Re: [PATCH 0/6] Allow to run the old testsuite in parallel

[Ralf Wildenhues]

Hi Gary,

* Gary V. Vaughan wrote on Sun, Aug 22, 2010 at 04:05:12PM CEST:
> On 22 Aug 2010, at 17:55, Ralf Wildenhues wrote:
> > I've had this patch series half-done in my tree for a long time,
> > and have now taken the time to fix the missing bits.
> 
> Do you have (or could you easily move) this on a private topic branch
> already? If so, please push the topic branch to a public repo to make
> it easier for me to test.

Done now: it's named parallel-tests.

> I'd still like to roll the next release in a week or two though.  Do
> you think this patch series is stable enough to push just before that
> release?

I was definitely aiming for this release.  Actually, I have one more
patch to use Autobuild, and my plan was to then get the sysroot branch
merged and then build and test Libtool in all setups I have available
and send them to the Autobuild site for reference.  This parallel-tests
patch series makes it easier to submit logs of the failed tests.

> I'm leaning towards holding off until just after the release:
> anyone who wants to test that code can easily clone the post-release
> head of the master branch... and we don't run the risk of making a
> release that blows up for casual users in unforseen ways.  WDYT?

Well, I hope that I can convince you otherwise, but sure, after the
release is still better than never.

Maybe it helps to explain that I actually had the meat of this patch,
the parallel-tests enabling code, in my test tree for almost two years
now.  In fact, when I merged Akim's parallel-tests driver code, I used
Libtool as testing ground.  So, while obviously there can be bugs, and
due to size alone chances are there will be  ;-)  I'm optimistic that
this won't be too risky.

> > parallel-tests requires a new Automake for building the Libtool package
> > itself.  Users of Libtool macros should still be able to cope with
> > Autoconf 2.59 and Automake 1.9.6.  I tested this on GNU/Linux.
> 
> So, with a distribution's preinstalled libtool package, we are still
> able to work fully with older autotools, and the Automake 1.11.1
> requirement is only for running the old testsuite?  Or only for
> running the old testsuite in parallel?

The new Automake is merely needed for the toplevel Makefile.  It is not
needed for the old tests/demo*/ directories and not for running any of
the tests.  Autoconf 2.62 is purely needed for the toplevel configure
and for creating scripts that use getopt.m4sh.  I verified each of these
two requirements, by running autoreconf-2.59 in all old tests/demo*/
directories, having Autoconf 2.59 and Automake 1.9.6 first in $PATH, and
running 'make -k check' that way, successfully, on GNU/Linux.

> I'd like to try it out (preferably without extracting the patches
> from emails on my laptop, transferring them to the test servers,
> and applying them there) before giving the all clear.

Of course.

> But, I think
> it is not worth your expending too much more effort on enhancing
> this series if my plan to migrate those tests to Autotest comes to
> fruition in the next month or two, so if I don't get back with a
> review in 72hrs, and if you are confident that we're not opening
> ourselves up to a slew of reports from casual early adopters who
> have trouble with the altered testsuite, then in principle I have
> no problem with merging.

Great.  Thanks!
Ralf