Re: [BUG] Git 2.38.0-rc1 t1800 message text comparison

From: René Scharfe <l.s.r@web.de>
Date: May 22, 2023, 20:39 UTC
Message-ID: <2e3be2e9-191d-fbfd-54c6-a3958e35637c@web.de>
In-Reply-To: <013701d90f80$692ba9c0$3b82fd40$@nexbridge.com>

Am 14.12.22 um 06:53 schrieb rsbecker@nexbridge.com:

Show 43 quoted lines

> On September 23, 2022 4:43 PM, I wrote:
>> On September 22, 2022 3:03 PM, I wrote:
>>> On September 22, 2022 3:02 PM, I wrote:
>>>> Rc1 is looking good except for this test.
>>>>
>>>> We get a diff as follows:
>>>>
>>>> -fatal: cannot run bad-hooks/test-hook: ...
>>>> +fatal: cannot exec 'bad-hooks/test-hook': Permission denied
>>>>
>>>> It looks like the pattern
>>>> sed -e s/test-hook: .*/test-hook: .../
>>>>
>>>> needs to be a bit extended. Adding
>>>>
>>>> sed -e s/exec/run/ | send -e s/["']//g
>>>>
>>>> might help clear off the other noise.
>>
>> A patch that might work is as follows:
>>
>> diff --git a/t/t1800-hook.sh b/t/t1800-hook.sh index 43fcb7c0bf..9a723631a2
>> 100755
>> --- a/t/t1800-hook.sh
>> +++ b/t/t1800-hook.sh
>> @@ -173,7 +173,10 @@ test_expect_success 'git hook run a hook with a bad
>> shebang' '
>>                -c core.hooksPath=bad-hooks \
>>                hook run test-hook >out 2>err &&
>>        test_must_be_empty out &&
>> -       sed -e "s/test-hook: .*/test-hook: .../" <err >actual &&
>> +       quot=`echo "\047"` &&
>> +       sed -e "s/exec/run/" <err | \
>> +               sed -e "s/$quot//g" | \
>> +               sed -e "s/test-hook: .*/test-hook: .../" >actual &&
>>        test_cmp expect actual
>> '
>>
>> This does not require setting up a prerequisite for NonStop and the
> technique
>> might make the MING code easier but adding a change from spawn to run.
>
> This is still broken on NonStop. Is there any hope of a resolution?

So trying to execute an executable file consisting only of the line "#!/bad/path/no/spaces" causes NonStop to report "Permission denied"? That message text belongs to error code EACCES, not to EPERM ("Operation not permitted"), right?

POSIX allows execve to return EACCES if the file to execute is not a regular file and executing that file type is not supported or if permissions are missing to one of its path components.

Either you have something called /bad that is not a regular file (e.g. a directory) -- then it's just a matter of changing the test to use a different supposedly non-existing filename, perhaps by creating and deleting a temporary file for just that purpose.

Or NonStop correctly returns ENOEXEC on the first execve(2) call in run-command.c and returns ENOACCES on the fallback with SHELL_PATH (default value /usr/coreutils/bin/bash), because you are not allowed to execute that shell. Then SHELL_PATH needs to be corrected.

Or valid shells need to be placed in some kind of allow list to be accepted in #!-lines, lest NonStop returns ENOACCES on them. Or NonStop is simply reporting a bogus error code for some reason. In those cases you probably need an execve) compat/ wrapper that corrects the error code.

Or I'm missing something here, which is a relatively safe bet. Anyway, depending on the cause of the "Permission denied" message, loosening the textual comparison in the test might not be enough. There may not be a point for end users to distinguish between "exec" vs. "run", but the silent_exec_failure feature of run-command depends on the error code being ENOENT.

René