Johannes Schindelin venit, vidit, dixit 2022-06-15 16:50:40:
> Hi Michael,
Hallo Dscho!
> On Tue, 14 Jun 2022, Michael J Gruber wrote:
>
> > 0527ccb1b5 ("add -i: default to the built-in implementation", 2021-11-30)
> > switched to the implementation which fixed to subtest. Mark them as
> > expect_success now.
>
> Good catch!
I'm no list regular anymore, but still a "next+ regular". While
experimenting with my own patch I noticed something got fixed
unexpectedly. That goes to show that these unexpected successes
(from expect_failure) go unnoticed too easily. I had missed this on my
regular rebuilds.
> However... that commit specifically contains this change:
>
> diff --git a/ci/run-build-and-tests.sh b/ci/run-build-and-tests.sh
> index cc62616d806..660ebe8d108 100755
> --- a/ci/run-build-and-tests.sh
> +++ b/ci/run-build-and-tests.sh
> @@ -29,7 +29,7 @@ linux-gcc)
> export GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS=1
> export GIT_TEST_MULTI_PACK_INDEX=1
> export GIT_TEST_MULTI_PACK_INDEX_WRITE_BITMAP=1
> - export GIT_TEST_ADD_I_USE_BUILTIN=1
> + export GIT_TEST_ADD_I_USE_BUILTIN=0
> export GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME=master
> export GIT_TEST_WRITE_REV_INDEX=1
> export GIT_TEST_CHECKOUT_WORKERS=2
>
> The intention is to have t3701 be run with the non-built-in version of
> `git add -i` in the `linux-gcc` job, and I am surprised that those two
> tests do not fail for you in that case.
>
> Did you run this through the CI builds?
That's why I mentioned "no list regular" - I didn't know about that knob
nor the intention to have the test suite run with either implementation
(rather than switching to the new one for good).
I do local builds, usually with
```
DEVELOPER=1 (which I had to disable during the bisect run; gcc12...)
DEFAULT_TEST_TARGET=prove
GIT_PROVE_OPTS=--jobs 4
GIT_TEST_OPTS=--root=/dev/shm/t --chain-lint
SHELL_PATH=/bin/dash
SKIP_DASHED_BUILT_INS=y
```
in config.mak. Nothing else strikes me as potentially relevant.
Ævar noticed this and has a better version of my patch, I think.
Michael
