Re: recommend use of gperf version 3

[Andrew Haley <aph at redhat dot com>]

Ranjit Mathew writes:
 > On 6/7/05, Bruno Haible <bruno@clisp.org> wrote:
 > > Hello Ranjit,
 > > 
 > > >   They should also be tested with Jacks:
 > > >
 > > >   http://gcc.gnu.org/install/test.html#TOC2
 > > >
 > > > There *should not* be any regressions, but you never
 > > > know...
 > > >
 > > > I assume that you bootstrapped and tested for regressions
 > > > after you applied your patch. Were you able to measure any
 > > > improvements in performance?
 > > 
 > > Sorry, I did not do all this (re-bootstrapping, regression-testing, measuring).
 > 
 > I did a rebuild of the Java bits (gcc/java, libjava) once with the
 > old table and once with the new table (built using gperf-3.0.1
 > and your command-line options). The build time went
 > slightly up from 20m 24.850s to 20m 28.315s (~0.2%, that
 > is noise).
 > 
 > There were no regressions in the testsuite (with Jacks included).
 > 
 > 
 > > But I know gperf well enough to be confident in the modified command-line
 > > options.
 > 
 > I didn't realise you were the co-author of gperf.
 > 
 > 
 > > I'm hoping that someone from the GCC developers likes the patch and commits
 > > it for me, and that the performance change may show up in the automatic
 > > builds.
 > 
 > Either a global maintainer has to approve it or the
 > maintainers of the C++ and the Java front-ends have
 > to approve the respective bits.

Thanks for doing the tests.  I'm OK with the Java bits.

Andrew.