>Anyway I was able to repro consistently when doing a full test run, and was 
>able to pass or skip always with the attached.

The patch works for me.

>> The smallest number that passed for me was 108336
>>   "(ulimit -v $(($vm+108336)) && fold 2>err >/dev/full)"
>Over 100MiB ?
>I wonder why the other ulimits in other tests OK?

Could the 100MB be loading the CentOS 7 locale data?
-rw-r--r-- 1 root root 106176928 Jul 18  2022 /usr/lib/locale/locale-archive
Maybe calling the library functions that fold needs to avoid breaking multibyte 
characters have a side-effect of loading the locale data.
My CentOS 7 server has glibc-2.17-326.el7_9.x86_64

William

________________________________
From: Pádraig Brady <[email protected]> on behalf of Pádraig Brady 
<[email protected]>
Sent: Sunday, September 21, 2025 11:53 AM
To: William Bader <[email protected]>; Collin Funk 
<[email protected]>
Cc: [email protected] <[email protected]>; [email protected] 
<[email protected]>; Bruno Haible <[email protected]>
Subject: Re: new snapshot available: coreutils-9.7.327-71a8c.tar.xz

On 21/09/2025 04:12, William Bader wrote:
>  >$ make check TESTS=tests/fold/fold-zero-width.sh VERBOSE=yes
>
> That still gave me
>    fatal: making test-suite.log: failed to create 
> tests/fold/fold-zero-width.sh.trs
>    fatal: making test-suite.log: failed to create 
> tests/fold/fold-zero-width.sh.log

You need to add SUBDIRS=. so that it doesn't try to do the same TESTS in 
gnulib-tests.
But in my testing on a Centos6 VM you need to run all tests to trigger this 
anyway.

> and it showed
>    # PASS:  1
> even with
>    "(ulimit -v $(($vm+0)) && fold 2>err >/dev/full)"
> My CentOS 7 server has make-3.82-24.el7.x86_64
>
> So, I just ran the full tests with
>    make check VERBOSE=yes
> and did a binary search.
> The smallest number that passed for me was 108336
>   "(ulimit -v $(($vm+108336)) && fold 2>err >/dev/full)"
Over 100MiB ?
I wonder why the other ulimits in other tests OK?
I suppose this test is applying ulimits in a loop which might be a trigger.
Perhaps there is some weird kernel bug on these older Linux kernels
where the ulimit is interpreted as bytes rather than kilobytes sometimes.

Anyway I was able to repro consistently when doing a full test run,
and was able to pass or skip always with the attached.

cheers,
Padraig

Reply via email to