created October 14, 2010 · complexity basic · author Fritzophrenic · version 7.0
If you frequently work with files saved with mixed line-ending styles, so that Vim incorrectly detects the fileformat, you might wish Vim could automatically read the file in the correct format without requiring you to correct the fileformat manually.
This autocmd in your vimrc will do just that. If it finds any occurrences of a carriage return at the end of a line, and the file was read in Unix format, it will automatically re-read the file in DOS format. This will automatically correct the issue for good when you save the file, but it is most useful for times when you cannot save the file for some reason (maybe it is in a code module that is "locked down" for example).
autocmd BufReadPost * nested \ if !exists('b:reload_dos') && !&binary && &ff=='unix' && (0 < search('\r$', 'nc')) | \ let b:reload_dos = 1 | \ e ++ff=dos | \ endif
Most of this is self-explanatory, but there are a couple tricks to it:
- The 'n' flag is used on the search, so that the cursor does not move.
- The autocmd is nested. This allows other autocmds, like those that load syntax highlighting, etc., to also fire on the reload.
- Because the autocmd is nested, it will itself fire again on the reload. Thus, we use the
b:reload_dos
variable to detect when it has already fired so we do not keep firing recursively.
This can be improved a bit. It might be nice to have a warning of some sort when this auto-reload occurs. Or maybe you want to change as few lines as possible, so you can manually convert the fileformat to the one that changes the fewest lines (for example, if only 2 lines out of 12000 have DOS-style endings, why convert the whole file to DOS?). Also, the detection of DOS-format endings uses a search, which could take a very long time on very large files.
The following version (when possible) limits the length of time allowed for the search before giving up, and additionally gives a message to the user by redirecting the output of an :s
command with the 'n' flag to count matches without making changes and parsing it to output a message like, "File has 2 DOS-style lined endings out of 12000 lines."
" When loading a file, if it reads in as Unix, but has a DOS line ending, " and is not in binary mode, reload it in DOS format. Do this AFTER loading " last known position to prevent always opening on last line. " " Time out the search after 1/10 second. Timeouts only available in 7.1.211 " and up, so just risk long loads in earlier versions. if (v:version > 701 || v:version == 701 && has("patch211")) autocmd BufReadPost * nested \ if !exists('b:reload_dos') && !&binary && &ff=='unix' && (0 < search('\r$', 'nc', 0, 100)) | \ let b:reload_dos = 1 | \ redir => s:num_dos_ends | \ silent %s#\r$##n | \ redir END | \ e ++ff=dos | \ echomsg "File has ". \ substitute(s:num_dos_ends, '^\_.\{-}\(\d\+\).*', '\1', ''). \ " DOS-style line endings out of ".line('$')." lines." | \ endif else autocmd BufReadPost * nested \ if !exists('b:reload_dos') && !&binary && &ff=='unix' && (0 < search('\r$', 'nc')) | \ let b:reload_dos = 1 | \ e ++ff=dos | \ endif endif autocmd BufReadPost * if exists('b:reload_dos') | unlet b:reload_dos | endif
Caveats[]
- It is not trivial to make this work alongside plugins which also reload the buffer. See comments below.
- If you use a plugin which relies on byte counts within a file (like eclim), buffers reloaded in this way will no longer match the byte counts in the file on disk, so any communication about the byte position can easily fail.
Comments[]
To set the 'fileformat'
to whatever will change the fewer lines, replace the echomsg
command in the above snippet by the following:
\ let s:l_tot = line('$') | \ let s:l_dos = 0 + s:num_dos_ends | \ echomsg s:l_dos 'DOS-style line endings found in' s:l_tot 'lines' | \ if &noro && &ma | \ if s:l_tot > s:l_dos * 2 | \ setlocal ff=unix | \ echomsg 'File format set to UNIX' | \ else | \ echomsg 'File format set to DOS' | \ endif | \ endif |
(the next statement is the final endif
).
Notes:
- This applies to Vim 7.1.211 or later; how to do it for earlier versions too is left as an exercise to the student.
- The Unix/Dos switch is bypassed if the file is 'readonly' or 'nomodifiable' at the time of the BufReadPost event. Either option can be set afterwards by a modeline.
- This takes advantage of the fact that adding zero to a String gives the Number found at the start of the string (or zero if the string starts with a non-numeric character or is empty) — see (of all places) :help octal
— Tonymec 04:41, September 21, 2011 (UTC)
Doesn't work for me. It looks like you're trying to cleverly auto-convert the first number in the substitute string to a number, but when I try that it always comes out with zero, because the string contains something like:
^@^@3 matches on 3 lines^@3 matches on 3 lines^@
Using the substitute call used in the main script seems to fix it, that's probably why it was there in the first place.
- ah, strange. I didn't check and thus noticed neither the nulls nor the double-printing. Well, let's say I tried to be too clever and got my feet caught in the carpet. — Tonymec 07:31, September 23, 2011 (UTC)
I have &ff shown in my statusline so I didn't need the additional echo about the decision made, and I prefer ignoring &modifiable/&readonly and instead setting nomodified after converting back to Unix (not technically true, but close enough to being true for the reasons I use this script).
It was non-trivial to make all this work well with plugins which also reload the buffer, like AutoFenc! Here's what I ended up with, some of it can probably be merged:
" When loading a file, if it reads in as Unix, but has a DOS line ending, " and is not in binary mode, reload it in DOS format. Do this AFTER loading " last known position to prevent always opening on last line. " " Time out the search after 1/10 second. Timeouts only available in 7.1.211 " and up, so just risk long loads in Unix. if (v:version > 701 || v:version == 701 && has("patch211")) autocmd BufReadPost * nested \ if !exists('b:reload_DOS') && !exists('b:nested_DOS_reload') && !&binary && &ff=='unix' && (0 < search('\r$', 'n', 0, 100)) | \ redir => s:num_dos_ends | \ silent %s#\r$##n | \ redir END | \ let b:reload_DOS = 1 | \ let b:nested_DOS_reload = 1 | \ let s:l_tot = line('$') | \ let s:l_dos = substitute(s:num_dos_ends, '^\_.\{-}\(\d\+\).*', '\1', '') | \ echomsg s:l_dos 'DOS-style line endings found in' s:l_tot 'lines' | \ e ++ff=dos | \ unlet b:reload_DOS | \ endif else autocmd BufReadPost * nested \ if !exists('b:reload_DOS') && !&binary && &ff=='unix' && (0 < search('\r$', 'nc')) | \ let b:reload_DOS = 1 | \ let b_nested_DOS_reload = 1 | \ e ++ff=dos | \ unlet b:reload_DOS | \ endif endif " Only unlet the stopper variable after all autocmds (including in plugins) " have fired which may want to reload the entire buffer (hopefully). " " Also set the ff to unix here if appropriate since anything reloading the " file has no way to know the correct ff except for what it has already been " set to. if has('gui_running') let s:event = 'GUIEnter' else let s:event = 'VimEnter' endif exec 'autocmd '.s:event.' * autocmd file_handling BufReadPost * '. \ 'if !exists("b:reload_DOS") |'. \ 'if exists("b:nested_DOS_reload") |'. \ 'unlet b:nested_DOS_reload |'. \ 'if exists("s:l_tot") && s:l_tot > s:l_dos * 2 |'. \ ' let s:dos_mod_sav = &l:modified |'. \ ' setlocal ff=unix |'. \ ' let &l:modified=s:dos_mod_sav |'. \ 'endif |'. \ 'endif |'. \ 'endif'