bug#45651: 28.0.50; emacs native-comp crashes with pidigits bench

bug#45651: 28.0.50; emacs native-comp crashes with pidigits bench

[Michael Schmidt]

[-- Attachment #1: Type: text/html, Size: 31023 bytes --]

bug#45651: 28.0.50; emacs native-comp crashes with pidigits bench

[Andrea Corallo via Bug reports for GNU Emacs, the Swiss army knife of text editors]

Michael Schmidt <michael.a.schmidt@email.de> writes:

> Hi!
>  
> Running the pidigits benchmark crashes emacs with a SIGTERM.
> All other benchmarks from the elip-benchmarks-1.9 suite passed and were
> moved in another dir for this run.
> I tested both with and without libgmp.
>  

Hi Michael,

Interesting, I can't reproduce on my machine.

I suspect may due to excessive memory consumption on your system, is it
possible?  Could you have a run monitoring it?  Maybe also having a look
into 'dmesg' output may give some info about why the process got killed.

  Andrea

bug#45651: Aw: Re: bug#45651: 28.0.50; emacs native-comp crashes with pidigits bench

[Michael Schmidt]

[-- Attachment #1: Type: text/html, Size: 1875 bytes --]

bug#45651: Aw: Re: bug#45651: 28.0.50; emacs native-comp crashes with pidigits bench

[Andrea Corallo via Bug reports for GNU Emacs, the Swiss army knife of text editors]

Michael Schmidt <michael.a.schmidt@email.de> writes:

> Hi Andrea,
>  
> [here an extended version with cc to 45651@debbugs.gnu.org]
>  
> Clean build with
>
>     emacs-8ad983c4acef60a80e8d6b6ba891b1ef957f2d7c.tar.gz  
>  
> Fedorea 33 VirtualBox 24GB+10GB swap: pidigits passed with ~61% peak memory consumption.
> Fedorea 33 VirtualBox 16GB+10GB swap: pidigits passed with ~98% peak memory consumption.
> Fedorea 33 VirtualBox 8GB  +10GB swap: pidigits crashed as described before. Please no "Eight gigabytes of memory and
> constantly swaping" ;-)
> dmesg unchaged after boot, no further info.
>  
> A build without nativecomp passed pidigits with ~1.5% memory consumption.
> Really looks like a memory leak.

Yeah I think this is a bug in the benchmark.  This is compiled at speed
3 and the programmer should be responsible to "garbage-collect-maybe" in
hot loops if this is required.

If this theory is correct you should see the issue disappearing adding
~(declare (speed 2))~ to `elb-pidigits'.

  Andrea

bug#45651: Aw: Re: bug#45651: 28.0.50; emacs native-comp crashes with pidigits bench

[Andrea Corallo via Bug reports for GNU Emacs, the Swiss army knife of text editors]

Hi Michael,

sorry for the late replay, I've fixed with 5a37ce17ea elisp-benchmarks
as discussed and now GC can do its job in `elb-pidigits'.

I'm closing this and works on my system, happy to reopen in case.

Thanks!

  Andrea