Debugging M4 [Was Re: speedup bootstrap]

[Gary V. Vaughan]

Ralf Wildenhues wrote:
> > > You know, I looked at sppeding up bootstrap for myself about a year ago, 
> > > but couldn't figure out the cause.
>
> Don't be fooled into thinking that this patch needed less than a few
> months, on and off.  I don't know how experts profile m4 scripts, but I
> first tried removing macros one by one (starting with a good guess) and
> ended up manually #define'ing DEBUG_SYM in m4-1.4.3 together with a form
> of manual statistical profiling: interrupt m4 in gdb every now and then,
> look where it's at.  :)

I don't consider myself an expert on debugging m4 code (although I'm
perhaps an expert on debugging the engine itself), and I too add manual
#defines of the various DEBUG_* cpp symbols and recompile, and make
judicious use of gdb to step through things.

I have a vague recollection of having seen an m4 engine profiling patch
and/or discussion on this list or Rene's site a few years ago.  And the
M4-2.0 architecture is supposed to help with writing introspection plugins.
I have a profiler plugin on my TODO list for post 2.0 release.

> valgrind's massif for overall memory usage profiling showed me m4 uses
> little heap until it starts outputting; that's what finally got me to
> get to the point that we needed to make use of storage in order to gain
> on execution time.  Some off-list discussion with Stepan (thanks, BTW!)
> got me on the right track.

Valgrind is x86 specific unfortunately :-(  I'll have to wait for Mactel
boxes before I can use it...

Cheers,
        Gary.
--
Gary V. Vaughan      ())_.  address@hidden,gnu.org}
Research Scientist   ( '/   http://tkd.kicks-ass.net
GNU Hacker           / )=   http://www.gnu.org/software/libtool
Technical Author   `(_~)_   http://sources.redhat.com/autobook

signature.asc
Description: OpenPGP digital signature