[PATCH 1/1 V5] gdb : Signal to pstack/gdb kills the attached process.

[Pedro Alves]

Adding another question to the list below.  (I haven't tried to reproduce this yet myself, btw.)

On 2024-05-10 21:19, Pedro Alves wrote:
> Hi.
> 
> Just wanted to let you know that I've read all the discussion around this until this
> email I'm replying to, and started thinking about it a bit.  Unfortunately this is one of
> those areas in GDB where the right change is rarely immediately obvious (to me).
> 
> Some questions:
> 
>  - If you ctrl-c to abort the attach, do we really abort the 
>    attach properly?  Or do we stay attached in some half broken state?
> 
>  - Below you mention pstack, where can we find it?  And you mention
>    that ctrl-c is pressed while that is printing a stack.  I'm assuming
>    that's a backtrace command.  I'm confused in that case, as if that is
>    so, then we should already be past the initial attach.  The question
>    would then becomes, shouldn't gdb have the terminal at that point?
>    How come it does not?

#3 - The patch description states:

 > Problem: While gdb is attaching an inferior, if ctrl-c is pressed in the
 > middle of the process attach,  the sigint is passed to the debugged
 > process.  This triggers the exit of the inferior.

This SIGINT passing is done with "kill(-pgrp, SIGINT)".  How does that manage
to trigger the exit of the inferior at all?  ptrace should intercept the
SIGINT before the inferior ever sees it.  Did it not?

Or could it be that the real issue is that because that sends the SIGINT
to all the processes in the inferior's pgrp, we kill more processes than
the one we're attaching to, and those processes exiting cause the inferior
to exit as well.  If so, then this is orthogonal to the initial attach,
and can happen after the attach as well.  There is a bug open about this
on bugzilla.

Pedro Alves

> 
> I'm wondering whether Baris's patch to eliminate the inferior
> continuations would help with this, as it probably makes the attaching
> sequence synchronous.  I should probably look at that one.
> 
> Pedro Alves
>