
|
\ 'php': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'pic': { 'left': ';' },
|
|
\ 'pike': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'pilrc': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'pine': { 'left': '#' },
|
|
\ 'plm': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'plsql': { 'left': '-- ', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'po': { 'left': '#' },
|
|
\ 'poscar': { 'left': '!' },
|
|
\ 'postscr': { 'left': '%' },
|
|
\ 'pov': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'povini': { 'left': ';' },
|
|
\ 'ppd': { 'left': '%' },
|
|
\ 'ppwiz': { 'left': ';;' },
|
|
\ 'praat': { 'left': '#' },
|
|
\ 'privoxy': { 'left': '#' },
|
|
\ 'processing': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'prolog': { 'left': '%', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'ps1': { 'left': '#' },
|
|
\ 'psf': { 'left': '#' },
|
|
\ 'ptcap': { 'left': '#' },
|
|
\ 'pug': { 'left': '//-', 'leftAlt': '//' },
|
|
\ 'puppet': { 'left': '#' },
|
|
\ 'pyrex': { 'left': '# ', 'leftAlt': '#' },
|
|
\ 'python': { 'left': '# ', 'leftAlt': '#' },
|
|
\ 'r': { 'left': '#', 'leftAlt': '#''' },
|
|
\ 'racket': { 'left': ';', 'nested': 1, 'leftAlt': '#|', 'rightAlt': '|#', 'nestedAlt': 1 },
|
|
\ 'radiance': { 'left': '#' },
|
|
\ 'ratpoison': { 'left': '#' },
|
|
\ 'rc': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'rebol': { 'left': ';' },
|
|
\ 'registry': { 'left': ';' },
|
|
\ 'remind': { 'left': '#' },
|
|
\ 'renpy': { 'left': '# ' },
|
|
\ 'resolv': { 'left': '#' },
|
|
\ 'rgb': { 'left': '!' },
|
|
\ 'rib': { 'left': '#' },
|
|
\ 'rmd': { 'left': '#' },
|
|
\ 'robot': { 'left': '#' },
|
|
\ 'robots': { 'left': '#' },
|
|
\ 'rspec': { 'left': '#' },
|
|
\ 'ruby': { 'left': '#' },
|
|
\ 'rust': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'sa': { 'left': '--' },
|
|
\ 'samba': { 'left': ';', 'leftAlt': '#' },
|
|
\ 'sass': { 'left': '//', 'leftAlt': '/*' },
|
|
\ 'sather': { 'left': '--' },
|
|
\ 'scala': { 'left': '//', 'nested': 1, 'leftAlt': '/*', 'rightAlt': '*/', 'nestedAlt': 1 },
|
|
\ 'scheme': { 'left': ';', 'nested': 1, 'leftAlt': '#|', 'rightAlt': '|#', 'nestedAlt': 1 },
|
|
\ 'scilab': { 'left': '//' },
|
|
\ 'scons': { 'left': '#' },
|
|
\ 'scsh': { 'left': ';' },
|
|
\ 'scss': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'sdc': { 'left': '#' },
|
|
\ 'sed': { 'left': '#' },
|
|
\ 'sgmldecl': { 'left': '--', 'right': '--' },
|
|
\ 'sgmllnx': { 'left': '<!--', 'right': '-->' },
|
|
\ 'sh': { 'left': '#' },
|
|
\ 'shader_test': { 'left': '#' },
|
|
\ 'sicad': { 'left': '*' },
|
|
\ 'sile': { 'left': '%' },
|
|
\ 'simula': { 'left': '%', 'leftAlt': '--' },
|
|
\ 'sinda': { 'left': '$' },
|
|
\ 'skill': { 'left': ';' },
|
|
\ 'slang': { 'left': '%' },
|
|
\ 'slice': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'slim': { 'left': '/', 'leftAlt': '/!' },
|
|
\ 'slrnrc': { 'left': '%' },
|
|
\ 'sls': { 'left': '#' },
|
|
\ 'sm': { 'left': '#' },
|
|
\ 'smarty': { 'left': '{*', 'right': '*}' },
|
|
\ 'smil': { 'left': '<!', 'right': '>' },
|
|
\ 'smith': { 'left': ';' },
|
|
\ 'sml': { 'left': '(*', 'right': '*)', 'nested': 1 },
|
|
\ 'snippets': { 'left': '#' },
|
|
\ 'snnsnet': { 'left': '#' },
|
|
\ 'snnspat': { 'left': '#' },
|
|
\ 'snnsres': { 'left': '#' },
|
|
\ 'snobol4': { 'left': '*' },
|
|
\ 'spec': { 'left': '#' },
|
|
\ 'specman': { 'left': '//' },
|
|
\ 'spectre': { 'left': '//', 'leftAlt': '*' },
|
|
\ 'spice': { 'left': '$' },
|
|
\ 'spin': { 'left': '''', 'leftAlt': '{', 'rightAlt': '}' },
|
|
\ 'sql': { 'left': '-- ' },
|
|
\ 'sqlforms': { 'left': '-- ' },
|
|
\ 'sqlj': { 'left': '-- ' },
|
|
\ 'sqr': { 'left': '!' },
|
|
\ 'squid': { 'left': '#' },
|
|
\ 'ss': { 'left': ';', 'leftAlt': '#|', 'rightAlt': '|#' },
|
|
\ 'sshconfig': { 'left': '#' },
|
|
\ 'sshdconfig': { 'left': '#' },
|
|
\ 'st': { 'left': '"' },
|
|
\ 'stan': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'stp': { 'left': '/*', 'right': '*/', 'leftAlt': '//' },
|
|
\ 'supercollider': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'swift': { 'left': '/*', 'right': '*/', 'leftAlt': '//' },
|
|
\ 'systemverilog': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'tads': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'tags': { 'left': ';' },
|
|
\ 'tak': { 'left': '$' },
|
|
\ 'tasm': { 'left': ';' },
|
|
\ 'tcl': { 'left': '#' },
|
|
\ 'teak': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'terraform': { 'left': '#', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'tex': { 'left': '%' },
|
|
\ 'texinfo': { 'left': "@c " },
|
|
\ 'texmf': { 'left': '%' },
|
|
\ 'tf': { 'left': '#' },
|
|
\ 'tidy': { 'left': '#' },
|
|
\ 'tli': { 'left': '#' },
|
|
\ 'tmux': { 'left': '#' },
|
|
\ 'toml': { 'left': '#' },
|
|
\ 'trasys': { 'left': "$" },
|
|
\ 'tsalt': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'tsscl': { 'left': '#' },
|
|
\ 'tssgm': { 'left': "comment = '", 'right': "'" },
|
|
\ 'ttl': { 'left': '#' },
|
|
\ 'tup': { 'left': '#' },
|
|
\ 'twig': { 'left': '{#', 'right': '#}' },
|
|
\ 'txt2tags': { 'left': '%' },
|
|
\ 'typescript': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'uc': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'uil': { 'left': '!' },
|
|
\ 'upstart': { 'left': '#' },
|
|
\ 'vala': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'vasp': { 'left': '!' },
|
|
\ 'vb': { 'left': "'" },
|
|
\ 'velocity': { 'left': "##", 'right': "", 'leftAlt': '#*', 'rightAlt': '*#' },
|
|
\ 'vera': { 'left': '/*', 'right': '*/', 'leftAlt': '//' },
|
|
\ 'verilog': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'verilog_systemverilog': { 'left': '//', 'leftAlt': '/*', 'rightAlt': '*/' },
|
|
\ 'vgrindefs': { 'left': '#' },
|
|
\ 'vhdl': { 'left': '--' },
|
|
\ 'vimperator': { 'left': '"' },
|
|
\ 'virata': { 'left': '%' },
|
|
\ 'vrml': { 'left': '#' },
|
|
\ 'vsejcl': { 'left': '/*' },
|
|
\ 'webmacro': { 'left': '##' },
|
|
\ 'wget': { 'left': '#' },
|
|
\ 'wikipedia': { 'left': '<!--', 'right': '-->' },
|
|
\ 'winbatch': { 'left': ';' },
|
|
\ 'wml': { 'left': '#' },
|
|
\ 'wvdial': { 'left': ';' },
|
|
\ 'xdefaults': { 'left': '!' },
|
|
\ 'xkb': { 'left': '//' },
|
|
\ 'xmath': { 'left': '#' },
|
|
\ 'xpm2': { 'left': '!' },
|
|
\ 'xquery': { 'left': '(:', 'right': ':)' },
|
|
\ 'yaml': { 'left': '#' },
|
|
\ 'z8a': { 'left': ';' }
|
|
\ }
|
|
|
|
let g:NERDDelimiterMap = s:delimiterMap
|
|
|
|
if exists("g:NERDCustomDelimiters")
|
|
call extend(s:delimiterMap, g:NERDCustomDelimiters)
|
|
endif
|
|
|
|
" Section: Comment mapping functions, autocommands and commands {{{1
|
|
" ============================================================================
|
|
" Section: Comment enabler autocommands {{{2
|
|
" ============================================================================
|
|
|
|
augroup NERDCommenter
|
|
|
|
"if the user enters a buffer or reads a buffer then we gotta set up
|
|
"the comment delimiters for that new filetype
|
|
autocmd BufEnter,BufRead * :call s:SetUpForNewFiletype(&filetype, 0)
|
|
|
|
"if the filetype of a buffer changes, force the script to reset the
|
|
"delimiters for the buffer
|
|
autocmd Filetype * :call s:SetUpForNewFiletype(&filetype, 1)
|
|
augroup END
|
|
|
|
|
|
" Function: s:SetUpForNewFiletype(filetype) function {{{2
|
|
" This function is responsible for setting up buffer scoped variables for the
|
|
" given filetype.
|
|
"
|
|
" Args:
|
|
" -filetype: the filetype to set delimiters for
|
|
" -forceReset: 1 if the delimiters should be reset if they have already be
|
|
" set for this buffer.
|
|
"
|
|
function s:SetUpForNewFiletype(filetype, forceReset)
|
|
let ft = a:filetype
|
|
|
|
"for compound filetypes, if we don't know how to handle the full filetype
|
|
"then break it down and use the first part that we know how to handle
|
|
if ft =~ '\.' && !has_key(s:delimiterMap, ft)
|
|
let filetypes = split(a:filetype, '\.')
|
|
for i in filetypes
|
|
if has_key(s:delimiterMap, i)
|
|
let ft = i
|
|
break
|
|
endif
|
|
endfor
|
|
endif
|
|
|
|
let b:NERDSexyComMarker = ''
|
|
|
|
if has_key(s:delimiterMap, ft)
|
|
let b:NERDCommenterDelims = s:delimiterMap[ft]
|
|
for i in ['left', 'leftAlt', 'right', 'rightAlt']
|
|
if !has_key(b:NERDCommenterDelims, i)
|
|
let b:NERDCommenterDelims[i] = ''
|
|
endif
|
|
endfor
|
|
for i in ['nested', 'nestedAlt']
|
|
if !has_key(b:NERDCommenterDelims, i)
|
|
let b:NERDCommenterDelims[i] = 0
|
|
endif
|
|
endfor
|
|
" if g:NERD_<filetype>_alt_style is defined, use the alternate style
|
|
let b:NERDCommenterFirstInit = getbufvar(1,"NERDCommenterFirstInit")
|
|
if exists('g:NERDAltDelims_'.ft) && eval('g:NERDAltDelims_'.ft) && !b:NERDCommenterFirstInit
|
|
call s:SwitchToAlternativeDelimiters(0)
|
|
let b:NERDCommenterFirstInit = 1
|
|
endif
|
|
else
|
|
let b:NERDCommenterDelims = s:CreateDelimMapFromCms()
|
|
endif
|
|
|
|
endfunction
|
|
|
|
function s:CreateDelimMapFromCms()
|
|
if &ft == '' && exists('g:NERDDefaultDelims')
|
|
let delims = g:NERDDefaultDelims
|
|
for i in ['left', 'leftAlt', 'right', 'rightAlt']
|
|
if !has_key(delims, i)
|
|
let delims[i] = ''
|
|
endif
|
|
endfor
|
|
return delims
|
|
endif
|
|
return {
|
|
\ 'left': substitute(&commentstring, '\([^ \t]*\)\s*%s.*', '\1', ''),
|
|
\ 'right': substitute(&commentstring, '.*%s\s*\(.*\)', '\1', 'g'),
|
|
\ 'nested': 0,
|
|
\ 'leftAlt': '',
|
|
\ 'rightAlt': '',
|
|
\ 'nestedAlt': 0}
|
|
endfunction
|
|
|
|
" Function: s:SwitchToAlternativeDelimiters(printMsgs) function {{{2
|
|
" This function is used to swap the delimiters that are being used to the
|
|
" alternative delimiters for that filetype. For example, if a c++ file is
|
|
" being edited and // comments are being used, after this function is called
|
|
" /**/ comments will be used.
|
|
"
|
|
" Args:
|
|
" -printMsgs: if this is 1 then a message is echoed to the user telling them
|
|
" if this function changed the delimiters or not
|
|
function s:SwitchToAlternativeDelimiters(printMsgs)
|
|
if exists('*NERDCommenter_before')
|
|
exe "call NERDCommenter_before()"
|
|
endif
|
|
"if both of the alternative delimiters are empty then there is no
|
|
"alternative comment style so bail out
|
|
if b:NERDCommenterDelims['leftAlt'] == '' && b:NERDCommenterDelims['rightAlt'] == ''
|
|
if a:printMsgs
|
|
call s:NerdEcho("Cannot use alternative delimiters, none are specified", 0)
|
|
endif
|
|
return 0
|
|
endif
|
|
|
|
"save the current delimiters
|
|
let tempLeft = s:Left()
|
|
let tempRight = s:Right()
|
|
let tempNested = s:Nested()
|
|
|
|
"swap current delimiters for alternative
|
|
let b:NERDCommenterDelims['left'] = b:NERDCommenterDelims['leftAlt']
|
|
let b:NERDCommenterDelims['right'] = b:NERDCommenterDelims['rightAlt']
|
|
"set information on whether these are nested
|
|
let b:NERDCommenterDelims['nested'] = b:NERDCommenterDelims['nestedAlt']
|
|
|
|
"set the previously current delimiters to be the new alternative ones
|
|
let b:NERDCommenterDelims['leftAlt'] = tempLeft
|
|
let b:NERDCommenterDelims['rightAlt'] = tempRight
|
|
let b:NERDCommenterDelims['nestedAlt'] = tempNested
|
|
|
|
"tell the user what comment delimiters they are now using
|
|
if a:printMsgs
|
|
call s:NerdEcho("Now using " . s:Left() . " " . s:Right() . " to delimit comments", 1)
|
|
endif
|
|
|
|
if exists('*NERDCommenter_after')
|
|
exe "call NERDCommenter_after()"
|
|
endif
|
|
|
|
return 1
|
|
endfunction
|
|
|
|
" Section: Comment delimiter add/removal functions {{{1
|
|
" ============================================================================
|
|
" Function: s:AppendCommentToLine(){{{2
|
|
" This function appends comment delimiters at the EOL and places the cursor in
|
|
" position to start typing the comment
|
|
function s:AppendCommentToLine()
|
|
let left = s:Left({'space': 1})
|
|
let right = s:Right({'space': 1})
|
|
|
|
" get the length of the right delimiter
|
|
let lenRight = strlen(right)
|
|
|
|
let isLineEmpty = strlen(getline(".")) == 0
|
|
let insOrApp = (isLineEmpty==1 ? 'i' : 'A')
|
|
|
|
"stick the delimiters down at the end of the line. We have to format the
|
|
"comment with spaces as appropriate
|
|
execute ":normal! " . insOrApp . (isLineEmpty ? '' : ' ') . left . right
|
|
|
|
" if there is a right delimiter then we gotta move the cursor left
|
|
" by the length of the right delimiter so we insert between the delimiters
|
|
if lenRight > 0
|
|
let leftMoveAmount = lenRight - 1
|
|
execute ":normal! " . leftMoveAmount . "h"
|
|
startinsert
|
|
else
|
|
startinsert!
|
|
endif
|
|
endfunction
|
|
|
|
" Function: s:CommentBlock(top, bottom, lSide, rSide, forceNested ) {{{2
|
|
" This function is used to comment out a region of code. This region is
|
|
" specified as a bounding box by arguments to the function.
|
|
"
|
|
" Args:
|
|
" -top: the line number for the top line of code in the region
|
|
" -bottom: the line number for the bottom line of code in the region
|
|
" -lSide: the column number for the left most column in the region
|
|
" -rSide: the column number for the right most column in the region
|
|
" -forceNested: a flag indicating whether comments should be nested
|
|
function s:CommentBlock(top, bottom, lSide, rSide, forceNested )
|
|
" we need to create local copies of these arguments so we can modify them
|
|
let top = a:top
|
|
let bottom = a:bottom
|
|
let lSide = a:lSide
|
|
let rSide = a:rSide
|
|
|
|
"if the top or bottom line starts with tabs we have to adjust the left and
|
|
"right boundaries so that they are set as though the tabs were spaces
|
|
let topline = getline(top)
|
|
let bottomline = getline(bottom)
|
|
if s:HasLeadingTabs(topline, bottomline)
|
|
|
|
"find out how many tabs are in the top line and adjust the left
|
|
"boundary accordingly
|
|
let numTabs = s:NumberOfLeadingTabs(topline)
|
|
if lSide < numTabs
|
|
let lSide = &ts * lSide
|
|
else
|
|
let lSide = (lSide - numTabs) + (&ts * numTabs)
|
|
endif
|
|
|
|
"find out how many tabs are in the bottom line and adjust the right
|
|
"boundary accordingly
|
|
let numTabs = s:NumberOfLeadingTabs(bottomline)
|
|
let rSide = (rSide - numTabs) + (&ts * numTabs)
|
|
endif
|
|
|
|
"we must check that bottom IS actually below top, if it is not then we
|
|
"swap top and bottom. Similarly for left and right.
|
|
if bottom < top
|
|
let temp = top
|
|
let top = bottom
|
|
let bottom = top
|
|
endif
|
|
if rSide < lSide
|
|
let temp = lSide
|
|
let lSide = rSide
|
|
let rSide = temp
|
|
endif
|
|
|
|
"if the current delimiters aren't multipart then we will switch to the
|
|
"alternative delimiters (if THEY are) as the comment will be better and more
|
|
"accurate with multipart delimiters
|
|
let switchedDelims = 0
|
|
if !s:Multipart() && g:NERDAllowAnyVisualDelims && s:AltMultipart()
|
|
let switchedDelims = 1
|
|
call s:SwitchToAlternativeDelimiters(0)
|
|
endif
|
|
|
|
"start the commenting from the top and keep commenting till we reach the
|
|
"bottom
|
|
let currentLine=top
|
|
while currentLine <= bottom
|
|
|
|
"check if we are allowed to comment this line
|
|
if s:CanCommentLine(a:forceNested, currentLine)
|
|
|
|
"convert the leading tabs into spaces
|
|
let theLine = getline(currentLine)
|
|
let lineHasLeadTabs = s:HasLeadingTabs(theLine)
|
|
if lineHasLeadTabs
|
|
let theLine = s:ConvertLeadingTabsToSpaces(theLine)
|
|
endif
|
|
|
|
"don't comment lines that begin after the right boundary of the
|
|
"block unless the user has specified to do so
|
|
if theLine !~ '^ \{' . rSide . '\}' || !g:NERDBlockComIgnoreEmpty
|
|
|
|
"attempt to place the cursor in on the left of the boundary box,
|
|
"then check if we were successful, if not then we cant comment this
|
|
"line
|
|
call setline(currentLine, theLine)
|
|
if s:CanPlaceCursor(currentLine, lSide)
|
|
|
|
let leftSpaced = s:Left({'space': 1})
|
|
let rightSpaced = s:Right({'space': 1})
|
|
|
|
"stick the left delimiter down
|
|
let theLine = strpart(theLine, 0, lSide-1) . leftSpaced . strpart(theLine, lSide-1)
|
|
|
|
if s:Multipart()
|
|
"stick the right delimiter down
|
|
let theLine = strpart(theLine, 0, rSide+strlen(leftSpaced)) . rightSpaced . strpart(theLine, rSide+strlen(leftSpaced))
|
|
|
|
let firstLeftDelim = s:FindDelimiterIndex(s:Left(), theLine)
|
|
let lastRightDelim = s:LastIndexOfDelim(s:Right(), theLine)
|
|
|
|
if firstLeftDelim != -1 && lastRightDelim != -1
|
|
let searchStr = strpart(theLine, 0, lastRightDelim)
|
|
let searchStr = strpart(searchStr, firstLeftDelim+strlen(s:Left()))
|
|
|
|
"replace the outer most delimiters in searchStr with
|
|
"place-holders
|
|
let theLineWithPlaceHolders = s:ReplaceDelims(s:Left(), s:Right(), g:NERDLPlace, g:NERDRPlace, searchStr)
|
|
|
|
"add the right delimiter onto the line
|
|
let theLine = strpart(theLine, 0, firstLeftDelim+strlen(s:Left())) . theLineWithPlaceHolders . strpart(theLine, lastRightDelim)
|
|
endif
|
|
endif
|
|
endif
|
|
endif
|
|
|
|
"restore tabs if needed
|
|
if lineHasLeadTabs
|
|
let theLine = s:ConvertLeadingSpacesToTabs(theLine)
|
|
endif
|
|
|
|
if g:NERDTrimTrailingWhitespace == 1
|
|
let theLine = s:TrimTrailingWhitespace(theLine)
|
|
endif
|
|
|
|
call setline(currentLine, theLine)
|
|
endif
|
|
|
|
let currentLine = currentLine + 1
|
|
endwhile
|
|
|
|
"if we switched delimiterss then we gotta go back to what they were before
|
|
if switchedDelims == 1
|
|
call s:SwitchToAlternativeDelimiters(0)
|
|
endif
|
|
endfunction
|
|
|
|
" Function: s:CommentLines(forceNested, alignLeft, alignRight, firstLine, lastLine) {{{2
|
|
" This function comments a range of lines.
|
|
"
|
|
" Args:
|
|
" -forceNested: a flag indicating whether the called is requesting the comment
|
|
" to be nested if need be
|
|
" -align: should be "left", "start", "both" or "none"
|
|
" -firstLine/lastLine: the top and bottom lines to comment
|
|
function s:CommentLines(forceNested, align, firstLine, lastLine)
|
|
" we need to get the left and right indexes of the leftmost char in the
|
|
" block of of lines and the right most char so that we can do alignment of
|
|
" the delimiters if the user has specified
|
|
let leftAlignIndx = a:align == "start" ? 0 : s:LeftMostIndx(a:forceNested, 0, a:firstLine, a:lastLine)
|
|
let rightAlignIndx = s:RightMostIndx(a:forceNested, 0, a:firstLine, a:lastLine)
|
|
|
|
" gotta add the length of the left delimiter onto the rightAlignIndx cos
|
|
" we'll be adding a left delimiter to the line
|
|
let rightAlignIndx = rightAlignIndx + strlen(s:Left({'space': 1}))
|
|
|
|
" now we actually comment the lines. Do it line by line
|
|
let currentLine = a:firstLine
|
|
while currentLine <= a:lastLine
|
|
|
|
" get the next line, check commentability and convert spaces to tabs
|
|
let theLine = getline(currentLine)
|
|
let lineHasLeadingTabs = s:HasLeadingTabs(theLine)
|
|
let theLine = s:ConvertLeadingTabsToSpaces(theLine)
|
|
if s:CanCommentLine(a:forceNested, currentLine)
|
|
"if the user has specified forceNesting then we check to see if we
|
|
"need to switch delimiters for place-holders
|
|
if a:forceNested && g:NERDUsePlaceHolders && !s:Nested()
|
|
let theLine = s:SwapOuterMultiPartDelimsForPlaceHolders(theLine)
|
|
endif
|
|
|
|
" find out if the line is commented using normal delimiters and/or
|
|
" alternate ones
|
|
let isCommented = s:IsCommented(s:Left(), s:Right(), theLine) || s:IsCommented(s:Left({'alt': 1}), s:Right({'alt': 1}), theLine)
|
|
|
|
" check if we can comment this line
|
|
if !isCommented || g:NERDUsePlaceHolders || s:Multipart()
|
|
if a:align == "left" || a:align == "start" || a:align == "both"
|
|
let theLine = s:AddLeftDelimAligned(s:Left({'space': 1}), theLine, leftAlignIndx)
|
|
else
|
|
let theLine = s:AddLeftDelim(s:Left({'space': 1}), theLine)
|
|
endif
|
|
if a:align == "both"
|
|
let theLine = s:AddRightDelimAligned(s:Right({'space': 1}), theLine, rightAlignIndx)
|
|
else
|
|
let theLine = s:AddRightDelim(s:Right({'space': 1}), theLine)
|
|
endif
|
|
endif
|
|
endif
|
|
|
|
" restore leading tabs if appropriate
|
|
if lineHasLeadingTabs
|
|
let theLine = s:ConvertLeadingSpacesToTabs(theLine)
|
|
endif
|
|
|
|
if g:NERDTrimTrailingWhitespace == 1
|
|
let theLine = s:TrimTrailingWhitespace(theLine)
|
|
endif
|
|
|
|
" we are done with this line
|
|
call setline(currentLine, theLine)
|
|
let currentLine = currentLine + 1
|
|
endwhile
|
|
|
|
endfunction
|
|
|
|
" Function: s:CommentLinesMinimal(firstLine, lastLine) {{{2
|
|
" This function comments a range of lines in a minimal style. I
|
|
"
|
|
" Args:
|
|
" -firstLine/lastLine: the top and bottom lines to comment
|
|
function s:CommentLinesMinimal(firstLine, lastLine)
|
|
"check that minimal comments can be done on this filetype
|
|
if !s:HasMultipartDelims()
|
|
throw 'NERDCommenter.Delimiters exception: Minimal comments can only be used for filetypes that have multipart delimiters'
|
|
endif
|
|
|
|
let sexyNested = s:SexyNested()
|
|
|
|
"if we need to use place holders for the comment, make sure they are
|
|
"enabled for this filetype, or the delimiterss allow nesting
|
|
if !g:NERDUsePlaceHolders && !sexyNested && s:DoesBlockHaveMultipartDelim(a:firstLine, a:lastLine)
|
|
throw 'NERDCommenter.Settings exception: Place holders are required but disabled.'
|
|
endif
|
|
|
|
"get the left and right delimiters to smack on
|
|
let left = s:GetSexyComLeft(g:NERDSpaceDelims,0)
|
|
let right = s:GetSexyComRight(g:NERDSpaceDelims,0)
|
|
|
|
"make sure all multipart delimiters on the lines are replaced with
|
|
"placeholders to prevent illegal syntax
|
|
if !sexyNested
|
|
let currentLine = a:firstLine
|
|
while(currentLine <= a:lastLine)
|
|
let theLine = getline(currentLine)
|
|
let theLine = s:ReplaceDelims(left, right, g:NERDLPlace, g:NERDRPlace, theLine)
|
|
call setline(currentLine, theLine)
|
|
let currentLine = currentLine + 1
|
|
endwhile
|
|
endif
|
|
|
|
"add the delimiter to the top line
|
|
let theLine = getline(a:firstLine)
|
|
let lineHasLeadingTabs = s:HasLeadingTabs(theLine)
|
|
let theLine = s:ConvertLeadingTabsToSpaces(theLine)
|
|
let theLine = s:AddLeftDelim(left, theLine)
|
|
if lineHasLeadingTabs
|
|
let theLine = s:ConvertLeadingSpacesToTabs(theLine)
|
|
endif
|
|
call setline(a:firstLine, theLine)
|
|
|
|
"add the delimiter to the bottom line
|
|
let theLine = getline(a:lastLine)
|
|
let lineHasLeadingTabs = s:HasLeadingTabs(theLine)
|
|
let theLine = s:ConvertLeadingTabsToSpaces(theLine)
|
|
let theLine = s:AddRightDelim(right, theLine)
|
|
if lineHasLeadingTabs
|
|
let theLine = s:ConvertLeadingSpacesToTabs(theLine)
|
|
endif
|
|
|
|
if g:NERDTrimTrailingWhitespace == 1
|
|
let theLine = s:TrimTrailingWhitespace(theLine)
|
|
endif
|
|
|
|
call setline(a:lastLine, theLine)
|
|
endfunction
|
|
|
|
" Function: s:CommentLinesSexy(topline, bottomline) function {{{2
|
|
" This function is used to comment lines in the 'Sexy' style. E.g., in c:
|
|
" /*
|
|
" * This is a sexy comment
|
|
" */
|
|
" Args:
|
|
" -topline: the line number of the top line in the sexy comment
|
|
" -bottomline: the line number of the bottom line in the sexy comment
|
|
function s:CommentLinesSexy(topline, bottomline)
|
|
let left = s:GetSexyComLeft(0, 0)
|
|
let right = s:GetSexyComRight(0, 0)
|
|
|
|
"check if we can do a sexy comment with the available delimiters
|
|
if left == -1 || right == -1
|
|
throw 'NERDCommenter.Delimiters exception: cannot perform sexy comments with available delimiters.'
|
|
endif
|
|
|
|
"make sure the lines aren't already commented sexually or we can nest
|
|
if !s:CanSexyCommentLines(a:topline, a:bottomline)
|
|
throw 'NERDCommenter.Nesting exception: cannot nest sexy comments'
|
|
endif
|
|
|
|
|
|
let sexyComMarker = s:GetSexyComMarker(0,0)
|
|
let sexyComMarkerSpaced = s:GetSexyComMarker(1,0)
|
|
|
|
|
|
" we jam the comment as far to the right as possible
|
|
let leftAlignIndx = s:LeftMostIndx(1, 1, a:topline, a:bottomline)
|
|
|
|
"check if we should use the compact style i.e that the left/right
|
|
"delimiters should appear on the first and last lines of the code and not
|
|
"on separate lines above/below the first/last lines of code
|
|
if g:NERDCompactSexyComs
|
|
let spaceString = (g:NERDSpaceDelims ? s:spaceStr : '')
|
|
|
|
"comment the top line
|
|
let theLine = getline(a:topline)
|
|
let lineHasTabs = s:HasLeadingTabs(theLine)
|
|
if lineHasTabs
|
|
let theLine = s:ConvertLeadingTabsToSpaces(theLine)
|
|
endif
|
|
if !s:SexyNested()
|
|
let theLine = s:SwapOuterMultiPartDelimsForPlaceHolders(theLine)
|
|
endif
|
|
let theLine = s:AddLeftDelimAligned(left . spaceString, theLine, leftAlignIndx)
|
|
if lineHasTabs
|
|
let theLine = s:ConvertLeadingSpacesToTabs(theLine)
|
|
endif
|
|
call setline(a:topline, theLine)
|
|
|
|
"comment the bottom line
|
|
if a:bottomline != a:topline
|
|
let theLine = getline(a:bottomline)
|
|
let lineHasTabs = s:HasLeadingTabs(theLine)
|
|
if lineHasTabs
|
|
let theLine = s:ConvertLeadingTabsToSpaces(theLine)
|
|
endif
|
|
if !s:SexyNested()
|
|
let theLine = s:SwapOuterMultiPartDelimsForPlaceHolders(theLine)
|
|
endif
|
|
endif
|
|
let theLine = s:AddRightDelim(spaceString . right, theLine)
|
|
if lineHasTabs
|
|
let theLine = s:ConvertLeadingSpacesToTabs(theLine)
|
|
endif
|
|
call setline(a:bottomline, theLine)
|
|
else
|
|
|
|
" add the left delimiter one line above the lines that are to be commented
|
|
call cursor(a:topline, 1)
|
|
execute 'normal! O'
|
|
let theLine = repeat(' ', leftAlignIndx) . left
|
|
|
|
" Make sure tabs are respected
|
|
if !&expandtab
|
|
let theLine = s:ConvertLeadingSpacesToTabs(theLine)
|
|
endif
|
|
call setline(a:topline, theLine)
|
|
|
|
" add the right delimiter after bottom line (we have to add 1 cos we moved
|
|
" the lines down when we added the left delimiter
|
|
call cursor(a:bottomline+1, 1)
|
|
execute 'normal! o'
|
|
let theLine = repeat(' ', leftAlignIndx) . repeat(' ', strlen(left)-strlen(sexyComMarker)) . right
|
|
|
|
" Make sure tabs are respected
|
|
if !&expandtab
|
|
let theLine = s:ConvertLeadingSpacesToTabs(theLine)
|
|
endif
|
|
call setline(a:bottomline+2, theLine)
|
|
|
|
endif
|
|
|
|
" go thru each line adding the sexyComMarker marker to the start of each
|
|
" line in the appropriate place to align them with the comment delimiters
|
|
let currentLine = a:topline+1
|
|
while currentLine <= a:bottomline + !g:NERDCompactSexyComs
|
|
" get the line and convert the tabs to spaces
|
|
let theLine = getline(currentLine)
|
|
let lineHasTabs = s:HasLeadingTabs(theLine)
|
|
if lineHasTabs
|
|
let theLine = s:ConvertLeadingTabsToSpaces(theLine)
|
|
endif
|
|
|
|
if !s:SexyNested()
|
|
let theLine = s:SwapOuterMultiPartDelimsForPlaceHolders(theLine)
|
|
endif
|
|
|
|
" add the sexyComMarker
|
|
let theLine = repeat(' ', leftAlignIndx) . repeat(' ', strlen(left)-strlen(sexyComMarker)) . sexyComMarkerSpaced . strpart(theLine, leftAlignIndx)
|
|
|
|
if lineHasTabs
|
|
let theLine = s:ConvertLeadingSpacesToTabs(theLine)
|
|
endif
|
|
|
|
if g:NERDTrimTrailingWhitespace == 1
|
|
let theLine = s:TrimTrailingWhitespace(theLine)
|
|
endif
|
|
|
|
" set the line and move onto the next one
|
|
call setline(currentLine, theLine)
|
|
let currentLine = currentLine + 1
|
|
endwhile
|
|
|
|
endfunction
|
|
|
|
" Function: s:CommentLinesToggle(forceNested, firstLine, lastLine) {{{2
|
|
" Applies "toggle" commenting to the given range of lines
|
|
"
|
|
" Args:
|
|
" -forceNested: a flag indicating whether the called is requesting the comment
|
|
" to be nested if need be
|
|
" -firstLine/lastLine: the top and bottom lines to comment
|
|
function s:CommentLinesToggle(forceNested, firstLine, lastLine)
|
|
let currentLine = a:firstLine
|
|
|
|
let align = g:NERDDefaultAlign
|
|
let leftAlignIndx = align == "start" ? 0 : s:LeftMostIndx(a:forceNested, 0, a:firstLine, a:lastLine)
|
|
let rightAlignIndx = s:RightMostIndx(a:forceNested, 0, a:firstLine, a:lastLine)
|
|
let rightAlignIndx = rightAlignIndx + strlen(s:Left({'space': 1}))
|
|
|
|
while currentLine <= a:lastLine
|
|
|
|
" get the next line, check commentability and convert spaces to tabs
|
|
let theLine = getline(currentLine)
|
|
let lineHasLeadingTabs = s:HasLeadingTabs(theLine)
|
|
let theLine = s:ConvertLeadingTabsToSpaces(theLine)
|
|
if s:CanToggleCommentLine(a:forceNested, currentLine)
|
|
|
|
"if the user has specified forceNesting then we check to see if we
|
|
"need to switch delimiters for place-holders
|
|
if g:NERDUsePlaceHolders && !s:Nested()
|
|
let theLine = s:SwapOuterMultiPartDelimsForPlaceHolders(theLine)
|
|
endif
|
|
|
|
if align == 'left' || align == 'start' || align == 'both'
|
|
let theLine = s:AddLeftDelimAligned(s:Left({'space': 1}), theLine, leftAlignIndx)
|
|
else
|
|
let theLine = s:AddLeftDelim(s:Left({'space': 1}), theLine)
|
|
endif
|
|
if align == "both"
|
|
let theLine = s:AddRightDelimAligned(s:Right({'space': 1}), theLine, rightAlignIndx)
|
|
else
|
|
let theLine = s:AddRightDelim(s:Right({'space': 1}), theLine)
|
|
endif
|
|
endif
|
|
|
|
" restore leading tabs if appropriate
|
|
if lineHasLeadingTabs
|
|
let theLine = s:ConvertLeadingSpacesToTabs(theLine)
|
|
endif
|
|
|
|
if g:NERDTrimTrailingWhitespace == 1
|
|
let theLine = s:TrimTrailingWhitespace(theLine)
|
|
endif
|
|
|
|
" we are done with this line
|
|
call setline(currentLine, theLine)
|
|
let currentLine = currentLine + 1
|
|
endwhile
|
|
|
|
endfunction
|
|
|
|
" Function: s:CommentRegion(topline, topCol, bottomLine, bottomCol) function {{{2
|
|
" This function comments chunks of text selected in visual mode.
|
|
" It will comment exactly the text that they have selected.
|
|
" Args:
|
|
" -topLine: the line number of the top line in the sexy comment
|
|
" -topCol: top left column for this comment
|
|
" -bottomline: the line number of the bottom line in the sexy comment
|
|
" -bottomCol: the bottom right column for this comment
|
|
" -forceNested: whether the caller wants comments to be nested if the
|
|
" line(s) are already commented
|
|
function s:CommentRegion(topLine, topCol, bottomLine, bottomCol, forceNested)
|
|
|
|
"switch delimiters (if we can) if the current set isn't multipart
|
|
let switchedDelims = 0
|
|
if !s:Multipart() && s:AltMultipart() && !g:NERDAllowAnyVisualDelims
|
|
let switchedDelims = 1
|
|
call s:SwitchToAlternativeDelimiters(0)
|
|
endif
|
|
|
|
"if there is only one line in the comment then just do it
|
|
if a:topLine == a:bottomLine
|
|
call s:CommentBlock(a:topLine, a:bottomLine, a:topCol, a:bottomCol, a:forceNested)
|
|
|
|
"there are multiple lines in the comment
|
|
else
|
|
"comment the top line
|
|
call s:CommentBlock(a:topLine, a:topLine, a:topCol, strlen(getline(a:topLine)), a:forceNested)
|
|
|
|
"comment out all the lines in the middle of the comment
|
|
let topOfRange = a:topLine+1
|
|
let bottomOfRange = a:bottomLine-1
|
|
if topOfRange <= bottomOfRange
|
|
call s:CommentLines(a:forceNested, g:NERDDefaultAlign, topOfRange, bottomOfRange)
|
|
endif
|
|
|
|
"comment the bottom line
|
|
let bottom = getline(a:bottomLine)
|
|
let numLeadingSpacesTabs = strlen(substitute(bottom, '^\([ \t]*\).*$', '\1', ''))
|
|
call s:CommentBlock(a:bottomLine, a:bottomLine, numLeadingSpacesTabs+1, a:bottomCol, a:forceNested)
|
|
|
|
endif
|
|
|
|
"stick the cursor back on the char it was on before the comment
|
|
call cursor(a:topLine, a:topCol + strlen(s:Left()) + g:NERDSpaceDelims)
|
|
|
|
"if we switched delimiters then we gotta go back to what they were before
|
|
if switchedDelims == 1
|
|
call s:SwitchToAlternativeDelimiters(0)
|
|
endif
|
|
|
|
endfunction
|
|
|
|
" Function: s:InvertComment(firstLine, lastLine) function {{{2
|
|
" Inverts the comments on the lines between and including the given line
|
|
" numbers i.e all commented lines are uncommented and vice versa
|
|
" Args:
|
|
" -firstLine: the top of the range of lines to be inverted
|
|
" -lastLine: the bottom of the range of lines to be inverted
|
|
function s:InvertComment(firstLine, lastLine)
|
|
|
|
" go thru all lines in the given range
|
|
let currentLine = a:firstLine
|
|
while currentLine <= a:lastLine
|
|
let theLine = getline(currentLine)
|
|
|
|
let sexyComBounds = s:FindBoundingLinesOfSexyCom(currentLine)
|
|
|
|
" if the line is commented normally, uncomment it
|
|
if s:IsCommentedFromStartOfLine(s:Left(), theLine) || s:IsCommentedFromStartOfLine(s:Left({'alt': 1}), theLine)
|
|
call s:UncommentLines(currentLine, currentLine)
|
|
let currentLine = currentLine + 1
|
|
|
|
" check if the line is commented sexually
|
|
elseif !empty(sexyComBounds)
|
|
let numLinesBeforeSexyComRemoved = s:NumLinesInBuf()
|
|
call s:UncommentLinesSexy(sexyComBounds[0], sexyComBounds[1])
|
|
|
|
"move to the line after last line of the sexy comment
|
|
let numLinesAfterSexyComRemoved = s:NumLinesInBuf()
|
|
let currentLine = sexyComBounds[1] - (numLinesBeforeSexyComRemoved - numLinesAfterSexyComRemoved) + 1
|
|
|
|
" the line isn't commented
|
|
else
|
|
call s:CommentLinesToggle(1, currentLine, currentLine)
|
|
let currentLine = currentLine + 1
|
|
endif
|
|
|
|
endwhile
|
|
endfunction
|
|
|
|
function! NERDCommentIsLineCommented(lineNo)
|
|
let theLine = getline(a:lineNo)
|
|
return s:IsInSexyComment(a:lineNo) || s:IsCommentedFromStartOfLine(s:Left(), theLine) || s:IsCommentedFromStartOfLine(s:Left({'alt': 1}), theLine)
|
|
endfunction
|
|
|
|
" Function: NERDComment(mode, type) function {{{2
|
|
" This function is a Wrapper for the main commenting functions
|
|
"
|
|
" Args:
|
|
" -mode: a character indicating the mode in which the comment is requested:
|
|
" 'n' for Normal mode, 'x' for Visual mode
|
|
" -type: the type of commenting requested. Can be 'Sexy', 'Invert',
|
|
" 'Minimal', 'Toggle', 'AlignLeft', 'AlignBoth', 'Comment',
|
|
" 'Nested', 'ToEOL', 'Append', 'Insert', 'Uncomment', 'Yank'
|
|
function! NERDComment(mode, type) range
|
|
if exists('*NERDCommenter_before')
|
|
exe "call NERDCommenter_before()"
|
|
endif
|
|
|
|
let isVisual = a:mode =~ '[vsx]'
|
|
|
|
if !exists("g:did_load_ftplugin") || g:did_load_ftplugin != 1
|
|
call s:NerdEcho("filetype plugins should be enabled. See :help NERDComInstallation and :help :filetype-plugin-on", 0)
|
|
endif
|
|
|
|
if isVisual
|
|
let firstLine = line("'<")
|
|
let lastLine = line("'>")
|
|
let firstCol = col("'<")
|
|
let lastCol = col("'>") - (&selection == 'exclusive' ? 1 : 0)
|
|
else
|
|
let firstLine = a:firstline
|
|
let lastLine = a:lastline
|
|
endif
|
|
"
|
|
" Save options we need to change so we can recover them later
|
|
let state = s:SetupStateBeforeLineComment(firstLine, lastLine)
|
|
|
|
let countWasGiven = (!isVisual && firstLine != lastLine)
|
|
|
|
let forceNested = (a:type ==? 'Nested' || g:NERDDefaultNesting)
|
|
|
|
if a:type ==? 'Comment' || a:type ==? 'Nested'
|
|
if isVisual && visualmode() == "\<C-V>"
|
|
call s:CommentBlock(firstLine, lastLine, firstCol, lastCol, forceNested)
|
|
elseif isVisual && visualmode() == "v" && (g:NERDCommentWholeLinesInVMode==0 || (g:NERDCommentWholeLinesInVMode==2 && s:HasMultipartDelims()))
|
|
call s:CommentRegion(firstLine, firstCol, lastLine, lastCol, forceNested)
|
|
else
|
|
call s:CommentLines(forceNested, g:NERDDefaultAlign, firstLine, lastLine)
|
|
endif
|
|
|
|
elseif a:type ==? 'AlignLeft' || a:type ==? 'AlignBoth'
|
|
let align = "none"
|
|
if a:type ==? "AlignLeft"
|
|
let align = "left"
|
|
elseif a:type ==? "AlignBoth"
|
|
let align = "both"
|
|
endif
|
|
call s:CommentLines(forceNested, align, firstLine, lastLine)
|
|
|
|
elseif a:type ==? 'Invert'
|
|
call s:InvertComment(firstLine, lastLine)
|
|
|
|
elseif a:type ==? 'Sexy'
|
|
try
|
|
call s:CommentLinesSexy(firstLine, lastLine)
|
|
catch /NERDCommenter.Delimiters/
|
|
call s:CommentLines(forceNested, g:NERDDefaultAlign, firstLine, lastLine)
|
|
catch /NERDCommenter.Nesting/
|
|
call s:NerdEcho("Sexy comment aborted. Nested sexy cannot be nested", 0)
|
|
endtry
|
|
|
|
elseif a:type ==? 'Toggle'
|
|
if g:NERDToggleCheckAllLines ==# 0
|
|
let theLine = getline(firstLine)
|
|
if s:IsInSexyComment(firstLine) || s:IsCommentedFromStartOfLine(s:Left(), theLine) || s:IsCommentedFromStartOfLine(s:Left({'alt': 1}), theLine)
|
|
call s:UncommentLines(firstLine, lastLine)
|
|
else
|
|
call s:CommentLinesToggle(forceNested, firstLine, lastLine)
|
|
endif
|
|
else
|
|
let l:commentAllLines = 0
|
|
for i in range(firstLine, lastLine)
|
|
let theLine = getline(i)
|
|
" if have one line no comment, then comment all lines
|
|
if !s:IsInSexyComment(firstLine) && !s:IsCommentedFromStartOfLine(s:Left(), theLine) && !s:IsCommentedFromStartOfLine(s:Left({'alt': 1}), theLine)
|
|
let l:commentAllLines = 1
|
|
break
|
|
else
|
|
endif
|
|
endfor
|
|
if l:commentAllLines ==# 1
|
|
call s:CommentLinesToggle(forceNested, firstLine, lastLine)
|
|
else
|
|
call s:UncommentLines(firstLine, lastLine)
|
|
endif
|
|
endif
|
|
|
|
elseif a:type ==? 'Minimal'
|
|
try
|
|
call s:CommentLinesMinimal(firstLine, lastLine)
|
|
catch /NERDCommenter.Delimiters/
|
|
call s:NerdEcho("Minimal comments can only be used for filetypes that have multipart delimiters.", 0)
|
|
catch /NERDCommenter.Settings/
|
|
call s:NerdEcho("Place holders are required but disabled.", 0)
|
|
endtry
|
|
|
|
elseif a:type ==? 'ToEOL'
|
|
call s:SaveScreenState()
|
|
call s:CommentBlock(firstLine, firstLine, col("."), col("$")-1, 1)
|
|
call s:RestoreScreenState()
|
|
|
|
elseif a:type ==? 'Append'
|
|
call s:AppendCommentToLine()
|
|
|
|
elseif a:type ==? 'Insert'
|
|
call s:PlaceDelimitersAndInsBetween()
|
|
|
|
elseif a:type ==? 'Uncomment'
|
|
call s:UncommentLines(firstLine, lastLine)
|
|
|
|
elseif a:type ==? 'Yank'
|
|
if isVisual
|
|
normal! gvy
|
|
elseif countWasGiven
|
|
execute firstLine .','. lastLine .'yank'
|
|
else
|
|
normal! yy
|
|
endif
|
|
execute firstLine .','. lastLine .'call NERDComment("'. a:mode .'", "Comment")'
|
|
endif
|
|
|
|
call s:RecoverStateAfterLineComment(state)
|
|
|
|
if isVisual
|
|
let nlines = lastLine - firstLine
|
|
silent! call repeat#set("V" . nlines . "jo" . "\<Plug>NERDCommenter". a:type)
|
|
else
|
|
silent! call repeat#set("\<Plug>NERDCommenter". a:type)
|
|
endif
|
|
|
|
if exists('*NERDCommenter_after')
|
|
exe "call NERDCommenter_after()"
|
|
endif
|
|
|
|
endfunction
|
|
|
|
" Function: s:PlaceDelimitersAndInsBetween() function {{{2
|
|
" This is function is called to place comment delimiters down and place the
|
|
" cursor between them
|
|
function s:PlaceDelimitersAndInsBetween()
|
|
" get the left and right delimiters without any escape chars in them
|
|
let left = s:Left({'space': 1})
|
|
let right = s:Right({'space': 1})
|
|
|
|
let theLine = getline(".")
|
|
let lineHasLeadTabs = s:HasLeadingTabs(theLine) || (theLine =~ '^ *$' && !&expandtab)
|
|
|
|
"convert tabs to spaces and adjust the cursors column to take this into
|
|
"account
|
|
let untabbedCol = s:UntabbedCol(theLine, col("."))
|
|
call setline(line("."), s:ConvertLeadingTabsToSpaces(theLine))
|
|
call cursor(line("."), untabbedCol)
|
|
|
|
" get the length of the right delimiter
|
|
let lenRight = strlen(right)
|
|
|
|
let isDelimOnEOL = col(".") >= strlen(getline("."))
|
|
|
|
" if the cursor is in the first col then we gotta insert rather than
|
|
" append the comment delimiters here
|
|
let insOrApp = (col(".")==1 ? 'i' : 'a')
|
|
|
|
" place the delimiters down. We do it differently depending on whether
|
|
" there is a left AND right delimiter
|
|
if lenRight > 0
|
|
execute ":normal! " . insOrApp . left . right
|
|
execute ":normal! " . lenRight . "h"
|
|
else
|
|
execute ":normal! " . insOrApp . left
|
|
endif
|
|
silent! normal! l
|
|
|
|
"if needed convert spaces back to tabs and adjust the cursors col
|
|
"accordingly
|
|
if lineHasLeadTabs
|
|
let tabbedCol = s:TabbedCol(getline("."), col("."))
|
|
call setline(line("."), s:ConvertLeadingSpacesToTabs(getline(".")))
|
|
call cursor(line("."), tabbedCol)
|
|
endif
|
|
|
|
if isDelimOnEOL && lenRight == 0
|
|
startinsert!
|
|
else
|
|
startinsert
|
|
endif
|
|
endfunction
|
|
|
|
" Function: s:RemoveDelimiters(left, right, line) {{{2
|
|
" this function is called to remove the first left comment delimiter and the
|
|
" last right delimiter of the given line.
|
|
"
|
|
" The arguments left and right must be strings. If there is no right delimiter (as
|
|
" is the case for e.g vim file comments) them the argument right should be ""
|
|
"
|
|
" Args:
|
|
" -left: the left comment delimiter
|
|
" -right: the right comment delimiter
|
|
" -line: the line to remove the delimiters from
|
|
function s:RemoveDelimiters(left, right, line)
|
|
|
|
let l:left = a:left
|
|
let l:right = a:right
|
|
let lenLeft = strlen(left)
|
|
let lenRight = strlen(right)
|
|
|
|
let delimsSpaced = (g:NERDSpaceDelims || g:NERDRemoveExtraSpaces)
|
|
|
|
let line = a:line
|
|
|
|
"look for the left delimiter, if we find it, remove it.
|
|
let leftIndx = s:FindDelimiterIndex(a:left, line)
|
|
if leftIndx != -1
|
|
let line = strpart(line, 0, leftIndx) . strpart(line, leftIndx+lenLeft)
|
|
|
|
"if the user has specified that there is a space after the left delimiter
|
|
"then check for the space and remove it if it is there
|
|
if delimsSpaced && strpart(line, leftIndx, s:lenSpaceStr) == s:spaceStr
|
|
let line = strpart(line, 0, leftIndx) . strpart(line, leftIndx+s:lenSpaceStr)
|
|
endif
|
|
endif
|
|
|
|
"look for the right delimiter, if we find it, remove it
|
|
let rightIndx = s:LastIndexOfDelim(a:right, line)
|
|
if rightIndx != -1
|
|
let line = strpart(line, 0, rightIndx) . strpart(line, rightIndx+lenRight)
|
|
|
|
"if the user has specified that there is a space before the right delimiter
|
|
"then check for the space and remove it if it is there
|
|
if delimsSpaced && strpart(line, rightIndx-s:lenSpaceStr, s:lenSpaceStr) == s:spaceStr && (s:Multipart() || s:AltMultipart())
|
|
let line = strpart(line, 0, rightIndx-s:lenSpaceStr) . strpart(line, rightIndx)
|
|
endif
|
|
endif
|
|
|
|
return line
|
|
endfunction
|
|
|
|
" Function: s:SetupStateBeforeLineComment(topLine, bottomLine) {{{2
|
|
" Changes ignorecase and foldmethod options before commenting lines and saves
|
|
" their original values in a dict, which is returned as a result
|
|
"
|
|
" Args:
|
|
" topLine: the top line of the visual selection to uncomment
|
|
" bottomLine: the bottom line of the visual selection to uncomment
|
|
"
|
|
" Return: a dict with the state prior to configuration changes
|
|
"
|
|
function s:SetupStateBeforeLineComment(topLine, bottomLine)
|
|
let state = {'foldmethod' : &foldmethod,
|
|
\'ignorecase' : &ignorecase}
|
|
|
|
" Vim's foldmethods are evaluated every time we use 'setline', which can
|
|
" make commenting wide ranges of lines VERY slow. We'll change it to
|
|
" manual, do the commenting stuff and recover it later. To avoid slowing
|
|
" down commenting few lines, we avoid doing this for ranges smaller than
|
|
" 10 lines
|
|
if a:bottomLine - a:topLine >= 10 && &foldmethod != "manual"
|
|
set foldmethod=manual
|
|
endif
|
|
|
|
" we want case sensitivity when commenting
|
|
set noignorecase
|
|
|
|
return state
|
|
endfunction
|
|
|
|
" Function: s:RecoverStateAfterLineComment(state) {{{2
|
|
" Receives the state returned by s:SetupStateBeforeLineComment and restores
|
|
" the state accordingly
|
|
"
|
|
" Args:
|
|
" state: the top line of the visual selection to uncomment
|
|
" bottomLine: the bottom line of the visual selection to uncomment
|
|
function s:RecoverStateAfterLineComment(state)
|
|
if a:state['foldmethod'] != &foldmethod
|
|
let &foldmethod = a:state['foldmethod']
|
|
endif
|
|
if a:state['ignorecase'] != &ignorecase
|
|
let &ignorecase = a:state['ignorecase']
|
|
endif
|
|
endfunction
|
|
|
|
" Function: s:TrimTrailingWhitespace(line) {{{2
|
|
" This function removes all the trailing whitespace
|
|
" Args:
|
|
" -line: the target line
|
|
function s:TrimTrailingWhitespace(line)
|
|
let toReturn = substitute(a:line, '\s\+$', '', 'g')
|
|
return toReturn
|
|
endfunction
|
|
|
|
" Function: s:UncommentLines(topLine, bottomLine) {{{2
|
|
" This function uncomments the given lines
|
|
"
|
|
" Args:
|
|
" topLine: the top line of the visual selection to uncomment
|
|
" bottomLine: the bottom line of the visual selection to uncomment
|
|
function s:UncommentLines(topLine, bottomLine)
|
|
"make local copies of a:firstline and a:lastline and, if need be, swap
|
|
"them around if the top line is below the bottom
|
|
let l:firstline = a:topLine
|
|
let l:lastline = a:bottomLine
|
|
if firstline > lastline
|
|
let firstline = lastline
|
|
let lastline = a:topLine
|
|
endif
|
|
|
|
"go thru each line uncommenting each line removing sexy comments
|
|
let currentLine = firstline
|
|
while currentLine <= lastline
|
|
|
|
"check the current line to see if it is part of a sexy comment
|
|
let sexyComBounds = s:FindBoundingLinesOfSexyCom(currentLine)
|
|
if !empty(sexyComBounds)
|
|
|
|
"we need to store the number of lines in the buffer before the comment is
|
|
"removed so we know how many lines were removed when the sexy comment
|
|
"was removed
|
|
let numLinesBeforeSexyComRemoved = s:NumLinesInBuf()
|
|
|
|
call s:UncommentLinesSexy(sexyComBounds[0], sexyComBounds[1])
|
|
|
|
"move to the line after last line of the sexy comment
|
|
let numLinesAfterSexyComRemoved = s:NumLinesInBuf()
|
|
let numLinesRemoved = numLinesBeforeSexyComRemoved - numLinesAfterSexyComRemoved
|
|
let currentLine = sexyComBounds[1] - numLinesRemoved + 1
|
|
let lastline = lastline - numLinesRemoved
|
|
|
|
"no sexy com was detected so uncomment the line as normal
|
|
else
|
|
call s:UncommentLinesNormal(currentLine, currentLine)
|
|
let currentLine = currentLine + 1
|
|
endif
|
|
endwhile
|
|
|
|
endfunction
|
|
|
|
" Function: s:UncommentLinesSexy(topline, bottomline) {{{2
|
|
" This function removes all the comment characters associated with the sexy
|
|
" comment spanning the given lines
|
|
" Args:
|
|
" -topline/bottomline: the top/bottom lines of the sexy comment
|
|
function s:UncommentLinesSexy(topline, bottomline)
|
|
let left = s:GetSexyComLeft(0,1)
|
|
let right = s:GetSexyComRight(0,1)
|
|
|
|
|
|
"check if it is even possible for sexy comments to exist with the
|
|
"available delimiters
|
|
if left == -1 || right == -1
|
|
throw 'NERDCommenter.Delimiters exception: cannot uncomment sexy comments with available delimiters.'
|
|
endif
|
|
|
|
let leftUnEsc = s:GetSexyComLeft(0,0)
|
|
let rightUnEsc = s:GetSexyComRight(0,0)
|
|
|
|
let sexyComMarker = s:GetSexyComMarker(0, 1)
|
|
let sexyComMarkerUnEsc = s:GetSexyComMarker(0, 0)
|
|
|
|
"the markerOffset is how far right we need to move the sexyComMarker to
|
|
"line it up with the end of the left delimiter
|
|
let markerOffset = strlen(leftUnEsc)-strlen(sexyComMarkerUnEsc)
|
|
|
|
" go thru the intermediate lines of the sexy comment and remove the
|
|
" sexy comment markers (e.g., the '*'s on the start of line in a c sexy
|
|
" comment)
|
|
let currentLine = a:topline+1
|
|
while currentLine < a:bottomline
|
|
let theLine = getline(currentLine)
|
|
|
|
" remove the sexy comment marker from the line. We also remove the
|
|
" space after it if there is one and if appropriate options are set
|
|
let sexyComMarkerIndx = stridx(theLine, sexyComMarkerUnEsc)
|
|
if strpart(theLine, sexyComMarkerIndx+strlen(sexyComMarkerUnEsc), s:lenSpaceStr) == s:spaceStr && g:NERDSpaceDelims
|
|
let theLine = strpart(theLine, 0, sexyComMarkerIndx - markerOffset) . strpart(theLine, sexyComMarkerIndx+strlen(sexyComMarkerUnEsc)+s:lenSpaceStr)
|
|
else
|
|
let theLine = strpart(theLine, 0, sexyComMarkerIndx - markerOffset) . strpart(theLine, sexyComMarkerIndx+strlen(sexyComMarkerUnEsc))
|
|
endif
|
|
|
|
let theLine = s:SwapOuterPlaceHoldersForMultiPartDelims(theLine)
|
|
|
|
let theLine = s:ConvertLeadingWhiteSpace(theLine)
|
|
|
|
if g:NERDTrimTrailingWhitespace == 1
|
|
let theLine = s:TrimTrailingWhitespace(theLine)
|
|
endif
|
|
|
|
" move onto the next line
|
|
call setline(currentLine, theLine)
|
|
let currentLine = currentLine + 1
|
|
endwhile
|
|
|
|
" gotta make a copy of a:bottomline cos we modify the position of the
|
|
" last line it if we remove the topline
|
|
let bottomline = a:bottomline
|
|
|
|
" get the first line so we can remove the left delimiter from it
|
|
let theLine = getline(a:topline)
|
|
|
|
" if the first line contains only the left delimiter then just delete it
|
|
if theLine =~ '^[ \t]*' . left . '[ \t]*$' && !g:NERDCompactSexyComs
|
|
call cursor(a:topline, 1)
|
|
normal! dd
|
|
let bottomline = bottomline - 1
|
|
|
|
" topline contains more than just the left delimiter
|
|
else
|
|
|
|
" remove the delimiter. If there is a space after it
|
|
" then remove this too if appropriate
|
|
let delimIndx = stridx(theLine, leftUnEsc)
|
|
if strpart(theLine, delimIndx+strlen(leftUnEsc), s:lenSpaceStr) == s:spaceStr && g:NERDSpaceDelims
|
|
let theLine = strpart(theLine, 0, delimIndx) . strpart(theLine, delimIndx+strlen(leftUnEsc)+s:lenSpaceStr)
|
|
else
|
|
let theLine = strpart(theLine, 0, delimIndx) . strpart(theLine, delimIndx+strlen(leftUnEsc))
|
|
endif
|
|
let theLine = s:SwapOuterPlaceHoldersForMultiPartDelims(theLine)
|
|
call setline(a:topline, theLine)
|
|
endif
|
|
|
|
" get the last line so we can remove the right delimiter
|
|
let theLine = getline(bottomline)
|
|
|
|
" if the bottomline contains only the right delimiter then just delete it
|
|
if theLine =~ '^[ \t]*' . right . '[ \t]*$'
|
|
call cursor(bottomline, 1)
|
|
normal! dd
|
|
|
|
" the last line contains more than the right delimiter
|
|
else
|
|
" remove the right delimiter. If there is a space after it and
|
|
" if the appropriate options are set then remove this too.
|
|
let delimIndx = s:LastIndexOfDelim(rightUnEsc, theLine)
|
|
if strpart(theLine, delimIndx+strlen(leftUnEsc), s:lenSpaceStr) == s:spaceStr && g:NERDSpaceDelims
|
|
let theLine = strpart(theLine, 0, delimIndx) . strpart(theLine, delimIndx+strlen(rightUnEsc)+s:lenSpaceStr)
|
|
else
|
|
let theLine = strpart(theLine, 0, delimIndx) . strpart(theLine, delimIndx+strlen(rightUnEsc))
|
|
endif
|
|
|
|
" if the last line also starts with a sexy comment marker then we
|
|
" remove this as well
|
|
if theLine =~ '^[ \t]*' . sexyComMarker
|
|
|
|
" remove the sexyComMarker. If there is a space after it then
|
|
" remove that too
|
|
let sexyComMarkerIndx = stridx(theLine, sexyComMarkerUnEsc)
|
|
if strpart(theLine, sexyComMarkerIndx+strlen(sexyComMarkerUnEsc), s:lenSpaceStr) == s:spaceStr && g:NERDSpaceDelims
|
|
let theLine = strpart(theLine, 0, sexyComMarkerIndx - markerOffset ) . strpart(theLine, sexyComMarkerIndx+strlen(sexyComMarkerUnEsc)+s:lenSpaceStr)
|
|
else
|
|
let theLine = strpart(theLine, 0, sexyComMarkerIndx - markerOffset ) . strpart(theLine, sexyComMarkerIndx+strlen(sexyComMarkerUnEsc))
|
|
endif
|
|
endif
|
|
|
|
let theLine = s:SwapOuterPlaceHoldersForMultiPartDelims(theLine)
|
|
call setline(bottomline, theLine)
|
|
endif
|
|
|
|
" remove trailing whitespaces for first and last line
|
|
if g:NERDTrimTrailingWhitespace == 1
|
|
let theLine = getline(a:bottomline)
|
|
let theLine = s:TrimTrailingWhitespace(theLine)
|
|
call setline(a:bottomline, theLine)
|
|
let theLine = getline(a:topline)
|
|
let theLine = s:TrimTrailingWhitespace(theLine)
|
|
call setline(a:topline, theLine)
|
|
endif
|
|
endfunction
|
|
|
|
" Function: s:UncommentLineNormal(line) {{{2
|
|
" uncomments the given line and returns the result
|
|
" Args:
|
|
" -line: the line to uncomment
|
|
function s:UncommentLineNormal(line)
|
|
let line = a:line
|
|
|
|
"get the positions of all delimiter types on the line
|
|
let indxLeft = s:FindDelimiterIndex(s:Left(), line)
|
|
let indxLeftAlt = s:FindDelimiterIndex(s:Left({'alt': 1}), line)
|
|
let indxRight = s:LastIndexOfDelim(s:Right(), line)
|
|
let indxRightAlt = s:LastIndexOfDelim(s:Right({'alt': 1}), line)
|
|
|
|
"get the comment status on the line so we know how it is commented
|
|
let lineCommentStatus = s:IsCommentedOutermost(s:Left(), s:Right(), s:Left({'alt': 1}), s:Right({'alt': 1}), line)
|
|
|
|
"it is commented with s:Left() and s:Right() so remove these delimiters
|
|
if lineCommentStatus == 1
|
|
let line = s:RemoveDelimiters(s:Left(), s:Right(), line)
|
|
|
|
"it is commented with s:Left({'alt': 1}) and s:Right({'alt': 1}) so remove these delimiters
|
|
elseif lineCommentStatus == 2 && g:NERDRemoveAltComs
|
|
let line = s:RemoveDelimiters(s:Left({'alt': 1}), s:Right({'alt': 1}), line)
|
|
|
|
"it is not properly commented with any delimiters so we check if it has
|
|
"any random left or right delimiters on it and remove the outermost ones
|
|
else
|
|
"remove the outer most left comment delimiter
|
|
if indxLeft != -1 && (indxLeft < indxLeftAlt || indxLeftAlt == -1)
|
|
let line = s:RemoveDelimiters(s:Left(), '', line)
|
|
elseif indxLeftAlt != -1 && g:NERDRemoveAltComs
|
|
let line = s:RemoveDelimiters(s:Left({'alt': 1}), '', line)
|
|
endif
|
|
|
|
"remove the outer most right comment delimiter
|
|
if indxRight != -1 && (indxRight < indxRightAlt || indxRightAlt == -1)
|
|
let line = s:RemoveDelimiters('', s:Right(), line)
|
|
elseif indxRightAlt != -1 && g:NERDRemoveAltComs
|
|
let line = s:RemoveDelimiters('', s:Right({'alt': 1}), line)
|
|
endif
|
|
endif
|
|
|
|
|
|
let indxLeftPlace = s:FindDelimiterIndex(g:NERDLPlace, line)
|
|
let indxRightPlace = s:FindDelimiterIndex(g:NERDRPlace, line)
|
|
|
|
let right = s:Right()
|
|
let left = s:Left()
|
|
if !s:Multipart()
|
|
let right = s:Right({'alt': 1})
|
|
let left = s:Left({'alt': 1})
|
|
endif
|
|
|
|
|
|
"if there are place-holders on the line then we check to see if they are
|
|
"the outermost delimiters on the line. If so then we replace them with
|
|
"real delimiters
|
|
if indxLeftPlace != -1
|
|
if (indxLeftPlace < indxLeft || indxLeft==-1) && (indxLeftPlace < indxLeftAlt || indxLeftAlt==-1)
|
|
let line = s:ReplaceDelims(g:NERDLPlace, g:NERDRPlace, left, right, line)
|
|
endif
|
|
elseif indxRightPlace != -1
|
|
if (indxRightPlace < indxLeft || indxLeft==-1) && (indxLeftPlace < indxLeftAlt || indxLeftAlt==-1)
|
|
let line = s:ReplaceDelims(g:NERDLPlace, g:NERDRPlace, left, right, line)
|
|
endif
|
|
|
|
endif
|
|
|
|
let line = s:ConvertLeadingWhiteSpace(line)
|
|
|
|
if g:NERDTrimTrailingWhitespace == 1
|
|
let line = s:TrimTrailingWhitespace(line)
|
|
endif
|
|
|
|
return line
|
|
endfunction
|
|
|
|
" Function: s:UncommentLinesNormal(topline, bottomline) {{{2
|
|
" This function is called to uncomment lines that aren't a sexy comment
|
|
" Args:
|
|
" -topline/bottomline: the top/bottom line numbers of the comment
|
|
function s:UncommentLinesNormal(topline, bottomline)
|
|
let currentLine = a:topline
|
|
while currentLine <= a:bottomline
|
|
let line = getline(currentLine)
|
|
call setline(currentLine, s:UncommentLineNormal(line))
|
|
let currentLine = currentLine + 1
|
|
endwhile
|
|
endfunction
|
|
|
|
|
|
" Section: Other helper functions {{{1
|
|
" ============================================================================
|
|
|
|
" Function: s:AddLeftDelim(delim, theLine) {{{2
|
|
" Args:
|
|
function s:AddLeftDelim(delim, theLine)
|
|
return substitute(a:theLine, '^\([ \t]*\)', '\1' . a:delim, '')
|
|
endfunction
|
|
|
|
" Function: s:AddLeftDelimAligned(delim, theLine) {{{2
|
|
" Args:
|
|
function s:AddLeftDelimAligned(delim, theLine, alignIndx)
|
|
|
|
"if the line is not long enough then bung some extra spaces on the front
|
|
"so we can align the delimiter properly
|
|
let theLine = a:theLine
|
|
if strlen(theLine) < a:alignIndx
|
|
let theLine = repeat(' ', a:alignIndx - strlen(theLine))
|
|
endif
|
|
|
|
return strpart(theLine, 0, a:alignIndx) . a:delim . strpart(theLine, a:alignIndx)
|
|
endfunction
|
|
|
|
" Function: s:AddRightDelim(delim, theLine) {{{2
|
|
" Args:
|
|
function s:AddRightDelim(delim, theLine)
|
|
if a:delim == ''
|
|
return a:theLine
|
|
else
|
|
return substitute(a:theLine, '$', a:delim, '')
|
|
endif
|
|
endfunction
|
|
|
|
" Function: s:AddRightDelimAligned(delim, theLine, alignIndx) {{{2
|
|
" Args:
|
|
function s:AddRightDelimAligned(delim, theLine, alignIndx)
|
|
if a:delim == ""
|
|
return a:theLine
|
|
else
|
|
|
|
" when we align the right delimiter we are just adding spaces
|
|
" so we get a string containing the needed spaces (it
|
|
" could be empty)
|
|
let extraSpaces = ''
|
|
let extraSpaces = repeat(' ', a:alignIndx-strlen(a:theLine))
|
|
|
|
" add the right delimiter
|
|
return substitute(a:theLine, '$', extraSpaces . a:delim, '')
|
|
endif
|
|
endfunction
|
|
|
|
" Function: s:AltMultipart() {{{2
|
|
" returns 1 if the alternative delimiters are multipart
|
|
function s:AltMultipart()
|
|
return b:NERDCommenterDelims['rightAlt'] != ''
|
|
endfunction
|
|
|
|
" Function: s:AltNested() {{{2
|
|
" returns 1 if the alternate multipart (if any) delimiters allow nesting
|
|
function s:AltNested()
|
|
return b:NERDCommenterDelims['nestedAlt']
|
|
endfunction
|
|
|
|
" Function: s:CanCommentLine(forceNested, line) {{{2
|
|
"This function is used to determine whether the given line can be commented.
|
|
"It returns 1 if it can be and 0 otherwise
|
|
"
|
|
" Args:
|
|
" -forceNested: a flag indicating whether the caller wants comments to be nested
|
|
" if the current line is already commented
|
|
" -lineNum: the line number of the line to check for commentability
|
|
function s:CanCommentLine(forceNested, lineNum)
|
|
let theLine = getline(a:lineNum)
|
|
|
|
" make sure we don't comment lines that are just spaces or tabs or empty,
|
|
" unless configured otherwise
|
|
if g:NERDCommentEmptyLines == 0 && theLine =~ "^[ \t]*$"
|
|
return 0
|
|
endif
|
|
|
|
"if the line is part of a sexy comment then just flag it...
|
|
if s:IsInSexyComment(a:lineNum)
|
|
return 0
|
|
endif
|
|
|
|
let isCommented = s:IsCommentedNormOrSexy(a:lineNum)
|
|
|
|
"if the line isn't commented return true
|
|
if !isCommented
|
|
return 1
|
|
endif
|
|
|
|
"if the line is commented but nesting is allowed then return true
|
|
if s:Nested() || (a:forceNested && (!s:Multipart() || g:NERDUsePlaceHolders))
|
|
return 1
|
|
endif
|
|
|
|
return 0
|
|
endfunction
|
|
|
|
" Function: s:CanPlaceCursor(line, col) {{{2
|
|
" returns 1 if the cursor can be placed exactly in the given position
|
|
function s:CanPlaceCursor(line, col)
|
|
let c = col(".")
|
|
let l = line(".")
|
|
call cursor(a:line, a:col)
|
|
let success = (line(".") == a:line && col(".") == a:col)
|
|
call cursor(l,c)
|
|
return success
|
|
endfunction
|
|
|
|
" Function: s:CanSexyCommentLines(topline, bottomline) {{{2
|
|
" Return: 1 if the given lines can be commented sexually, 0 otherwise
|
|
function s:CanSexyCommentLines(topline, bottomline)
|
|
" see if the selected regions have any sexy comments
|
|
" however, if the language allows nested comments,
|
|
" we allow nested sexy comments
|
|
if s:SexyNested()
|
|
return 1
|
|
endif
|
|
let currentLine = a:topline
|
|
while(currentLine <= a:bottomline)
|
|
if s:IsInSexyComment(currentLine)
|
|
return 0
|
|
endif
|
|
let currentLine = currentLine + 1
|
|
endwhile
|
|
return 1
|
|
endfunction
|
|
" Function: s:CanToggleCommentLine(forceNested, line) {{{2
|
|
"This function is used to determine whether the given line can be toggle commented.
|
|
"It returns 1 if it can be and 0 otherwise
|
|
"
|
|
" Args:
|
|
" -lineNum: the line number of the line to check for commentability
|
|
function s:CanToggleCommentLine(forceNested, lineNum)
|
|
let theLine = getline(a:lineNum)
|
|
if (s:IsCommentedFromStartOfLine(s:Left(), theLine) || s:IsCommentedFromStartOfLine(s:Left({'alt': 1}), theLine)) && !a:forceNested
|
|
return 0
|
|
endif
|
|
|
|
" make sure we don't comment lines that are just spaces or tabs or empty,
|
|
" unless configured otherwise
|
|
if g:NERDCommentEmptyLines == 0 && theLine =~ "^[ \t]*$"
|
|
return 0
|
|
endif
|
|
|
|
"if the line is part of a sexy comment then just flag it...
|
|
if s:IsInSexyComment(a:lineNum)
|
|
return 0
|
|
endif
|
|
|
|
return 1
|
|
endfunction
|
|
|
|
" Function: s:ConvertLeadingSpacesToTabs(line) {{{2
|
|
" This function takes a line and converts all leading tabs on that line into
|
|
" spaces
|
|
"
|
|
" Args:
|
|
" -line: the line whose leading tabs will be converted
|
|
function s:ConvertLeadingSpacesToTabs(line)
|
|
let toReturn = a:line
|
|
while toReturn =~ '^\t*' . s:TabSpace() . '\(.*\)$'
|
|
let toReturn = substitute(toReturn, '^\(\t*\)' . s:TabSpace() . '\(.*\)$' , '\1\t\2' , "")
|
|
endwhile
|
|
|
|
return toReturn
|
|
endfunction
|
|
|
|
|
|
" Function: s:ConvertLeadingTabsToSpaces(line) {{{2
|
|
" This function takes a line and converts all leading spaces on that line into
|
|
" tabs
|
|
"
|
|
" Args:
|
|
" -line: the line whose leading spaces will be converted
|
|
function s:ConvertLeadingTabsToSpaces(line)
|
|
let toReturn = a:line
|
|
while toReturn =~ '^\( *\)\t'
|
|
let toReturn = substitute(toReturn, '^\( *\)\t', '\1' . s:TabSpace() , "")
|
|
endwhile
|
|
|
|
return toReturn
|
|
endfunction
|
|
|
|
" Function: s:ConvertLeadingWhiteSpace(line) {{{2
|
|
" Converts the leading white space to tabs/spaces depending on &ts
|
|
"
|
|
" Args:
|
|
" -line: the line to convert
|
|
function s:ConvertLeadingWhiteSpace(line)
|
|
let toReturn = a:line
|
|
while toReturn =~ '^ *\t'
|
|
let toReturn = substitute(toReturn, '^ *\zs\t\ze', s:TabSpace(), "g")
|
|
endwhile
|
|
|
|
if !&expandtab
|
|
let toReturn = s:ConvertLeadingSpacesToTabs(toReturn)
|
|
endif
|
|
|
|
return toReturn
|
|
endfunction
|
|
|
|
|
|
" Function: s:CountNonESCedOccurances(str, searchstr, escChar) {{{2
|
|
" This function counts the number of substrings contained in another string.
|
|
" These substrings are only counted if they are not escaped with escChar
|
|
" Args:
|
|
" -str: the string to look for searchstr in
|
|
" -searchstr: the substring to search for in str
|
|
" -escChar: the escape character which, when preceding an instance of
|
|
" searchstr, will cause it not to be counted
|
|
function s:CountNonESCedOccurances(str, searchstr, escChar)
|
|
"get the index of the first occurrence of searchstr
|
|
let indx = stridx(a:str, a:searchstr)
|
|
|
|
"if there is an instance of searchstr in str process it
|
|
if indx != -1
|
|
"get the remainder of str after this instance of searchstr is removed
|
|
let lensearchstr = strlen(a:searchstr)
|
|
let strLeft = strpart(a:str, indx+lensearchstr)
|
|
|
|
"if this instance of searchstr is not escaped, add one to the count
|
|
"and recurse. If it is escaped, just recurse
|
|
if !s:IsEscaped(a:str, indx, a:escChar)
|
|
return 1 + s:CountNonESCedOccurances(strLeft, a:searchstr, a:escChar)
|
|
else
|
|
return s:CountNonESCedOccurances(strLeft, a:searchstr, a:escChar)
|
|
endif
|
|
endif
|
|
endfunction
|
|
" Function: s:DoesBlockHaveDelim(delim, top, bottom) {{{2
|
|
" Returns 1 if the given block of lines has a delimiter (a:delim) in it
|
|
" Args:
|
|
" -delim: the comment delimiter to check the block for
|
|
" -top: the top line number of the block
|
|
" -bottom: the bottom line number of the block
|
|
function s:DoesBlockHaveDelim(delim, top, bottom)
|
|
let currentLine = a:top
|
|
while currentLine < a:bottom
|
|
let theline = getline(currentLine)
|
|
if s:FindDelimiterIndex(a:delim, theline) != -1
|
|
return 1
|
|
endif
|
|
let currentLine = currentLine + 1
|
|
endwhile
|
|
return 0
|
|
endfunction
|
|
|
|
" Function: s:DoesBlockHaveMultipartDelim(top, bottom) {{{2
|
|
" Returns 1 if the given block has a >= 1 multipart delimiter in it
|
|
" Args:
|
|
" -top: the top line number of the block
|
|
" -bottom: the bottom line number of the block
|
|
function s:DoesBlockHaveMultipartDelim(top, bottom)
|
|
if s:HasMultipartDelims()
|
|
if s:Multipart()
|
|
return s:DoesBlockHaveDelim(s:Left(), a:top, a:bottom) || s:DoesBlockHaveDelim(s:Right(), a:top, a:bottom)
|
|
else
|
|
return s:DoesBlockHaveDelim(s:Left({'alt': 1}), a:top, a:bottom) || s:DoesBlockHaveDelim(s:Right({'alt': 1}), a:top, a:bottom)
|
|
endif
|
|
endif
|
|
return 0
|
|
endfunction
|
|
|
|
|
|
" Function: s:Esc(str) {{{2
|
|
" Escapes all the tricky chars in the given string
|
|
function s:Esc(str)
|
|
let charsToEsc = '*/\."&$+'
|
|
return escape(a:str, charsToEsc)
|
|
endfunction
|
|
|
|
" Function: s:FindDelimiterIndex(delimiter, line) {{{2
|
|
" This function is used to get the string index of the input comment delimiter
|
|
" on the input line. If no valid comment delimiter is found in the line then
|
|
" -1 is returned
|
|
" Args:
|
|
" -delimiter: the delimiter we are looking to find the index of
|
|
" -line: the line we are looking for delimiter on
|
|
function s:FindDelimiterIndex(delimiter, line)
|
|
|
|
"make sure the delimiter isn't empty otherwise we go into an infinite loop.
|
|
if a:delimiter == ""
|
|
return -1
|
|
endif
|
|
|
|
|
|
let l:delimiter = a:delimiter
|
|
let lenDel = strlen(l:delimiter)
|
|
|
|
"get the index of the first occurrence of the delimiter
|
|
let delIndx = stridx(a:line, l:delimiter)
|
|
|
|
"keep looping thru the line till we either find a real comment delimiter
|
|
"or run off the EOL
|
|
while delIndx != -1
|
|
|
|
"if we are not off the EOL get the str before the possible delimiter
|
|
"in question and check if it really is a delimiter. If it is, return
|
|
"its position
|
|
if delIndx != -1
|
|
if s:IsDelimValid(l:delimiter, delIndx, a:line)
|
|
return delIndx
|
|
endif
|
|
endif
|
|
|
|
"we have not yet found a real comment delimiter so move past the
|
|
"current one we are looking at
|
|
let restOfLine = strpart(a:line, delIndx + lenDel)
|
|
let distToNextDelim = stridx(restOfLine , l:delimiter)
|
|
|
|
"if distToNextDelim is -1 then there is no more potential delimiters
|
|
"on the line so set delIndx to -1. Otherwise, move along the line by
|
|
"distToNextDelim
|
|
if distToNextDelim == -1
|
|
let delIndx = -1
|
|
else
|
|
let delIndx = delIndx + lenDel + distToNextDelim
|
|
endif
|
|
endwhile
|
|
|
|
"there is no comment delimiter on this line
|
|
return -1
|
|
endfunction
|
|
|
|
" Function: s:FindBoundingLinesOfSexyCom(lineNum) {{{2
|
|
" This function takes in a line number and tests whether this line number is
|
|
" the top/bottom/middle line of a sexy comment. If it is then the top/bottom
|
|
" lines of the sexy comment are returned
|
|
" Args:
|
|
" -lineNum: the line number that is to be tested whether it is the
|
|
" top/bottom/middle line of a sexy com
|
|
" Returns:
|
|
" A string that has the top/bottom lines of the sexy comment encoded in it.
|
|
" The format is 'topline,bottomline'. If a:lineNum turns out not to be the
|
|
" top/bottom/middle of a sexy comment then -1 is returned
|
|
function s:FindBoundingLinesOfSexyCom(lineNum)
|
|
|
|
"find which delimiters to look for as the start/end delimiters of the comment
|
|
let left = ''
|
|
let right = ''
|
|
if s:Multipart()
|
|
let left = s:Left({'esc': 1})
|
|
let right = s:Right({'esc': 1})
|
|
elseif s:AltMultipart()
|
|
let left = s:Left({'alt': 1, 'esc': 1})
|
|
let right = s:Right({'alt': 1, 'esc': 1})
|
|
else
|
|
return []
|
|
endif
|
|
|
|
let sexyComMarker = s:GetSexyComMarker(0, 1)
|
|
|
|
"initialise the top/bottom line numbers of the sexy comment to -1
|
|
let top = -1
|
|
let bottom = -1
|
|
|
|
let currentLine = a:lineNum
|
|
while top == -1 || bottom == -1
|
|
let theLine = getline(currentLine)
|
|
|
|
"check if the current line is the top of the sexy comment
|
|
if currentLine <= a:lineNum && theLine =~ '^[ \t]*' . left && theLine !~ '.*' . right && currentLine < s:NumLinesInBuf()
|
|
let top = currentLine
|
|
let currentLine = a:lineNum
|
|
|
|
"check if the current line is the bottom of the sexy comment
|
|
elseif theLine =~ '^[ \t]*' . right && theLine !~ '.*' . left && currentLine > 1
|
|
let bottom = currentLine
|
|
|
|
"the right delimiter is on the same line as the last sexyComMarker
|
|
elseif theLine =~ '^[ \t]*' . sexyComMarker . '.*' . right
|
|
let bottom = currentLine
|
|
|
|
"we have not found the top or bottom line so we assume currentLine is an
|
|
"intermediate line and look to prove otherwise
|
|
else
|
|
|
|
"if the line doesn't start with a sexyComMarker then it is not a sexy
|
|
"comment
|
|
if theLine !~ '^[ \t]*' . sexyComMarker
|
|
return []
|
|
endif
|
|
|
|
endif
|
|
|
|
"if top is -1 then we haven't found the top yet so keep looking up
|
|
if top == -1
|
|
let currentLine = currentLine - 1
|
|
"if we have found the top line then go down looking for the bottom
|
|
else
|
|
let currentLine = currentLine + 1
|
|
endif
|
|
|
|
endwhile
|
|
|
|
return [top, bottom]
|
|
endfunction
|
|
|
|
|
|
" Function: s:GetSexyComMarker() {{{2
|
|
" Returns the sexy comment marker for the current filetype.
|
|
"
|
|
" C style sexy comments are assumed if possible. If not then the sexy comment
|
|
" marker is the last char of the delimiter pair that has both left and right
|
|
" delimiters and has the longest left delimiter
|
|
"
|
|
" Args:
|
|
" -space: specifies whether the marker is to have a space string after it
|
|
" (the space string will only be added if NERDSpaceDelims is set)
|
|
" -esc: specifies whether the tricky chars in the marker are to be ESCed
|
|
function s:GetSexyComMarker(space, esc)
|
|
let sexyComMarker = b:NERDSexyComMarker
|
|
|
|
"if there is no hardcoded marker then we find one
|
|
if sexyComMarker == ''
|
|
|
|
"if the filetype has c style comments then use standard c sexy
|
|
"comments
|
|
if s:HasCStyleComments()
|
|
let sexyComMarker = '*'
|
|
else
|
|
"find a comment marker by getting the longest available left delimiter
|
|
"(that has a corresponding right delimiter) and taking the last char
|
|
let lenLeft = strlen(s:Left())
|
|
let lenLeftAlt = strlen(s:Left({'alt': 1}))
|
|
let left = ''
|
|
let right = ''
|
|
if s:Multipart() && lenLeft >= lenLeftAlt
|
|
let left = s:Left()
|
|
elseif s:AltMultipart()
|
|
let left = s:Left({'alt': 1})
|
|
else
|
|
return -1
|
|
endif
|
|
|
|
"get the last char of left
|
|
let sexyComMarker = strpart(left, strlen(left)-1)
|
|
endif
|
|
endif
|
|
|
|
if a:space && g:NERDSpaceDelims
|
|
let sexyComMarker = sexyComMarker . s:spaceStr
|
|
endif
|
|
|
|
if a:esc
|
|
let sexyComMarker = s:Esc(sexyComMarker)
|
|
endif
|
|
|
|
return sexyComMarker
|
|
endfunction
|
|
|
|
" Function: s:SexyNested() {{{2
|
|
" Returns 1 if the sexy delimeters allow nesting
|
|
" TODO this is ugly copy&paste from the GetSexyComLeft/Right functions,
|
|
" these could all be cleaned up
|
|
function s:SexyNested()
|
|
let lenLeft = strlen(s:Left())
|
|
let lenLeftAlt = strlen(s:Left({'alt': 1}))
|
|
|
|
"assume c style sexy comments if possible
|
|
if s:HasCStyleComments()
|
|
return (s:Left() == '/*' && s:Nested()) || (s:Left({'alt': 1}) == '/*' && s:AltNested())
|
|
else
|
|
"grab the longest left delim that has a right
|
|
if s:Multipart() && lenLeft >= lenLeftAlt
|
|
return s:Nested()
|
|
elseif s:AltMultipart()
|
|
return s:AltNested()
|
|
else
|
|
return 0
|
|
endif
|
|
endif
|
|
endfunction
|
|
|
|
" Function: s:GetSexyComLeft(space, esc) {{{2
|
|
" Returns the left delimiter for sexy comments for this filetype or -1 if
|
|
" there is none. C style sexy comments are used if possible
|
|
" Args:
|
|
" -space: specifies if the delimiter has a space string on the end
|
|
" (the space string will only be added if NERDSpaceDelims is set)
|
|
" -esc: specifies whether the tricky chars in the string are ESCed
|
|
function s:GetSexyComLeft(space, esc)
|
|
let lenLeft = strlen(s:Left())
|
|
let lenLeftAlt = strlen(s:Left({'alt': 1}))
|
|
let left = ''
|
|
|
|
"assume c style sexy comments if possible
|
|
if s:HasCStyleComments()
|
|
let left = '/*'
|
|
else
|
|
"grab the longest left delimiter that has a right
|
|
if s:Multipart() && lenLeft >= lenLeftAlt
|
|
let left = s:Left()
|
|
elseif s:AltMultipart()
|
|
let left = s:Left({'alt': 1})
|
|
else
|
|
return -1
|
|
endif
|
|
endif
|
|
|
|
if a:space && g:NERDSpaceDelims
|
|
let left = left . s:spaceStr
|
|
endif
|
|
|
|
if a:esc
|
|
let left = s:Esc(left)
|
|
endif
|
|
|
|
return left
|
|
endfunction
|
|
|
|
" Function: s:GetSexyComRight(space, esc) {{{2
|
|
" Returns the right delimiter for sexy comments for this filetype or -1 if
|
|
" there is none. C style sexy comments are used if possible.
|
|
" Args:
|
|
" -space: specifies if the delimiter has a space string on the start
|
|
" (the space string will only be added if NERDSpaceDelims
|
|
" is specified for the current filetype)
|
|
" -esc: specifies whether the tricky chars in the string are ESCed
|
|
function s:GetSexyComRight(space, esc)
|
|
let lenLeft = strlen(s:Left())
|
|
let lenLeftAlt = strlen(s:Left({'alt': 1}))
|
|
let right = ''
|
|
|
|
"assume c style sexy comments if possible
|
|
if s:HasCStyleComments()
|
|
let right = '*/'
|
|
else
|
|
"grab the right delimiter that pairs with the longest left delimiter
|
|
if s:Multipart() && lenLeft >= lenLeftAlt
|
|
let right = s:Right()
|
|
elseif s:AltMultipart()
|
|
let right = s:Right({'alt': 1})
|
|
else
|
|
return -1
|
|
endif
|
|
endif
|
|
|
|
if a:space && g:NERDSpaceDelims
|
|
let right = s:spaceStr . right
|
|
endif
|
|
|
|
if a:esc
|
|
let right = s:Esc(right)
|
|
endif
|
|
|
|
return right
|
|
endfunction
|
|
|
|
" Function: s:HasMultipartDelims() {{{2
|
|
" Returns 1 if the current filetype has at least one set of multipart delimiters
|
|
function s:HasMultipartDelims()
|
|
return s:Multipart() || s:AltMultipart()
|
|
endfunction
|
|
|
|
" Function: s:HasLeadingTabs(...) {{{2
|
|
" Returns 1 if any of the given strings have leading tabs
|
|
function s:HasLeadingTabs(...)
|
|
for s in a:000
|
|
if s =~ '^\t.*'
|
|
return 1
|
|
end
|
|
endfor
|
|
return 0
|
|
endfunction
|
|
" Function: s:HasCStyleComments() {{{2
|
|
" Returns 1 if the current filetype has c style comment delimiters
|
|
function s:HasCStyleComments()
|
|
return (s:Left() == '/*' && s:Right() == '*/') || (s:Left({'alt': 1}) == '/*' && s:Right({'alt': 1}) == '*/')
|
|
endfunction
|
|
|
|
" Function: s:IsCommentedNormOrSexy(lineNum) {{{2
|
|
"This function is used to determine whether the given line is commented with
|
|
"either set of delimiters or if it is part of a sexy comment
|
|
"
|
|
" Args:
|
|
" -lineNum: the line number of the line to check
|
|
function s:IsCommentedNormOrSexy(lineNum)
|
|
let theLine = getline(a:lineNum)
|
|
|
|
"if the line is commented normally return 1
|
|
if s:IsCommented(s:Left(), s:Right(), theLine) || s:IsCommented(s:Left({'alt': 1}), s:Right({'alt': 1}), theLine)
|
|
return 1
|
|
endif
|
|
|
|
"if the line is part of a sexy comment return 1
|
|
if s:IsInSexyComment(a:lineNum)
|
|
return 1
|
|
endif
|
|
return 0
|
|
endfunction
|
|
|
|
" Function: s:IsCommented(left, right, line) {{{2
|
|
"This function is used to determine whether the given line is commented with
|
|
"the given delimiters
|
|
"
|
|
" Args:
|
|
" -line: the line that to check if commented
|
|
" -left/right: the left and right delimiters to check for
|
|
function s:IsCommented(left, right, line)
|
|
"if the line isn't commented return true
|
|
if s:FindDelimiterIndex(a:left, a:line) != -1 && (s:LastIndexOfDelim(a:right, a:line) != -1 || !s:Multipart())
|
|
return 1
|
|
endif
|
|
return 0
|
|
endfunction
|
|
|
|
" Function: s:IsCommentedFromStartOfLine(left, line) {{{2
|
|
"This function is used to determine whether the given line is commented with
|
|
"the given delimiters at the start of the line i.e the left delimiter is the
|
|
"first thing on the line (apart from spaces\tabs)
|
|
"
|
|
" Args:
|
|
" -line: the line that to check if commented
|
|
" -left: the left delimiter to check for
|
|
function s:IsCommentedFromStartOfLine(left, line)
|
|
let theLine = s:ConvertLeadingTabsToSpaces(a:line)
|
|
let numSpaces = strlen(substitute(theLine, '^\( *\).*$', '\1', ''))
|
|
let delimIndx = s:FindDelimiterIndex(a:left, theLine)
|
|
return delimIndx == numSpaces
|
|
endfunction
|
|
|
|
" Function: s:IsCommentedOutermost(left, right, leftAlt, rightAlt, line) {{{2
|
|
" Finds the type of the outermost delimiters on the line
|
|
"
|
|
" Args:
|
|
" -line: the line that to check if the outermost comments on it are
|
|
" left/right
|
|
" -left/right: the left and right delimiters to check for
|
|
" -leftAlt/rightAlt: the left and right alternative delimiters to check for
|
|
"
|
|
" Returns:
|
|
" 0 if the line is not commented with either set of delimiters
|
|
" 1 if the line is commented with the left/right delimiter set
|
|
" 2 if the line is commented with the leftAlt/rightAlt delim set
|
|
function s:IsCommentedOutermost(left, right, leftAlt, rightAlt, line)
|
|
"get the first positions of the left delimiters and the last positions of the
|
|
"right delimiters
|
|
let indxLeft = s:FindDelimiterIndex(a:left, a:line)
|
|
let indxLeftAlt = s:FindDelimiterIndex(a:leftAlt, a:line)
|
|
let indxRight = s:LastIndexOfDelim(a:right, a:line)
|
|
let indxRightAlt = s:LastIndexOfDelim(a:rightAlt, a:line)
|
|
|
|
"check if the line has a left delimiter before a leftAlt delimiter
|
|
if (indxLeft <= indxLeftAlt || indxLeftAlt == -1) && indxLeft != -1
|
|
"check if the line has a right delimiter after any rightAlt delimiter
|
|
if (indxRight > indxRightAlt && indxRight > indxLeft) || !s:Multipart()
|
|
return 1
|
|
endif
|
|
|
|
"check if the line has a leftAlt delimiter before a left delimiter
|
|
elseif (indxLeftAlt <= indxLeft || indxLeft == -1) && indxLeftAlt != -1
|
|
"check if the line has a rightAlt delimiter after any right delimiter
|
|
if (indxRightAlt > indxRight && indxRightAlt > indxLeftAlt) || !s:AltMultipart()
|
|
return 2
|
|
endif
|
|
else
|
|
return 0
|
|
endif
|
|
|
|
return 0
|
|
|
|
endfunction
|
|
|
|
|
|
" Function: s:IsDelimValid(delimiter, delIndx, line) {{{2
|
|
" This function is responsible for determining whether a given instance of a
|
|
" comment delimiter is a real delimiter or not. For example, in java the
|
|
" // string is a comment delimiter but in the line:
|
|
" System.out.println("//");
|
|
" it does not count as a comment delimiter. This function is responsible for
|
|
" distinguishing between such cases. It does so by applying a set of
|
|
" heuristics that are not fool proof but should work most of the time.
|
|
"
|
|
" Args:
|
|
" -delimiter: the delimiter we are validating
|
|
" -delIndx: the position of delimiter in line
|
|
" -line: the line that delimiter occurs in
|
|
"
|
|
" Returns:
|
|
" 0 if the given delimiter is not a real delimiter (as far as we can tell) ,
|
|
" 1 otherwise
|
|
function s:IsDelimValid(delimiter, delIndx, line)
|
|
"get the delimiter without the escchars
|
|
let l:delimiter = a:delimiter
|
|
|
|
"get the strings before and after the delimiter
|
|
let preComStr = strpart(a:line, 0, a:delIndx)
|
|
let postComStr = strpart(a:line, a:delIndx+strlen(delimiter))
|
|
|
|
"to check if the delimiter is real, make sure it isn't preceded by
|
|
"an odd number of quotes and followed by the same (which would indicate
|
|
"that it is part of a string and therefore is not a comment)
|
|
if !s:IsNumEven(s:CountNonESCedOccurances(preComStr, '"', "\\")) && !s:IsNumEven(s:CountNonESCedOccurances(postComStr, '"', "\\"))
|
|
return 0
|
|
endif
|
|
if !s:IsNumEven(s:CountNonESCedOccurances(preComStr, "'", "\\")) && !s:IsNumEven(s:CountNonESCedOccurances(postComStr, "'", "\\"))
|
|
return 0
|
|
endif
|
|
if !s:IsNumEven(s:CountNonESCedOccurances(preComStr, "`", "\\")) && !s:IsNumEven(s:CountNonESCedOccurances(postComStr, "`", "\\"))
|
|
return 0
|
|
endif
|
|
|
|
|
|
"if the comment delimiter is escaped, assume it isn't a real delimiter
|
|
if s:IsEscaped(a:line, a:delIndx, "\\")
|
|
return 0
|
|
endif
|
|
|
|
"vim comments are so fucking stupid!! Why the hell do they have comment
|
|
"delimiters that are used elsewhere in the syntax?!?! We need to check
|
|
"some conditions especially for vim
|
|
if &filetype == "vim"
|
|
if !s:IsNumEven(s:CountNonESCedOccurances(preComStr, '"', "\\"))
|
|
return 0
|
|
endif
|
|
|
|
"if the delimiter is on the very first char of the line or is the
|
|
"first non-tab/space char on the line then it is a valid comment delimiter
|
|
if a:delIndx == 0 || a:line =~ "^[ \t]\\{" . a:delIndx . "\\}\".*$"
|
|
return 1
|
|
endif
|
|
|
|
let numLeftParen =s:CountNonESCedOccurances(preComStr, "(", "\\")
|
|
let numRightParen =s:CountNonESCedOccurances(preComStr, ")", "\\")
|
|
|
|
"if the quote is inside brackets then assume it isn't a comment
|
|
if numLeftParen > numRightParen
|
|
return 0
|
|
endif
|
|
|
|
"if the line has an even num of unescaped "'s then we can assume that
|
|
"any given " is not a comment delimiter
|
|
if s:IsNumEven(s:CountNonESCedOccurances(a:line, "\"", "\\"))
|
|
return 0
|
|
endif
|
|
endif
|
|
|
|
return 1
|
|
|
|
endfunction
|
|
|
|
" Function: s:IsNumEven(num) {{{2
|
|
" A small function the returns 1 if the input number is even and 0 otherwise
|
|
" Args:
|
|
" -num: the number to check
|
|
function s:IsNumEven(num)
|
|
return (a:num % 2) == 0
|
|
endfunction
|
|
|
|
" Function: s:IsEscaped(str, indx, escChar) {{{2
|
|
" This function takes a string, an index into that string and an esc char and
|
|
" returns 1 if the char at the index is escaped (i.e if it is preceded by an
|
|
" odd number of esc chars)
|
|
" Args:
|
|
" -str: the string to check
|
|
" -indx: the index into str that we want to check
|
|
" -escChar: the escape char the char at indx may be ESCed with
|
|
function s:IsEscaped(str, indx, escChar)
|
|
"initialise numEscChars to 0 and look at the char before indx
|
|
let numEscChars = 0
|
|
let curIndx = a:indx-1
|
|
|
|
"keep going back thru str until we either reach the start of the str or
|
|
"run out of esc chars
|
|
while curIndx >= 0 && strpart(a:str, curIndx, 1) == a:escChar
|
|
|
|
"we have found another esc char so add one to the count and move left
|
|
"one char
|
|
let numEscChars = numEscChars + 1
|
|
let curIndx = curIndx - 1
|
|
|
|
endwhile
|
|
|
|
"if there is an odd num of esc chars directly before the char at indx then
|
|
"the char at indx is escaped
|
|
return !s:IsNumEven(numEscChars)
|
|
endfunction
|
|
|
|
" Function: s:IsInSexyComment(line) {{{2
|
|
" returns 1 if the given line number is part of a sexy comment
|
|
function s:IsInSexyComment(line)
|
|
return !empty(s:FindBoundingLinesOfSexyCom(a:line))
|
|
endfunction
|
|
|
|
" Function: s:IsSexyComment(topline, bottomline) {{{2
|
|
" This function takes in 2 line numbers and returns 1 if the lines between and
|
|
" including the given line numbers are a sexy comment. It returns 0 otherwise.
|
|
" Args:
|
|
" -topline: the line that the possible sexy comment starts on
|
|
" -bottomline: the line that the possible sexy comment stops on
|
|
function s:IsSexyComment(topline, bottomline)
|
|
|
|
"get the delimiter set that would be used for a sexy comment
|
|
let left = ''
|
|
let right = ''
|
|
if s:Multipart()
|
|
let left = s:Left()
|
|
let right = s:Right()
|
|
elseif s:AltMultipart()
|
|
let left = s:Left({'alt': 1})
|
|
let right = s:Right({'alt': 1})
|
|
else
|
|
return 0
|
|
endif
|
|
|
|
"swap the top and bottom line numbers around if need be
|
|
let topline = a:topline
|
|
let bottomline = a:bottomline
|
|
if bottomline < topline
|
|
topline = bottomline
|
|
bottomline = a:topline
|
|
endif
|
|
|
|
"if there is < 2 lines in the comment it cannot be sexy
|
|
if (bottomline - topline) <= 0
|
|
return 0
|
|
endif
|
|
|
|
"if the top line doesn't begin with a left delimiter then the comment isn't sexy
|
|
if getline(a:topline) !~ '^[ \t]*' . left
|
|
return 0
|
|
endif
|
|
|
|
"if there is a right delimiter on the top line then this isn't a sexy comment
|
|
if s:LastIndexOfDelim(right, getline(a:topline)) != -1
|
|
return 0
|
|
endif
|
|
|
|
"if there is a left delimiter on the bottom line then this isn't a sexy comment
|
|
if s:FindDelimiterIndex(left, getline(a:bottomline)) != -1
|
|
return 0
|
|
endif
|
|
|
|
"if the bottom line doesn't begin with a right delimiter then the comment isn't
|
|
"sexy
|
|
if getline(a:bottomline) !~ '^.*' . right . '$'
|
|
return 0
|
|
endif
|
|
|
|
let sexyComMarker = s:GetSexyComMarker(0, 1)
|
|
|
|
"check each of the intermediate lines to make sure they start with a
|
|
"sexyComMarker
|
|
let currentLine = a:topline+1
|
|
while currentLine < a:bottomline
|
|
let theLine = getline(currentLine)
|
|
|
|
if theLine !~ '^[ \t]*' . sexyComMarker
|
|
return 0
|
|
endif
|
|
|
|
"if there is a right delimiter in an intermediate line then the block isn't
|
|
"a sexy comment
|
|
if s:LastIndexOfDelim(right, theLine) != -1
|
|
return 0
|
|
endif
|
|
|
|
let currentLine = currentLine + 1
|
|
endwhile
|
|
|
|
"we have not found anything to suggest that this isn't a sexy comment so
|
|
return 1
|
|
|
|
endfunction
|
|
|
|
" Function: s:LastIndexOfDelim(delim, str) {{{2
|
|
" This function takes a string and a delimiter and returns the last index of
|
|
" that delimiter in string
|
|
" Args:
|
|
" -delim: the delimiter to look for
|
|
" -str: the string to look for delimiter in
|
|
function s:LastIndexOfDelim(delim, str)
|
|
let delim = a:delim
|
|
let lenDelim = strlen(delim)
|
|
|
|
"set index to the first occurrence of delimiter. If there is no occurrence then
|
|
"bail
|
|
let indx = s:FindDelimiterIndex(delim, a:str)
|
|
if indx == -1
|
|
return -1
|
|
endif
|
|
|
|
"keep moving to the next instance of delimiter in str till there is none left
|
|
while 1
|
|
|
|
"search for the next delimiter after the previous one
|
|
let searchStr = strpart(a:str, indx+lenDelim)
|
|
let indx2 = s:FindDelimiterIndex(delim, searchStr)
|
|
|
|
"if we find a delimiter update indx to record the position of it, if we
|
|
"don't find another delimiter then indx is the last one so break out of
|
|
"this loop
|
|
if indx2 != -1
|
|
let indx = indx + indx2 + lenDelim
|
|
else
|
|
break
|
|
endif
|
|
endwhile
|
|
|
|
return indx
|
|
|
|
endfunction
|
|
|
|
" Function: s:Left(...) {{{2
|
|
" returns left delimiter data
|
|
function s:Left(...)
|
|
let params = a:0 ? a:1 : {}
|
|
|
|
let delim = has_key(params, 'alt') ? b:NERDCommenterDelims['leftAlt'] : b:NERDCommenterDelims['left']
|
|
|
|
if delim == ''
|
|
return ''
|
|
endif
|
|
|
|
if has_key(params, 'space') && g:NERDSpaceDelims
|
|
let delim = delim . s:spaceStr
|
|
endif
|
|
|
|
if has_key(params, 'esc')
|
|
let delim = s:Esc(delim)
|
|
endif
|
|
|
|
return delim
|
|
endfunction
|
|
|
|
" Function: s:LeftMostIndx(countCommentedLines, countEmptyLines, topline, bottomline) {{{2
|
|
" This function takes in 2 line numbers and returns the index of the left most
|
|
" char (that is not a space or a tab) on all of these lines.
|
|
" Args:
|
|
" -countCommentedLines: 1 if lines that are commented are to be checked as
|
|
" well. 0 otherwise
|
|
" -countEmptyLines: 1 if empty lines are to be counted in the search
|
|
" -topline: the top line to be checked
|
|
" -bottomline: the bottom line to be checked
|
|
function s:LeftMostIndx(countCommentedLines, countEmptyLines, topline, bottomline)
|
|
|
|
" declare the left most index as an extreme value
|
|
let leftMostIndx = 1000
|
|
|
|
" go thru the block line by line updating leftMostIndx
|
|
let currentLine = a:topline
|
|
while currentLine <= a:bottomline
|
|
|
|
" get the next line and if it is allowed to be commented, or is not
|
|
" commented, check it
|
|
let theLine = getline(currentLine)
|
|
if a:countEmptyLines || theLine !~ '^[ \t]*$'
|
|
if a:countCommentedLines || (!s:IsCommented(s:Left(), s:Right(), theLine) && !s:IsCommented(s:Left({'alt': 1}), s:Right({'alt': 1}), theLine))
|
|
" convert spaces to tabs and get the number of leading spaces for
|
|
" this line and update leftMostIndx if need be
|
|
let theLine = s:ConvertLeadingTabsToSpaces(theLine)
|
|
let leadSpaceOfLine = strlen( substitute(theLine, '\(^[ \t]*\).*$','\1','') )
|
|
if leadSpaceOfLine < leftMostIndx
|
|
let leftMostIndx = leadSpaceOfLine
|
|
endif
|
|
endif
|
|
endif
|
|
|
|
" move on to the next line
|
|
let currentLine = currentLine + 1
|
|
endwhile
|
|
|
|
if leftMostIndx == 1000
|
|
return 0
|
|
else
|
|
return leftMostIndx
|
|
endif
|
|
endfunction
|
|
|
|
" Function: s:Multipart() {{{2
|
|
" returns 1 if the current delimiters are multipart
|
|
function s:Multipart()
|
|
return s:Right() != ''
|
|
endfunction
|
|
|
|
" Function: s:NerdEcho(msg, typeOfMsg) {{{2
|
|
" Args:
|
|
" -msg: the message to echo
|
|
" -typeOfMsg: 0 = warning message
|
|
" 1 = normal message
|
|
function s:NerdEcho(msg, typeOfMsg)
|
|
if a:typeOfMsg == 0
|
|
echohl WarningMsg
|
|
echom 'NERDCommenter:' . a:msg
|
|
echohl None
|
|
elseif a:typeOfMsg == 1
|
|
echom 'NERDCommenter:' . a:msg
|
|
endif
|
|
endfunction
|
|
|
|
" Function: s:Nested() {{{2
|
|
" returns 1 if the current multipart (if any) delimiters allow nesting
|
|
function s:Nested()
|
|
return b:NERDCommenterDelims['nested']
|
|
endfunction
|
|
|
|
" Function: s:NumberOfLeadingTabs(s) {{{2
|
|
" returns the number of leading tabs in the given string
|
|
function s:NumberOfLeadingTabs(s)
|
|
return strlen(substitute(a:s, '^\(\t*\).*$', '\1', ""))
|
|
endfunction
|
|
|
|
" Function: s:NumLinesInBuf() {{{2
|
|
" Returns the number of lines in the current buffer
|
|
function s:NumLinesInBuf()
|
|
return line('$')
|
|
endfunction
|
|
|
|
" Function: s:ReplaceDelims(toReplace1, toReplace2, replacor1, replacor2, str) {{{2
|
|
" This function takes in a string, 2 delimiters in that string and 2 strings
|
|
" to replace these delimiters with.
|
|
"
|
|
" Args:
|
|
" -toReplace1: the first delimiter to replace
|
|
" -toReplace2: the second delimiter to replace
|
|
" -replacor1: the string to replace toReplace1 with
|
|
" -replacor2: the string to replace toReplace2 with
|
|
" -str: the string that the delimiters to be replaced are in
|
|
function s:ReplaceDelims(toReplace1, toReplace2, replacor1, replacor2, str)
|
|
let line = s:ReplaceLeftMostDelim(a:toReplace1, a:replacor1, a:str)
|
|
let line = s:ReplaceRightMostDelim(a:toReplace2, a:replacor2, line)
|
|
return line
|
|
endfunction
|
|
|
|
" Function: s:ReplaceLeftMostDelim(toReplace, replacor, str) {{{2
|
|
" This function takes a string and a delimiter and replaces the left most
|
|
" occurrence of this delimiter in the string with a given string
|
|
"
|
|
" Args:
|
|
" -toReplace: the delimiter in str that is to be replaced
|
|
" -replacor: the string to replace toReplace with
|
|
" -str: the string that contains toReplace
|
|
function s:ReplaceLeftMostDelim(toReplace, replacor, str)
|
|
let toReplace = a:toReplace
|
|
let replacor = a:replacor
|
|
"get the left most occurrence of toReplace
|
|
let indxToReplace = s:FindDelimiterIndex(toReplace, a:str)
|
|
|
|
"if there IS an occurrence of toReplace in str then replace it and return
|
|
"the resulting string
|
|
if indxToReplace != -1
|
|
let line = strpart(a:str, 0, indxToReplace) . replacor . strpart(a:str, indxToReplace+strlen(toReplace))
|
|
return line
|
|
endif
|
|
|
|
return a:str
|
|
endfunction
|
|
|
|
" Function: s:ReplaceRightMostDelim(toReplace, replacor, str) {{{2
|
|
" This function takes a string and a delimiter and replaces the right most
|
|
" occurrence of this delimiter in the string with a given string
|
|
"
|
|
" Args:
|
|
" -toReplace: the delimiter in str that is to be replaced
|
|
" -replacor: the string to replace toReplace with
|
|
" -str: the string that contains toReplace
|
|
"
|
|
function s:ReplaceRightMostDelim(toReplace, replacor, str)
|
|
let toReplace = a:toReplace
|
|
let replacor = a:replacor
|
|
let lenToReplace = strlen(toReplace)
|
|
|
|
"get the index of the last delimiter in str
|
|
let indxToReplace = s:LastIndexOfDelim(toReplace, a:str)
|
|
|
|
"if there IS a delimiter in str, replace it and return the result
|
|
let line = a:str
|
|
if indxToReplace != -1
|
|
let line = strpart(a:str, 0, indxToReplace) . replacor . strpart(a:str, indxToReplace+strlen(toReplace))
|
|
endif
|
|
return line
|
|
endfunction
|
|
|
|
"FUNCTION: s:RestoreScreenState() {{{2
|
|
"
|
|
"Sets the screen state back to what it was when s:SaveScreenState was last
|
|
"called.
|
|
"
|
|
function s:RestoreScreenState()
|
|
if !exists("t:NERDComOldTopLine") || !exists("t:NERDComOldPos")
|
|
throw 'NERDCommenter exception: cannot restore screen'
|
|
endif
|
|
|
|
call cursor(t:NERDComOldTopLine, 0)
|
|
normal! zt
|
|
call setpos(".", t:NERDComOldPos)
|
|
endfunction
|
|
|
|
" Function: s:Right(...) {{{2
|
|
" returns right delimiter data
|
|
function s:Right(...)
|
|
let params = a:0 ? a:1 : {}
|
|
|
|
let delim = has_key(params, 'alt') ? b:NERDCommenterDelims['rightAlt'] : b:NERDCommenterDelims['right']
|
|
|
|
if delim == ''
|
|
return ''
|
|
endif
|
|
|
|
if has_key(params, 'space') && g:NERDSpaceDelims
|
|
let delim = s:spaceStr . delim
|
|
endif
|
|
|
|
if has_key(params, 'esc')
|
|
let delim = s:Esc(delim)
|
|
endif
|
|
|
|
return delim
|
|
endfunction
|
|
|
|
" Function: s:RightMostIndx(countCommentedLines, countEmptyLines, topline, bottomline) {{{2
|
|
" This function takes in 2 line numbers and returns the index of the right most
|
|
" char on all of these lines.
|
|
" Args:
|
|
" -countCommentedLines: 1 if lines that are commented are to be checked as
|
|
" well. 0 otherwise
|
|
" -countEmptyLines: 1 if empty lines are to be counted in the search
|
|
" -topline: the top line to be checked
|
|
" -bottomline: the bottom line to be checked
|
|
function s:RightMostIndx(countCommentedLines, countEmptyLines, topline, bottomline)
|
|
let rightMostIndx = -1
|
|
|
|
" go thru the block line by line updating rightMostIndx
|
|
let currentLine = a:topline
|
|
while currentLine <= a:bottomline
|
|
|
|
" get the next line and see if it is commentable, otherwise it doesn't
|
|
" count
|
|
let theLine = getline(currentLine)
|
|
if a:countEmptyLines || theLine !~ '^[ \t]*$'
|
|
|
|
if a:countCommentedLines || (!s:IsCommented(s:Left(), s:Right(), theLine) && !s:IsCommented(s:Left({'alt': 1}), s:Right({'alt': 1}), theLine))
|
|
|
|
" update rightMostIndx if need be
|
|
let theLine = s:ConvertLeadingTabsToSpaces(theLine)
|
|
let lineLen = strlen(theLine)
|
|
if lineLen > rightMostIndx
|
|
let rightMostIndx = lineLen
|
|
endif
|
|
endif
|
|
endif
|
|
|
|
" move on to the next line
|
|
let currentLine = currentLine + 1
|
|
endwhile
|
|
|
|
return rightMostIndx
|
|
endfunction
|
|
|
|
"FUNCTION: s:SaveScreenState() {{{2
|
|
"Saves the current cursor position in the current buffer and the window
|
|
"scroll position
|
|
function s:SaveScreenState()
|
|
let t:NERDComOldPos = getpos(".")
|
|
let t:NERDComOldTopLine = line("w0")
|
|
endfunction
|
|
|
|
" Function: s:SwapOuterMultiPartDelimsForPlaceHolders(line) {{{2
|
|
" This function takes a line and swaps the outer most multi-part delimiters for
|
|
" place holders
|
|
" Args:
|
|
" -line: the line to swap the delimiters in
|
|
"
|
|
function s:SwapOuterMultiPartDelimsForPlaceHolders(line)
|
|
" find out if the line is commented using normal delimiters and/or
|
|
" alternate ones
|
|
let isCommented = s:IsCommented(s:Left(), s:Right(), a:line)
|
|
let isCommentedAlt = s:IsCommented(s:Left({'alt': 1}), s:Right({'alt': 1}), a:line)
|
|
|
|
let line2 = a:line
|
|
|
|
"if the line is commented and there is a right delimiter, replace
|
|
"the delimiters with place-holders
|
|
if isCommented && s:Multipart()
|
|
let line2 = s:ReplaceDelims(s:Left(), s:Right(), g:NERDLPlace, g:NERDRPlace, a:line)
|
|
|
|
"similarly if the line is commented with the alternative
|
|
"delimiters
|
|
elseif isCommentedAlt && s:AltMultipart()
|
|
let line2 = s:ReplaceDelims(s:Left({'alt': 1}), s:Right({'alt': 1}), g:NERDLPlace, g:NERDRPlace, a:line)
|
|
endif
|
|
|
|
return line2
|
|
endfunction
|
|
|
|
" Function: s:SwapOuterPlaceHoldersForMultiPartDelims(line) {{{2
|
|
" This function takes a line and swaps the outermost place holders for
|
|
" multi-part delimiters
|
|
" Args:
|
|
" -line: the line to swap the delimiters in
|
|
"
|
|
function s:SwapOuterPlaceHoldersForMultiPartDelims(line)
|
|
let left = ''
|
|
let right = ''
|
|
if s:Multipart()
|
|
let left = s:Left()
|
|
let right = s:Right()
|
|
elseif s:AltMultipart()
|
|
let left = s:Left({'alt': 1})
|
|
let right = s:Right({'alt': 1})
|
|
endif
|
|
|
|
let line = s:ReplaceDelims(g:NERDLPlace, g:NERDRPlace, left, right, a:line)
|
|
return line
|
|
endfunction
|
|
" Function: s:TabbedCol(line, col) {{{2
|
|
" Gets the col number for given line and existing col number. The new col
|
|
" number is the col number when all leading spaces are converted to tabs
|
|
" Args:
|
|
" -line:the line to get the rel col for
|
|
" -col: the abs col
|
|
function s:TabbedCol(line, col)
|
|
let lineTruncated = strpart(a:line, 0, a:col)
|
|
let lineSpacesToTabs = substitute(lineTruncated, s:TabSpace(), '\t', 'g')
|
|
return strlen(lineSpacesToTabs)
|
|
endfunction
|
|
"FUNCTION: s:TabSpace() {{{2
|
|
"returns a string of spaces equal in length to &tabstop
|
|
function s:TabSpace()
|
|
let tabSpace = ""
|
|
let spacesPerTab = &tabstop
|
|
while spacesPerTab > 0
|
|
let tabSpace = tabSpace . " "
|
|
let spacesPerTab = spacesPerTab - 1
|
|
endwhile
|
|
return tabSpace
|
|
endfunction
|
|
|
|
" Function: s:UnEsc(str, escChar) {{{2
|
|
" This function removes all the escape chars from a string
|
|
" Args:
|
|
" -str: the string to remove esc chars from
|
|
" -escChar: the escape char to be removed
|
|
function s:UnEsc(str, escChar)
|
|
return substitute(a:str, a:escChar, "", "g")
|
|
endfunction
|
|
|
|
" Function: s:UntabbedCol(line, col) {{{2
|
|
" Takes a line and a col and returns the absolute column of col taking into
|
|
" account that a tab is worth 3 or 4 (or whatever) spaces.
|
|
" Args:
|
|
" -line:the line to get the abs col for
|
|
" -col: the col that doesn't take into account tabs
|
|
function s:UntabbedCol(line, col)
|
|
let lineTruncated = strpart(a:line, 0, a:col)
|
|
let lineTabsToSpaces = substitute(lineTruncated, '\t', s:TabSpace(), 'g')
|
|
return strlen(lineTabsToSpaces)
|
|
endfunction
|
|
" Section: Comment mapping and menu item setup {{{1
|
|
" ===========================================================================
|
|
|
|
" Create menu items for the specified modes. If a:combo is not empty, then
|
|
" also define mappings and show a:combo in the menu items.
|
|
function! s:CreateMaps(modes, target, desc, combo)
|
|
" Build up a map command like
|
|
" 'noremap <silent> <plug>NERDCommenterComment :call NERDComment("n", "Comment")'
|
|
let plug = '<plug>NERDCommenter' . a:target
|
|
let plug_start = 'noremap <silent> ' . plug . ' :call NERDComment("'
|
|
let plug_end = '", "' . a:target . '")<cr>'
|
|
" Build up a menu command like
|
|
" 'menu <silent> comment.Comment<Tab>\\cc <plug>NERDCommenterComment'
|
|
let menuRoot = get(['', 'comment', '&comment', '&Plugin.&comment'],
|
|
\ g:NERDMenuMode, '')
|
|
let menu_command = 'menu <silent> ' . menuRoot . '.' . escape(a:desc, ' ')
|
|
if strlen(a:combo)
|
|
let leader = exists('g:mapleader') ? g:mapleader : '\'
|
|
let menu_command .= '<Tab>' . escape(leader, '\') . a:combo
|
|
endif
|
|
let menu_command .= ' ' . (strlen(a:combo) ? plug : a:target)
|
|
" Execute the commands built above for each requested mode.
|
|
for mode in (a:modes == '') ? [''] : split(a:modes, '\zs')
|
|
if strlen(a:combo)
|
|
execute mode . plug_start . mode . plug_end
|
|
if g:NERDCreateDefaultMappings && !hasmapto(plug, mode)
|
|
execute mode . 'map <leader>' . a:combo . ' ' . plug
|
|
endif
|
|
endif
|
|
" Check if the user wants the menu to be displayed.
|
|
if g:NERDMenuMode != 0
|
|
execute mode . menu_command
|
|
endif
|
|
endfor
|
|
endfunction
|
|
call s:CreateMaps('nx', 'Comment', 'Comment', 'cc')
|
|
call s:CreateMaps('nx', 'Toggle', 'Toggle', 'c<space>')
|
|
call s:CreateMaps('nx', 'Minimal', 'Minimal', 'cm')
|
|
call s:CreateMaps('nx', 'Nested', 'Nested', 'cn')
|
|
call s:CreateMaps('n', 'ToEOL', 'To EOL', 'c$')
|
|
call s:CreateMaps('nx', 'Invert', 'Invert', 'ci')
|
|
call s:CreateMaps('nx', 'Sexy', 'Sexy', 'cs')
|
|
call s:CreateMaps('nx', 'Yank', 'Yank then comment', 'cy')
|
|
call s:CreateMaps('n', 'Append', 'Append', 'cA')
|
|
call s:CreateMaps('', ':', '-Sep-', '')
|
|
call s:CreateMaps('nx', 'AlignLeft', 'Left aligned', 'cl')
|
|
call s:CreateMaps('nx', 'AlignBoth', 'Left and right aligned', 'cb')
|
|
call s:CreateMaps('', ':', '-Sep2-', '')
|
|
call s:CreateMaps('nx', 'Uncomment', 'Uncomment', 'cu')
|
|
call s:CreateMaps('n', 'AltDelims', 'Switch Delimiters', 'ca')
|
|
call s:CreateMaps('i', 'Insert', 'Insert Comment Here', '')
|
|
call s:CreateMaps('', ':', '-Sep3-', '')
|
|
call s:CreateMaps('', ':help NERDCommenterContents<CR>', 'Help', '')
|
|
|
|
inoremap <silent> <plug>NERDCommenterInsert <SPACE><BS><ESC>:call NERDComment('i', "insert")<CR>
|
|
|
|
" switch to/from alternative delimiters (does not use wrapper function)
|
|
nnoremap <plug>NERDCommenterAltDelims :call <SID>SwitchToAlternativeDelimiters(1)<cr>
|
|
|
|
" This is a workaround to enable lazy-loading from supported plugin managers:
|
|
" See https://github.com/scrooloose/nerdcommenter/issues/176
|
|
call s:SetUpForNewFiletype(&filetype, 1)
|
|
|
|
" vim: set foldmethod=marker :
|