Re: format failure on diff --side-by-side

[Mark D. Baushke]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Frank Hemer <frank@hemer.org> writes:

> > Looking at the problem space, I don't see any way that cvs can do
> > anything to fix your problem without hacking on the diffutils sources
> > and we are trying to avoid forking diffutils if at all possible.
> 
> I absolutely understand that, however see below -
> 
> > I see the same behavior from the
> > ftp://alpha.gnu.org/gnu/diffutils/diffutils-2.8.7.tar.gz package that
> > was released 2004-04-13 as I see in the current cvs top-of-tree.
> 
> I cannot reproduce this using the above mentioned sources. After downloading 
> and compiling, I used the following cmd:
> diff --side-by-side -t -W100 file1.txt file2.txt
> where file1 represents rev1.1 and file2 rev.1.2
> The output is exactly what I expect:
> 
> 
> The cvs output however is different:-(

Yes, the cvs output is based on a captive version of GNU diffutils 2.7
and it seems that 2.7.2 is the first version for which I have sources
after 2.7 that does what you expect if you set the LC_* variables
properly (unlike what I did initially when I tested 2.8.7 of diffutils).

> > I suggest you report your problem to the bug-gnu-utils@gnu.org mailing
> > list to see if they can help you. If they can, then cvs will do our best
> > to refresh to a new release of diffutils on the feature branch.
> 
> I don't actually see what I should report. IMHO diffutils work as
> expected. Do I miss something?

No, I did. My initial test managed to set the variables, but not export
them to the child processes and so I got the same output with cvs as
with a newer 'diff' command. However, setting the LC_* variables,
exporting them and running the newer 'diff' command get the results you
expect.

So, one thing we probably need to schedule is upgrading the captive
version of diff to a 2.8.x version (where x==7 is the latest alpha
release from 2004-04-13 and x==1 is the latest official release from
2002-04-06).

However, this is still going to be a problem in client/server mode as
nowhere does the current FEATURE version of CVS call setlocale() and it
is certainly not sending that information from the client to the server
at present.

        -- Mark
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (FreeBSD)

iD8DBQFBqj6H3x41pRYZE/gRAn86AKCmhFaIGxGAkYTk/2hUv1n4WHvq2gCgjyvW
XbiwkSxBJu1eQJl1PZn2wsU=
=M1Pk
-----END PGP SIGNATURE-----