threads / rfc / 6956
Subject: [PATCH/RFC 0/8] use "git diff" in the test suite
Johannes SchindelinHi,
Now that --no-index is supported by git's diff, we can use much of the test suite to do the job, instead of relying on GNU diff.
This is a bit work-in-progress, since the handling of "-" as a filename is not fleshed out yet, and we really should have the "--quiet" option. But I got side-tracked by the edit-patch-series script, and did not want to hold these patches off any longer.
Ciao, Dscho
Johannes SchindelinRe: [PATCH/RFC 0/8] use "git diff" in the test suite
Hi,
somehow, patch 7/8 did not come through, I guess, so here is how to do it yourself:
$ perl -pi.bak -e "s/(^\s*|'|&&\s*)cmp -s /\1git diff /" *.sh
We should probably add a "--quiet" flag for the diff family first, and use that...
Ciao, Dscho
Junio C HamanoRe: [PATCH/RFC 0/8] use "git diff" in the test suite
Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:
> Now that --no-index is supported by git's diff, we can use much of the > test suite to do the job, instead of relying on GNU diff. > > This is a bit work-in-progress, since the handling of "-" as a filename is > not fleshed out yet, and we really should have the "--quiet" option. But I > got side-tracked by the edit-patch-series script, and did not want to hold > these patches off any longer.
As you said yourself [6/8], [7/8], and [8/8] are mechanical conversions, and I do not think it is too much effort involved on your side to "hold these patches off".
My preference is to apply [1/8] now, cook the result in 'next' for a few days longer to see if anybody complains or finds glitches in the diff-ni series so far, push it out to 'master' and then apply [2/8] and [3/8] while on 'master'. The remainder would be forked into a topic branch and cooked in 'pu' and then 'next' as usual.
Also I do not think we would necessarily want [7/8] and [8/8]. It would help with tests on broken distributions whose 'cmp' misbehaves when the output is redirected to /dev/null, but I do not think it is a good enough reason to make the tests slower.
Johannes SchindelinRe: [PATCH/RFC 0/8] use "git diff" in the test suite
Hi,
On Sun, 25 Feb 2007, Junio C Hamano wrote:
> Johannes Schindelin <Johannes.Schindelin@gmx.de> writes: > > > Now that --no-index is supported by git's diff, we can use much of the > > test suite to do the job, instead of relying on GNU diff. > > > > This is a bit work-in-progress, since the handling of "-" as a > > filename is not fleshed out yet, and we really should have the > > "--quiet" option. But I got side-tracked by the edit-patch-series > > script, and did not want to hold these patches off any longer. > > As you said yourself [6/8], [7/8], and [8/8] are mechanical conversions, > and I do not think it is too much effort involved on your side to "hold > these patches off". > > My preference is to apply [1/8] now, cook the result in 'next' for a few > days longer to see if anybody complains or finds glitches in the diff-ni > series so far, push it out to 'master' and then apply [2/8] and [3/8] > while on 'master'. The remainder would be forked into a topic branch > and cooked in 'pu' and then 'next' as usual.
I fully agree.
BTW I think that using 020600 as mode for "-" when it means stdin would fix [4/8]...
> Also I do not think we would necessarily want [7/8] and [8/8]. It would > help with tests on broken distributions whose 'cmp' misbehaves when the > output is redirected to /dev/null, but I do not think it is a good > enough reason to make the tests slower.
I think with the upcoming "--quiet", which exits early on error, performance should be about par. Actually, I went off in this direction because I still use an OpenSUSE 10.1 install DVD, which has this broken cmp...
But I do not care deeply. (It was a good test case for diff-ni, though...)
Ciao, Dscho