Re: [PATCH] Fix gdb crash due to SIGPIPE when the compile command fails

[Pedro Alves <pedro@palves.net>]

On 2021-06-04 2:39 p.m., Tom Tromey wrote:
>>>>>> "Bernd" == Bernd Edlinger <bernd.edlinger@hotmail.de> writes:
> 
> Bernd> I believe it is right to avoid the SIGPIPE before calling the plugin,
> Bernd> instead of doing that in gcc-trunk, since we don't know which version
> Bernd> we will be calling, and all versions I tried have failed like this.
> 
> That seems fine, but I think it would be better to install the handler
> just when working with the plugin, and then uninstall it afterward, sort
> of like what class scoped_ignore_sigttou does.

One thing I dislike here (and also about scoped_ignore_sigttou), is that the
signal disposition is process-wide.  

If we happen to have multiple threads that need to ignore the signal changing the
disposition like that, then they may race.

We can fix that by making the ignoring be per-thread instead by using
sigprocmask + sigtimedwait.  

We can also introduce a generic scoped_ignore_signal so that both scoped_ignore_sigttou
and scoped_ignore_pipe (and whatever other similar signal we may need to ignore) share
the same code.  

I gave it a quick try.  See the 4 patches at the top of the
users/palves/scoped_ignore_signal branch.

This was something that I've thought before about doing for SIGTTOU to be honest.
Seeing it being copied over to SIGPIPE was simply the trigger that made me
actually give it a try.  

WDYT?

Pedro Alves