t3404.6 breaks on master under GIT_FSMONITOR_TEST

> The GIT_FSMONITOR_TEST variable allows you to roundtrip the fsmonitor
> codpath in the whole test suite. On both Debian & CentOS this breaks for
> me:
>
>     (cd t && GIT_FSMONITOR_TEST=$PWD/t7519/fsmonitor-all ./t3404-rebase-interactive.sh -i)
>
> Whereas this works:
>
>     (cd t && GIT_FSMONITOR_TEST=$PWD/t7519/fsmonitor-all GIT_SKIP_TESTS=t3404.6 ./t3404-rebase-interactive.sh -i)
>
> The entirety of the rest of the test suite still passes with
> GIT_FSMONITOR_TEST.
>
> This has been failing ever since GIT_FSMONITOR_TEST was introduced in
> 883e248b8a ("fsmonitor: teach git to optionally utilize a file system
> monitor to speed up detecting new or changed files.", 2017-09-22). Under
> -v -x -i:
>
>     + echo test_must_fail: command succeeded: env FAKE_LINES=exec_echo_foo_>file1 1 git rebase -i HEAD^
>     test_must_fail: command succeeded: env FAKE_LINES=exec_echo_foo_>file1 1 git rebase -i HEAD^
>     + return 1
>     error: last command exited with $?=1
>     not ok 6 - rebase -i with the exec command checks tree cleanness
>     #
>     #               git checkout master &&
>     #               set_fake_editor &&
>     #               test_must_fail env FAKE_LINES="exec_echo_foo_>file1 1" git rebase -i HEAD^ &&
>
> Maybe once this is fixed running the test suite under GIT_FSMONITOR_TEST
> would be a useful Travis target, but I don't know the current status of
> adding new options to Travis.

> On Thu, Feb 01 2018, Ævar Arnfjörð Bjarmason wrote:
>
>> The GIT_FSMONITOR_TEST variable allows you to roundtrip the fsmonitor
>> codpath in the whole test suite. On both Debian & CentOS this breaks for
>> me:
>>
>>     (cd t && GIT_FSMONITOR_TEST=$PWD/t7519/fsmonitor-all ./t3404-rebase-interactive.sh -i)
>>
>> Whereas this works:
>>
>>     (cd t && GIT_FSMONITOR_TEST=$PWD/t7519/fsmonitor-all GIT_SKIP_TESTS=t3404.6 ./t3404-rebase-interactive.sh -i)
>>
>> The entirety of the rest of the test suite still passes with
>> GIT_FSMONITOR_TEST.
>>
>> This has been failing ever since GIT_FSMONITOR_TEST was introduced in
>> 883e248b8a ("fsmonitor: teach git to optionally utilize a file system
>> monitor to speed up detecting new or changed files.", 2017-09-22). Under
>> -v -x -i:
>>
>>     + echo test_must_fail: command succeeded: env FAKE_LINES=exec_echo_foo_>file1 1 git rebase -i HEAD^
>>     test_must_fail: command succeeded: env FAKE_LINES=exec_echo_foo_>file1 1 git rebase -i HEAD^
>>     + return 1
>>     error: last command exited with $?=1
>>     not ok 6 - rebase -i with the exec command checks tree cleanness
>>     #
>>     #               git checkout master &&
>>     #               set_fake_editor &&
>>     #               test_must_fail env FAKE_LINES="exec_echo_foo_>file1 1" git rebase -i HEAD^ &&
>>
>> Maybe once this is fixed running the test suite under GIT_FSMONITOR_TEST
>> would be a useful Travis target, but I don't know the current status of
>> adding new options to Travis.
>
> *Poke* at this again. Ben, or anyone else with knowledge of fsmonitor:
> Can you reproduce this?
>
> This failure along with the one I noted in
> https://public-inbox.org/git/87tvn2remn.fsf@evledraar.gmail.com/ is
> failing the tests on Linux when run with GIT_FSMONITOR_TEST.
>
> I'm looking at this again because SZEDER's patches to the split index
> reminded me again that we have these long-standing failures in rare test
> modes (see
> https://public-inbox.org/git/87va7ireuu.fsf@evledraar.gmail.com/ for the
> split index discussion).

> 
> On Thu, Sep 06 2018, Ævar Arnfjörð Bjarmason wrote:
> 
>> On Thu, Feb 01 2018, Ævar Arnfjörð Bjarmason wrote:
>>
>>> The GIT_FSMONITOR_TEST variable allows you to roundtrip the fsmonitor
>>> codpath in the whole test suite. On both Debian & CentOS this breaks for
>>> me:
>>>
>>>      (cd t && GIT_FSMONITOR_TEST=$PWD/t7519/fsmonitor-all ./t3404-rebase-interactive.sh -i)
>>>
>>> Whereas this works:
>>>
>>>      (cd t && GIT_FSMONITOR_TEST=$PWD/t7519/fsmonitor-all GIT_SKIP_TESTS=t3404.6 ./t3404-rebase-interactive.sh -i)
>>>
>>> The entirety of the rest of the test suite still passes with
>>> GIT_FSMONITOR_TEST.
>>>
>>> This has been failing ever since GIT_FSMONITOR_TEST was introduced in
>>> 883e248b8a ("fsmonitor: teach git to optionally utilize a file system
>>> monitor to speed up detecting new or changed files.", 2017-09-22). Under
>>> -v -x -i:
>>>
>>>      + echo test_must_fail: command succeeded: env FAKE_LINES=exec_echo_foo_>file1 1 git rebase -i HEAD^
>>>      test_must_fail: command succeeded: env FAKE_LINES=exec_echo_foo_>file1 1 git rebase -i HEAD^
>>>      + return 1
>>>      error: last command exited with $?=1
>>>      not ok 6 - rebase -i with the exec command checks tree cleanness
>>>      #
>>>      #               git checkout master &&
>>>      #               set_fake_editor &&
>>>      #               test_must_fail env FAKE_LINES="exec_echo_foo_>file1 1" git rebase -i HEAD^ &&
>>>
>>> Maybe once this is fixed running the test suite under GIT_FSMONITOR_TEST
>>> would be a useful Travis target, but I don't know the current status of
>>> adding new options to Travis.
>>
>> *Poke* at this again. Ben, or anyone else with knowledge of fsmonitor:
>> Can you reproduce this?
>>
>> This failure along with the one I noted in
>> https://public-inbox.org/git/87tvn2remn.fsf@evledraar.gmail.com/ is
>> failing the tests on Linux when run with GIT_FSMONITOR_TEST.
>>
>> I'm looking at this again because SZEDER's patches to the split index
>> reminded me again that we have these long-standing failures in rare test
>> modes (see
>> https://public-inbox.org/git/87va7ireuu.fsf@evledraar.gmail.com/ for the
>> split index discussion).
> 
> For what it's worth this is still broken, but more importantly (I'm not
> just keeping bumping the same thing) the only thing that's now broken
> under fsmonitor. I.e. my skip config is now GIT_SKIP_TESTS="t3404.7"
> whereas before 43f1180814 ("git-mv: allow submodules and fsmonitor to
> work together", 2018-09-10) I needed to add "t7411.3 t7411.4" to that.
>