v2.37.0-rc2 - Additional Breakages on NonStop x86 (was RE: [ANNOUNCE] Git v2.37.0-rc2)

v2.37.0-rc2 - Additional Breakages on NonStop x86 (was RE: [ANNOUNCE] Git v2.37.0-rc2)

[rsbecker]

On June 22, 2022 3:32 PM, Junio C Hamano wrote:
>A release candidate Git v2.37.0-rc2 is now available for testing at the usual places.
>It is comprised of 378 non-merge commits since v2.36.0, contributed by 64 people,
>19 of which are new faces [*].

RC2 had numerous new test failures on NonStop that were not present in RC1, RC0, or prior git releases, ignoring the @2 problem previously reported. Failures are:

t5516.113
t5601.8
t7031.3, 11
t7421 (all subtests)
t7500.46-50
t7528.4,5,8,15,16,20

This does require additional investigation as to what is happening, but as it stands, I do not think we can add this to our official package set.

I am sorry,
Randall

Re: v2.37.0-rc2 - Additional Breakages on NonStop x86 (was RE: [ANNOUNCE] Git v2.37.0-rc2)

[René Scharfe]

Am 23.06.22 um 14:55 schrieb rsbecker@nexbridge.com:
> On June 22, 2022 3:32 PM, Junio C Hamano wrote:
>> A release candidate Git v2.37.0-rc2 is now available for testing at the usual places.
>> It is comprised of 378 non-merge commits since v2.36.0, contributed by 64 people,
>> 19 of which are new faces [*].
>
> RC2 had numerous new test failures on NonStop that were not present in RC1, RC0, or prior git releases, ignoring the @2 problem previously reported. Failures are:
>
> t5516.113
> t5601.8
> t7031.3, 11
> t7421 (all subtests)
> t7500.46-50
> t7528.4,5,8,15,16,20

That's curious.  Here are the test names:

   t5516-fetch-push.sh
      113 - fetch warns or fails when using username:password
   t5601-clone.sh
      8 - clone warns or fails when using username:password
   t7031-verify-tag-signed-ssh.sh
      3 - verify and show ssh signatures
      11 - verify multiple tags ssh
   t7421-submodule-summary-add.sh
      1 - summary test environment setup
      2 - submodule summary output for initialized submodule
      3 - submodule summary output for deinitialized submodule
      4 - submodule summary output for submodules with changed paths
   t7500-commit-template-squash-signoff.sh
      46 - commit --squash works with -c
      47 - commit --squash works with -C for same commit
      48 - commit --squash works with -c for same commit
      49 - commit --squash works with editor
      50 - invalid message options when using --fixup
   t7528-signed-commit-ssh.sh
      4 - verify and show signatures
      5 - verify-commit exits failure on untrusted signature
      8 - verify-commit succeeds with commit date and key validity matching
      15 - show signed commit with signature
      16 - detect fudged signature
      20 - show bad signature with custom format

rc2 changed the following files:

   $ git diff --stat v2.37.0-rc1 v2.37.0-rc2
    Documentation/git-gc.txt         |  3 +--
    GIT-VERSION-GEN                  |  2 +-
    builtin/name-rev.c               |  2 +-
    contrib/completion/git-prompt.sh | 32 +++++++++++++++++---------------
    t/t9903-bash-prompt.sh           |  2 +-
    5 files changed, 21 insertions(+), 20 deletions(-)

The name-rev change only modifies a help message.  At first glance there
doesn't seem to be a way that these changes may influence the outcome of
those tests.

René

RE: v2.37.0-rc2 - Additional Breakages on NonStop x86 (was RE: [ANNOUNCE] Git v2.37.0-rc2)

[rsbecker]

On June 23, 2022 11:22 AM, René Scharfe wrote:
>Am 23.06.22 um 14:55 schrieb rsbecker@nexbridge.com:
>> On June 22, 2022 3:32 PM, Junio C Hamano wrote:
>>> A release candidate Git v2.37.0-rc2 is now available for testing at the usual
>places.
>>> It is comprised of 378 non-merge commits since v2.36.0, contributed
>>> by 64 people,
>>> 19 of which are new faces [*].
>>
>> RC2 had numerous new test failures on NonStop that were not present in RC1,
>RC0, or prior git releases, ignoring the @2 problem previously reported. Failures
>are:
>>
>> t5516.113
>> t5601.8
>> t7031.3, 11
>> t7421 (all subtests)
>> t7500.46-50
>> t7528.4,5,8,15,16,20
>
>That's curious.  Here are the test names:
>
>   t5516-fetch-push.sh
>      113 - fetch warns or fails when using username:password
>   t5601-clone.sh
>      8 - clone warns or fails when using username:password
>   t7031-verify-tag-signed-ssh.sh
>      3 - verify and show ssh signatures
>      11 - verify multiple tags ssh
>   t7421-submodule-summary-add.sh
>      1 - summary test environment setup
>      2 - submodule summary output for initialized submodule
>      3 - submodule summary output for deinitialized submodule
>      4 - submodule summary output for submodules with changed paths
>   t7500-commit-template-squash-signoff.sh
>      46 - commit --squash works with -c
>      47 - commit --squash works with -C for same commit
>      48 - commit --squash works with -c for same commit
>      49 - commit --squash works with editor
>      50 - invalid message options when using --fixup
>   t7528-signed-commit-ssh.sh
>      4 - verify and show signatures
>      5 - verify-commit exits failure on untrusted signature
>      8 - verify-commit succeeds with commit date and key validity matching
>      15 - show signed commit with signature
>      16 - detect fudged signature
>      20 - show bad signature with custom format
>
>rc2 changed the following files:
>
>   $ git diff --stat v2.37.0-rc1 v2.37.0-rc2
>    Documentation/git-gc.txt         |  3 +--
>    GIT-VERSION-GEN                  |  2 +-
>    builtin/name-rev.c               |  2 +-
>    contrib/completion/git-prompt.sh | 32 +++++++++++++++++---------------
>    t/t9903-bash-prompt.sh           |  2 +-
>    5 files changed, 21 insertions(+), 20 deletions(-)
>
>The name-rev change only modifies a help message.  At first glance there doesn't
>seem to be a way that these changes may influence the outcome of those tests.

Just to confirm, this is commit f770e9f39, correct?

I found a few related issues here. The biggest one is that we use Jenkins for our build. It looks like the TTY determinations are failing when in Jenkins but not outside. So an inaccessible stdin is causing test issues. Also when stderr=stdout is also causing issues. For t7031, there is a library conflict when we run with OpenSSH 8 with Git built with OpenSSL 1.1.1 vs. 3.0.4 - so that conflict is a configuration issue. We do all kinds of built/tests with those, so that explains that.

So other than t7031, the tests ran correctly when I did them manually, but not in Jenkins.

Re: v2.37.0-rc2 - Additional Breakages on NonStop x86 (was RE: [ANNOUNCE] Git v2.37.0-rc2)

[Junio C Hamano]

René Scharfe <l.s.r@web.de> writes:

> Am 23.06.22 um 14:55 schrieb rsbecker@nexbridge.com:
>> RC2 had numerous new test failures on NonStop that were not
>> present in RC1, RC0, or prior git releases, ignoring the @2
>> problem previously reported.

I am wondering if "@2 problem" is the only thing that were present
in -rc1 but was only noticed in -rc2, which presumably is because
some other changes in the reporter's build and test environment.

> rc2 changed the following files:
>
>    $ git diff --stat v2.37.0-rc1 v2.37.0-rc2
>     Documentation/git-gc.txt         |  3 +--
>     GIT-VERSION-GEN                  |  2 +-
>     builtin/name-rev.c               |  2 +-
>     contrib/completion/git-prompt.sh | 32 +++++++++++++++++---------------
>     t/t9903-bash-prompt.sh           |  2 +-
>     5 files changed, 21 insertions(+), 20 deletions(-)
>
> The name-rev change only modifies a help message.  At first glance there
> doesn't seem to be a way that these changes may influence the outcome of
> those tests.

Thanks for a thought.