Re: git failure on Solaris t3701-add-interactive.sh -- git version 5cc8f372509298d13632d8784bc851a587937550

Peff,

Thanks for looking.  I'm sure you have been busy preparing for the next release.

Sorry for the list omission on the last email, it was an oversight.

I just did the test suite on origin/next, so this test run was done
with git b0a67ea72573692994090d92e300cddaf20ac69d.

and here is the list of tests that are failing:

#       deleted:    t/t3701-add-interactive.sh
#       deleted:    t/t3900-i18n-commit.sh
#       deleted:    t/t3901-i18n-patch.sh
#       deleted:    t/t4116-apply-reverse.sh
#       deleted:    t/t4200-rerere.sh
#       deleted:    t/t5000-tar-tree.sh
#       deleted:    t/t7400-submodule-basic.sh
#       deleted:    t/t7401-submodule-summary.sh
#       deleted:    t/t7501-commit.sh
#       deleted:    t/t9001-send-email.sh

I'll send -v -i details on the ones that fail for me but not for you.

-Whit



On Mon, Mar 31, 2008 at 3:27 AM, Jeff King [off-list ref] wrote:

On Wed, Mar 26, 2008 at 04:57:38PM -0400, Whit Armstrong wrote:

quoted

Peff,

[I think the list might benefit from our discussion; did you mean to
respond off-list?]

Sorry for the slow response...I haven't had much time to look at this
stuff lately. This is an "off the top of my head" response; a lot of my
points will need followup patches and testing.

quoted

I have failures with the following tests on our Solaris setup.
However, I did use a more recent version of perl (5.10 installed in my
home dir) with these tests.

#       deleted:    t/t3900-i18n-commit.sh
#       deleted:    t/t3901-i18n-patch.sh

I think something is lacking in the Japanese charset support on Solaris.
Maybe there is some extra package that needs to be installed. I haven't
looked much further.

quoted

#       deleted:    t/t4116-apply-reverse.sh

This is due to a warning from Solaris tar, which doesn't like the pax
field in our generated tarfile. I think it's just being used to make a
copy of the tree, so we can probably get around it by using
git-checkout-index directly. More portable, and should be faster.

quoted

#       deleted:    t/t4118-apply-empty-context.sh

I don't have this in my list of broken tests. Might just be an
oversight on my part, though.

quoted

#       deleted:    t/t4200-rerere.sh

This definitely passed for me, since I submitted fixes. But there was
much discussion afterwards, so one of the followups might have re-broken
it.

quoted

#       deleted:    t/t8001-annotate.sh

I'm pretty sure this passed for me.

quoted

#       deleted:    t/t9001-send-email.sh

I couldn't even try this because of the perl version issues.

quoted

this is using origin/pu -- commit 9606c1c61e3a5b5acbb3f3dcccea37e00bcca145

Yikes. I would stick with 'next' at the most. Junio makes sure all tests
work (on Linux) on 'master', and I think 'next'. But 'pu' is a no-man's
land of quality.

quoted

Let me know if you want test.sh -v -i output for these.

For the ones that I marked as "passed", sure. For the other ones, either
we have the same failure, in which case I can generate it myself, or we
don't, in which case I need to fix my bug before your bug will make any
sense to me. :)

quoted

Also, is there a way to ask "make test" to continue the test suite
from a certain point?  I kept deleting the test that was failing, and
then restarting the whole "make test," which I'm sure you know is
quite long to run from start to finish.

No, there isn't. During my test-fixing spree, I just did something like:

 make test
 # oops, we failed in 3900; continue after
 make test t390[123]*.sh t4*.sh t5*.sh ...

and so forth, shrinking my commandline each time. Hack-ish, but what
we're doing is probably not worth implementing the 'resume at last
failure' feature (especially because most of the time you are fixing
_git_, not the test scripts, so it is a good idea to run the whole
suite).

What I'd really like is the ability to run the test scripts in parallel;
test-lib.sh would need to put each in its own directory.

-Peff