slow script execution when debugging large binaries

[Mike Frysinger]

On Thursday 10 August 2006 18:05, Daniel Jacobowitz wrote:
> On Thu, Aug 10, 2006 at 06:04:20PM -0400, Mike Frysinger wrote:
> > > But at a guess, you'll still have a symbol lookup on top of the
> > > profile.
> >
> > yeah, new one attached looks as you thought
>
> Well there you go then.  How are we getting into symbol lookups while
> handling internal variables?  A breakpoint on lookup_partial_symbol
> will probably tell you.

first time ive used gprof/looked at gdb source code so i could easily be off, 
but the call graph i posted gives the clues we need ?

looking at the gprof call graph, lookup_symbol() is only called by 
write_dollar_variable() ... looking at parse.c, write_dollar_variable() does 
a symbol lookup on purpose because of funky systems (like hppa) that allow 
system routines to begin with $ or $$

if i cheat and comment out lookup_symbol() here and force sym to be NULL, the 
speed test runs instantaneously even after loading up the kernel

perhaps a build time check to allow this behavior ?  force hppa targets to 
take the hit and not screw everyone else ? :)
-mike
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 827 bytes
Desc: not available
URL: <http://sourceware.org/pipermail/gdb/attachments/20060810/e7c0f318/attachment.sig>