At 2026-04-21T18:24:18-0700, Collin Funk wrote:
> "G. Branden Robinson" <[email protected]> writes:
> > [Colin wrote]:
> >> I suspect you need to do something like this:
> >> 
> >>     $ git clean -xdf
> >>     $ git submodule deinit --all --force
> >>     $ ./bootstrap && ./configure && make
> >
> > Yup, that did the trick!  Excellent.  I hope to have a report for
> > you about a measurable consequence of improved coreutils 9.11
> > performance.
> 
> Cool, glad to hear.

I have the report.  I don't think it will blow anyone away but it does
look like an improvement.

Method: I uninstalled coreutils 9.11.  I rebuilt groff (with
bootstrap/configure/make/make check), then ran "make -C build check" 10
times.  I used a single core on my 12-core system.  I reinstalled
coreutils 9.11, and repeated the procedure.

I am hopeful that the improvement lies outside the margin of error.

$ time for n in seq 10; do make -C build check; done
...
make: Leaving directory '/home/branden/src/GIT/groff/build'

real    1m2.396s
user    0m59.349s
sys     0m14.095s

$ time for n in seq 10; do make -C build check; done
...
make: Leaving directory '/home/branden/src/GIT/groff/build'

real    1m2.305s
user    0m59.836s
sys     0m13.462s

Spending over half a second less in the system is about a 5% improvement
in that respect, even if the wall clock difference doesn't look like
much.  And to be fair I probably hit--besides the shell and groff
executables themselves--grep harder than any single coreutils program.
But I _do_ use them.  od(1) comes in quite handy.  sed's in there too.

Over time I hope to learn some more sophisticated testing techniques
from coreutils's test suite.  Of groff's 342 automated tests, I wrote
all but about 5 and they're caveman simple.  Partly that's due to my own
knuckle-dragging nature and by design, a so far mostly failed effort to
attract contributions of test scripts from others.

> > Seems you're right.  Is it too much to hope for gnulib to add an
> > Autoconf test that detects its own staleness in this respect?  Not
> > quite the right list to ask, I know, but you may already know the
> > answer and if not, I know where to direct my query.
> 
> Moving to bug-gnulib, in case anyone else wants to comment. The Venn
> diagram of people that subscribe to bug-gnulib and coreutils is
> probably a single circle. But maybe my estimates are wrong.
[snipping]

I didn't have high hopes, but wanted to raise the issue just in case
there was an easy win to be had.  I've gotten comfortable enough with
groff's build system that it seldom surprises me even when it breaks,
but with other projects, a build failure can leave me with that newbie
programmer feeling as I stare slackjawed at an inscrutable crash.

Regards,
Branden

Attachment: signature.asc
Description: PGP signature

Reply via email to