[augeas-devel] [PATCH] enable automake-1.11 features: colorized and parallel tests, etc.

[David Lutterkort]

Hi Jim,

On Mon, 2009-06-15 at 16:31 +0200, Jim Meyering wrote:
> Now that automake-1.11 will soon be hitting F9, F10, F11,
> you may want to use some of its features.
> 
> Parallel tests is can be useful when you run parallel
> "make check" on a multi-core system.
> E.g., consider running "make -j9 check" on a quad-core system:

Heh .. yeah, I am usually even more reckless and just run 'make -j
check' ;)

> It took about 70 seconds before the changes.
> Adding automake's parallel-tests option shaved off only 5 or 10s
>   due to the fact that so much time was spent in a single test:
>   test-lenses.sh
> After splitting test-lenses.sh into its 54 separately-runnable tests,
>   and moving long-running tests "up" in the list, so that they start sooner,
>   a parallel "make check" completes in under 30 seconds.

Nice .. running the tests was getting to the point where it was starting
to annoy me ;)

>   Oddly, sometimes lens-grub.sh takes over 50(!) seconds and makes the total
>   test time approach 70s.  Other times it completes in just 5 seconds.

The grub lens contains some truly heinous regular expression, since it
also parses the Debian-specific embedded comments for update-grub.

>     $ for i in $(seq 10); do
>         env time --format %e \
>           ./augparse --nostdinc -I ../lenses ../lenses/tests/test_grub.aug
>       done

Strange .. I get runtimes that are pretty consistently 5s +/- .1s; I'll
run this for a little longer. Did you build with any special CFLAGS ?
But even when I build with CFLAGS='-O2 -g' and when I use the augparse
from the 0.5.0 RPM on my F10 machine, I get a very consistent picture.

> --------------------------------
> 
> automake's colorized-tests option is ok.
> Colors "PASS" in green and "FAIL" in red.
> Other colors for XPASS and XFAIL.  No big deal.
> 
> Likewise, I like silent-rules, but it's no big deal.

I don't have automake-1.11 on my F10 machine yet; before pushing it, I'd
like to play with it a little. I'll do that in the next few days.

> ----------
> 
> Along the way, I noticed that "make distcheck" was failing due to
> fa_sym.version not being distributed.  That's the 3rd patch below.

Cool .. pushed that one.

> ----------
> 
> As to whether it's ok to depend on automake-1.11, realize that
> the only people who will need it are those who build from a
> cloned directory.  If you build from a tarball, you don't need
> any version of automake.  And if you're developing, you can
> be expected to use the latest versions of autoconf and automake.

Yeah, I don't have big philosophical problems with requiring the latest
autotools - I do want to wait though until they've hit the Fedora stable
repos, even on my behind-the-curve F10 setup ;)

David