[darcs-users] Windows end-of-line characters

[Peter Strand]

Peter Maxwell wrote:
> This issue has been causing me problems.  Here is what I think I know:
> 
> Windows text file lines are terminated with CR LF, unix with just LF.
> 
> Most version control software automatically (sometimes optionaly) 
> converts text files to have native EOLs, so the same file will have CRLF 
> on windows and LF on unix.  GPG does something similar, in some modes.  
> So do email clients, web browsers etc.
> 
> darcs doesn't yet have a policy on end-of-line characters.  Current 
> behaviour is mostly to not do any such conversions.
...
> OK, that gets us to a consistent absence of EOL conversion.  Now, 
> thinking about what it would take to have deliberate EOL conversion:
> 
> I have a script that unpacks darcs patches emailed from windows, and as 
> a final step strips out any CR characters before passing it on to darcs 
> to apply, but patches more often travel by protocols that don't let a 
> script get a look in (scp and http (not as text/plain so no EOL 
> conversion)) so either patches should be always be stored with LF line 
> endings, even on windows, or darcs needs to convert either type to 
> native EOLs on any platform.

I think it is best to keep patches and metadata in a canonical format,
using LF as it is now, and do the conversion when darcs modifies files
in the working copy. Perhaps we could have a new category of files, in
addition to "boring" and "binaries", to specify that they should be
converted to or from the native text format when written or read?


/Peter