Re: [PATCH] gdb: iterate over targets, not inferiors, to commit resumed

[Simon Marchi <simark@simark.ca>]

On 3/4/24 04:28, Aktemur, Tankut Baris wrote:
> Hi Simon,
> 
> I see; thanks for pointing this out.  I'd then retract the patch.  Let's
> keep maybe_set_commit_resumed_all_targets and maybe_call_commit_resumed_all_targets
> both have the same iteration for consistency.  We can do the change in the future
> when the amd-dbgapi target is converted to a process_stratum_target.
> 
> Thanks
> -Baris

Great, thanks for your understanding.  We'll hopefully get that sorted
out this year.

In the back of my mind, I'm also worried about this for other (mostly
theoritical at this point) cases where we do things per-process-target,
ignoring that there might be other targets on top that we're bypassing.

I recently thought about this scenario, imagine we have the following
inferiors:

                | inf1             | inf2
target stratum  | my-thread-target |
process stratum | linux-nat        | linux-nat

Imagine the thread target is meant to pull events from the linux-nat
target below and translate them to some higher level events about
userspace thread.  When infrun pulls events, it selects a random
inferior and calls target_ops::wait with minus_one_ptid.  What if
inferior 2 is selected, and linux-nat returns an event about inferior 1,
that was meant to be processed by my-thread-target?

Simon