Re: [PATCH 1/4] gdb: add asserts in target.c for target_async_permitted

[Simon Marchi <simark@simark.ca>]

On 2021-11-23 9:08 a.m., Andrew Burgess via Gdb-patches wrote:
> The target_async_permitted flag allows a user to override whether a
> target can act in async mode or not.  Ideally we would check this flag
> in target_can_async_p and target_is_async_p, however, there are a few
> places in GDB where we call the can_async_p and is_async_p methods
> directly without calling through the target_* wrapper.

Off-hand, I would ask: is it correct for these places to call the target
methods directly?  Should we fix them?

Looking up these spots, I see that they are places where we call
can_async_p on targets possibly before they are pushed, like in
run_one_inferior, so we can't use the target_ wrapper there.

I wouldn't hate it if the targets themselves didn't have to check
target_async_permitted.  For example, linux_nat_target::can_async_p
would simply return true.  And the spots mentioned above could be
changed from:

  int async_p = mi_async && run_target->can_async_p ();

to use a new wrapper overload:

  int async_p = mi_async && target_can_async_p (run_target);

That new wrapper would check target_async_permitted.  I don't think that
is_async_p needs the same treatment: if we never make a target async
because target_async_permitted is false, then is_async_p will always
return false.

Simon