On Fri, Aug 30, 2013 at 9:58 PM, Teresa Johnson <tejohn...@google.com> wrote:
>>>>> Besides, we might also want to
>>>>> use the same machinery (dump_printf_loc etc) for dump file dumping.
>>>>> The current behavior of using '-details' to turn on opt-info-all
>>>>> messages for dump files are not desirable.
>>>>
>>>> Interestingly, this doesn't even work. When I do
>>>> -fdump-ipa-inline-details=stderr (with my patch containing the inliner
>>>> messages) I am not getting those inliner messages emitted to stderr.
>>>> Even though in dumpfile.c "details" is set to (TDF_DETAILS |
>>>> MSG_OPTIMIZED_LOCATIONS | MSG_MISSED_OPTIMIZATION | MSG_NOTE). I'm not
>>>> sure why, but will need to debug this.
>>>
>>> It works for vectorizer pass.
>>
>> Ok, let me see what is going on - I just confirmed that it is not
>> working for the loop unroller messages either.
>>
>
> Found the issue. The stream was incorrectly being closed when it was
> stderr/stdout. So only the dump output before the first dump_finish
> call was being emitted to stderr. I fixed this the same way the
> alt_dump_file was being handled just below - don't close if it is
> stderr/stdout. Confirmed that this fixes the problem.
>
> (So the real ratio between the volume of -fdump-...=stderr and
> -fopt-info is much higher than what I reported in an earlier email)
>
> Is the following patch ok, pending regression tests?
>
> 2013-08-30 Teresa Johnson <tejohn...@google.com>
>
> * dumpfile.c (dump_finish): Don't close stderr/stdout.
>
> Index: dumpfile.c
> ===================================================================
> --- dumpfile.c (revision 202059)
> +++ dumpfile.c (working copy)
> @@ -450,7 +450,8 @@ dump_finish (int phase)
> if (phase < 0)
> return;
> dfi = get_dump_file_info (phase);
> - if (dfi->pstream)
> + if (dfi->pstream && strcmp("stderr", dfi->pfilename) != 0
> + && strcmp("stdout", dfi->pfilename) != 0)
> fclose (dfi->pstream);
>
> if (dfi->alt_stream && strcmp("stderr", dfi->alt_filename) != 0
Yes, this is clearly a bug which I missed. Thanks for fixing it. Is it
feasible to add a test case for it?
Thanks,
Sharad
> Thanks,
> Teresa
