Patch: Accept CR/LF line breaks in exclude files

[Dave Dykstra]

On Thu, Feb 21, 2002 at 12:02:34PM -0800, Martin Pool wrote:
> On 21 Feb 2002, Dave Dykstra <dwd at bell-labs.com> wrote:
> 
> > On Thu, Feb 21, 2002 at 11:10:50AM -0600, Dave Dykstra wrote:
> > > On Tue, Feb 19, 2002 at 04:30:53PM -0600, Jordan Russell wrote:
> > > > use CR/LF (DOS-style) line breaks. It basically behaves as if the files are
> > > > empty. This is rather annoying if the exclude files are generated by Windows
> > > > systems (e.g. if you're using --cvs-exclude and rsync'ing the contents of a
> > > > Samba share that includes .cvsignore files).
> > > > 
> > > > Attached is a simple 1-line patch (for 2.5.2) which "fixes" this; it strips
> > > > off any CR character at the end of the line. Could it be committed?
> > > > 
> > > > Jordan Russell
> > > 
> > > Are you talking about on Cygwin or on some native unix?  I'm not sure it
> > > makes sense to modify rsync to strip off carriage returns from every file
> > > it reads.  On the other hand, that makes more sense than doing it for just
> > > exclude files.  On Cygwin it should, certainly, but I think the cygwin 
> > > option to handle that automatically has already been enabled.
> 
> > Oops, I should have read through all the messages first.  Rsync is not
> > doing that on Cygwin, and Ville Herva's patch addresses that.
> 
> There is one scenario where it will not, which is when running on
> Unix, but manipulating .cvsignore or other text files created by a
> Windoze program.  As far as I can see, this will not be handled by
> Ville Herva's good patch, because Unix does nothing special when you
> open a file in text mode.
> 
> (Seeing as we're on Samba.org we kind of ought to handle it :-)
> 
> Obviously files being transferred should always be sent byte-for-byte
> regardless of platform.

Right, I only meant files rsync interprets.

> However, when reading text formatted files for our own use, it is
> probably better to be "generous in what we accept" with regard to line
> endings.  I think we ought to accept a patch that makes rsync agnostic
> about CR, CRLF, or LF at the end of lines in password, config, and
> cvsignore files.  (I think Python now does this on all platforms.)
> 
> It's hard to imagine somebody specifically wanting a CR character in
> any of them to be treated literally.

Sounds reasonable to me, but I don't think it makes sense to fix just one
of them.  Probably there should be a util.c function to read a line from a
file descriptor that strips off any trailing CR or NL.  The cygwin-specific
patch from Ville Herva won't be needed then.

Something for 2.5.4, though; 2.5.3 can't wait.

- Dave