Re: Regression checking vs. my -shared-libgcc patch

[Mark Mitchell <mark at codesourcery dot com>]

>>>>> "Alexandre" == Alexandre Oliva <aoliva@redhat.com> writes:

    Alexandre> In fact, I'm surprised you have never even tried to build the
    Alexandre> regression-testing environment :-(

I did once.  It took hours.  It did not work the first time.  And that
was from a GNU/Linux box.  Geoff Keating sent me a script that made it
work better, but even with that there were a few tweaks I had to make.

There's no telling what might happen crossing from somewhere else;
realistically, there are lots of cross configurations that do not
work, for one reason or another.  Usually the problems are easy to
fix, but they're still present, and it takes a while to track them
down.

For the last several weeks, I've been busting my you-know-what trying
to fix platform-specific problems on platforms in which I have no
direct interest (either out of my own personal use or because of
CodeSourcery customers).  Other entities do provide commercial support
for those platforms, but volunteers to do the work have not been
available.  David Edelsohn has been working hard to help me solve the
problems, but he's a busy guy with a whole 'nother job to do. :-(

I see it as my job as GCC release manager to solve these kinds of
problems, and to step in to solve problems where others will not, so
I've stepped in to do the work.  That's why I made the change I did;
it fixed problems on AIX -- and, actually, on every other platform
using a shared libgcc as well.  Unfortunately, that change is
apparently tickling some kind of bug.  Probably, this bug involves
platform-specific issues in the building of libgcc, since the change
seems to work well on other platforms, including GNU/Linux, IRIX, and
AIX.  I'd bet that -shared-libgcc was never tested on the
regression-testing platform, and that it never worked, but, of course,
I cannot prove that.

In this case, I'd really appreciate it if someone more familiar with
cross environments, the PowerPC architecture, and the particular
embedded environment in question, and perhaps someone with direct
access to the regression-tester logs, would try to figure out what the
problem is.  That would be really helpful to me, as I continue to
battle the AIX exception-handling issues.  If the fix is not
immediately obvious, I will definitely try to help figure out what the
fix should be.

Since you are clearly skilled in working in this kind of cross
environment, would you care to take a look?

Thanks,

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com