Robert O'Callahan
Robert O'Callahan
Another option would be to extend `check_for_ioc_period_bug()` to try 32 if we get `EINVAL` with 1 and if that works, use a minimum of 32 thereafter. I like this option...
Let me know if this doesn't work for you.
ABC is building this code in C translation units so I have to do this differently in a C-compatible way.
Not sure what happened to the Mac GCC11 build in CI, but everything else seems fine and ready for review. This is basically https://github.com/berkeley-abc/abc/commit/3286179c4855810dc25a59f293274c5a7ff606ca
@aletempiac can you please take a look at this? I landed this in upstream ABC a while back.
Maybe `rr record --nested=release` should just not record at all, and always exec the child directly when not already running under rr? It seems a bit weird to have it...
This is a gnarly one! Here's what happens under rr: * The SIGSEGV causes a switch to the sigaltstack. * The SIGSEGV handler enters `pause`, which we try to syscall-buffer....
It is not obvious what is the best way to fix this. Probably the simplest thing to do is forcibly set the `SS_AUTODISARM` flag on calls to `sigaltstack()`. That flag...
Yes, it would. Thanks for reminding me --- I remembered discussing that approach, but I forgot you actually implemented it. I should take that over and get it landed.
You can go ahead and edit the wiki yourself.