Re: [draft patch 0/6] Split FYI and some review notes

[Tom Tromey <tromey@redhat.com>]

>>>>> "Gary" == Gary Benson <gbenson@redhat.com> writes:

Gary> Note that since Tom's ambiguous linespec work went in GDB needs to
Gary> update its list at every change.  You could seriously speed things up
Gary> if you could eliminate this need--I considered making it asynchronous
Gary> at one point, but couldn't get around it.

The plan is for the itset work to help with this; plus also perhaps
making a new objfile-limiting linespec, like "break libwhatever.so:function".

itsets will let us restrict the lookups to a given set of program
spaces; those left out will not have to have libraries loaded.

The linespec change will let us avoid loading debuginfo for other
libraries.

This may also need fine-grained breakpoint re-setting.  I filed this in
bugzilla recently.  I think the linespec thing is in there somewhere
too.

This approach assumes that the user knows about these things and cares
about gdb performance.  So maybe another tradeoff is better.  On balance
I tend to still like the ambiguous linespec approach, since I tend to
think predictability and uniformity is more important to more users than
debugger performance.


FWIW if you have patches to improve laziness, you can maybe test them by
arranging for your test program to crash -- don't set any breakpoints at
all.

Tom