Re: [RFC patch] non-release srctrees: --enable-targets=all & 64bit & -lmcheck

[Jan Kratochvil <jan.kratochvil@redhat.com>]

On Sun, 20 May 2012 15:59:16 +0200, Alan Modra wrote:
> > --enable-64-bit-bfd is required for full --enable-targets=all.
> 
> Always building with a 64-bit bfd hides another class of bug.

Yes, I run nightly build without --enable-64-bit-bfd here to test such bugs.

The goal of my proposal was to catch most of the bugs already by the
submitter.  We cannot want from all submitters to test too many combinations
(I do 3*17 = 46 runs; it equals a bit less due to some limitations).
So we should choose some best single catch-almost-all configuration.

While there are countercases I believe >90% of bugs get more caught by
--enable-targets=all --enable-64-bit-bfd than by any other settings.

After I know submitters run --enable-targets=all themselves I probably start
also runs without --enable-targets=all to test the other case, also I would
start new runs running -lmcheck.


> I'm not in favour of this patch.  Yes, it may help first-time
> contributors in testing their patches, but means yet another change in
> configure options for anyone past that stage.

I find the largest disadvantage that after switching to development=false for
the release the srctree becomes "untested".  Even -lmcheck may hide some bugs.
But there is also the pre-release gdb-x.y.90.z phase to catch those bugs.


I just find current regressions happenning almost daily without this patch to
be too boring.


Thanks,
Jan