Re: per thread breakpoints

[Andrew Cagney <ac131313 at cygnus dot com>]

Quality Quorum wrote:
> 
> Hi,
> 
> It seems to me that gdb core does NOT seriously support per thread breaks:
> 
> 1. target_insert_breakpoint() is unaware about thread associated with
>    a breakpoint.

Yes.  The original thread model was developed on systems that didn't
have kernel level threads (it wasn't SMP either :-).  Various
assumptions were made such as it wasn't actually possible to set a
breakpoint on a specific thread.  Instead all breakpoints were set
globally.

> 2. insert_breakpoints() does not do any dances with inferior pids before
>    and after calls to target_insert_breakpoint().
> 
> I would suppose it should be fixed ASAP (e.g. by adding pid argument to
> to the target_insert_breakpoint): it is a huge resource hog in
> case of remote debugging and it seems like linux threads would also
> greatly benefit from this improvement.

I don't think it is that easy.  I recently looked through the thread
code to see what would need changing.  At present GDB makes all sorts of
assumptions about breakpoints and their sideeffects.  From memory,
things like the way GDB iterates over breakpoints and resolves
duplicates would also need to be changed.

Rather than a simplistic tweek, I think there needs to be a clear robust
thread model that GDB can map onto all real thread implementations.

	Andrew