Hi Zack, Thanks for the swift follow-up!
At 2026-03-05T15:59:41-0500, Zack Weinberg wrote:
> On Thu, Mar 5, 2026, at 10:58 AM, G. Branden Robinson wrote:
> > I had to use a nasty hack to force serialization of two regression
> > tests that both can cause core dumps, and so can race with each
> > other in the TOCTTOU window on the existence of a "core" file.
>
> Are the core dumps necessary for the test?
Strictly? Probably not. I could alter the test cases to format some
hapax legomenon text after the applicable crash points observed in the
past and then grep the formatter output for it.
> If not, I would suggest wrapping each one with a shell script that
> does `ulimit -c 0` and then execs the main test program.
I kind of liked the idea of leaving a meaningful[1] core file in the
build tree so that it could be immediately inspected without having to
hack up the test script to take out the "ulimit -c 0".
Was I using the documented serialization mechanism incorrectly? Why
didn't it work?
Regards,
Branden
[1] assuming the person doing the build didn't employ "configure" and
"make" options so as to boil off all useful debugging information,
as distributors have traditionally loved to do
signature.asc
Description: PGP signature
