Re: [PATCH] [gdb] Fix assert in delete_breakpoint

[Tom Tromey <tom@tromey.com>]

>>>>> "Simon" == Simon Marchi <simark@simark.ca> writes:

Simon> On the other hand, I suspect that any memory read / write used to
Simon> implement GDB's internal logic should be considered by default unsafe to
Simon> interrupt.  Otherwise, that code should be written in a way that ensures
Simon> that should exceptions get thrown, the internal state is left in a
Simon> coherent... state, either by rolling back changes or otherwise.  That
Simon> sounds difficult, if not impossible.

Maybe I don't understand the context or something, because this seems
like a pretty big departure from historical practice.  In gdb,
exceptions can happen nearly anywhere and code must ordinarily be
exception-safe.

In this situation I think the issue is that some code was not, in a
subtle way.

Simon> The question is, how to decide where QUITs should be suppressed.  Here's
Simon> a random backtrace I just grabbed from a maybe_quit call under an
Simon> update_global_location_list call:

The danger with suppressing these quits is that then gdb can enter
uninterruptible states if something bad happens to the remote.  That
seems unfortunate.  At least assuming I understand correctly.

Tom