Message-ID: <53282d70-c217-d65e-2d30-7a5593d6c585@gmail.com>
Date:   Tue, 15 Jun 2021 21:57:47 +0200
From:   Maciej Falkowski <maciej.falkowski9@...il.com>
To:     Masahiro Yamada <masahiroy@...nel.org>,
        Nick Desaulniers <ndesaulniers@...gle.com>
Cc:     Nathan Chancellor <natechancellor@...il.com>,
        clang-built-linux <clang-built-linux@...glegroups.com>,
        LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 2/2] Makefile: clang-tools: Omit printing stack trace when
 KeyboardInterrupt is raised


On 5/22/21 5:03 AM, Masahiro Yamada wrote:
> On Sat, May 22, 2021 at 2:18 AM Nick Desaulniers
> <ndesaulniers@...gle.com> wrote:
>> On Thu, May 20, 2021 at 4:18 PM Maciej Falkowski
>> <maciej.falkowski9@...il.com> wrote:
>>> When user terminates the script (also implicitly through for example
>>> `make clang-analyzer`) by using
>>> Ctrl+C (or sending SIGINT more generally) the KeyboardInterrupt
>>> is raised printing stack trace of the execution as shown below:
>>>
>>> $ ./scripts/clang-tools/run-clang-tools.py clang-tidy ./compile_commands.json
>>> ^CTraceback (most recent call last):
>>>    File "./scripts/clang-tools/run-clang-tools.py", line 74, in <module>
>>>      main()
>>>    File "./scripts/clang-tools/run-clang-tools.py", line 70, in main
>>>      pool.map(run_analysis, datastore)
>>>    File "/usr/lib64/python3.8/multiprocessing/pool.py", line 364, in map
>>>      return self._map_async(func, iterable, mapstar, chunksize).get()
>>>    File "/usr/lib64/python3.8/multiprocessing/pool.py", line 765, in get
>>>      self.wait(timeout)
>>>    File "/usr/lib64/python3.8/multiprocessing/pool.py", line 762, in wait
>>>      self._event.wait(timeout)
>>>    File "/usr/lib64/python3.8/threading.py", line 558, in wait
>>> Process ForkPoolWorker-6:
>>> Process ForkPoolWorker-1:
>>> Process ForkPoolWorker-5:
>>> Process ForkPoolWorker-7:
>>> Process ForkPoolWorker-2:
>>> Process ForkPoolWorker-3:
>>> Process ForkPoolWorker-4:
>>> Process ForkPoolWorker-8:
>>>      signaled = self._cond.wait(timeout)
>>>    File "/usr/lib64/python3.8/threading.py", line 302, in wait
>>>      waiter.acquire()
>>> KeyboardInterrupt
>> With this applied,
>> $ make LLVM=1 LLVM_IAS=1 -j72 clang-analyzer
>> ^C
>> Process ForkPoolWorker-5:
>> make: *** [Makefile:1902: clang-analyzer] Error 130
>>
>> Reviewed-by: Nick Desaulniers <ndesaulniers@...gle.com>
>> Tested-by: Nick Desaulniers <ndesaulniers@...gle.com>
>>
>> Thanks for the patch!
>
> I am not a python expert, but is
> "Let's suppress back-trace when a user presses an interrupt"
> the common coding style?
I think that is not a common coding style.
In this case, the back-trace is verbose and it is caused
by a bug in multiprocessing Pool.

> If really so, do we need to do something similar in all python scripts?
> I do not know what is special about run-clang-tools.py.
I think no, I misunderstood the nature of the problem.
The verbose back-trace is a result of the multiprocessing Pool
not handling KeyboardInterrupt properly.
It is the old bug(bugs.python.org/issue8296), yet to be 
solved(bugs.python.org/issue22393).
For more info please see:
https://stackoverflow.com/questions/1408356/keyboard-interrupts-with-pythons-multiprocessing-pool

This issue is already once solved in a Linux source tree in a 
scripts/checkkconfigsymbols.py:

scripts/checkkconfigsymbols.py:312
     try:
         return check_symbols_helper(pool, ignore)
     except KeyboardInterrupt:
         pool.terminate()
         pool.join()
         sys.exit(1)

Also the comment:
scripts/checkkconfigsymbols.py:321
"""Helper method for check_symbols(). Used to catch keyboard interrupts in
check_symbols() in order to properly terminate running worker processes."""

> For example, if I press Ctrl-C while building Clang
> by using tc-build, I see a back-trace.
> I have never thought back-tracing was annoying.
>
> The exit code is 130 regardless of this patch.
Okay.
>
> BTW, I prefer not having "Makefile:" in the patch subject
> since this is not touching Makefile at all.
>
> I rather like "clang-tools:" or "scripts/clang-tools:".
>
Okay.

The solution should cover the whole bug, I think this patch
may be dismissed then.
Thank you for your feedback!

Best regards,
Maciej Falkowski

>
>
>
>
>
>>> The patch handles the raise of the KeyboardInterrupt and exits when occurred
>>> with code 130 as documented in: https://tldp.org/LDP/abs/html/exitcodes.html
>>>
>>> Signed-off-by: Maciej Falkowski <maciej.falkowski9@...il.com>
>>> ---
>>>   scripts/clang-tools/run-clang-tools.py | 5 ++++-
>>>   1 file changed, 4 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/scripts/clang-tools/run-clang-tools.py b/scripts/clang-tools/run-clang-tools.py
>>> index 38fc311d2e03..eb0e0ecfce24 100755
>>> --- a/scripts/clang-tools/run-clang-tools.py
>>> +++ b/scripts/clang-tools/run-clang-tools.py
>>> @@ -77,4 +77,7 @@ def main():
>>>
>>>
>>>   if __name__ == "__main__":
>>> -    main()
>>> +    try:
>>> +        main()
>>> +    except KeyboardInterrupt:
>>> +        sys.exit(130)
>>> --
>>> 2.26.3
>>>
>>
>> --
>> Thanks,
>> ~Nick Desaulniers
>
>
> --
> Best Regards
> Masahiro Yamada

Powered by blists - more mailing lists