Re: Does musl build error with gcc-4.9.2+ git repo resolved

[Jens Gustedt <jens.gustedt@inria.fr>]

[-- Attachment #1: Type: text/plain, Size: 1955 bytes --]

Am Mittwoch, den 11.03.2015, 08:02 -0400 schrieb Rich Felker:
> On Wed, Mar 11, 2015 at 09:52:35AM +0530, vlse wrote:
> > I want to know whether the gcc 4.9.x bug bugzilla id 61144 resolved
> > with latest git/svn branch of gcc-4.9.x.
> > 
> > That bugzilla page says bug 61144 has been resolved. Anybody has
> > confirmed it? Has musl changed it's build scripts to account that?
> 
> I haven't done detailed tests to confirm, but I believe the bug is
> fixed in 4.9.2. However there's another very serious bug that affects
> 4.9.2 and possibly all 4.9.x releases causing incorrect code gen:
> 
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65307
> 
> On the musl side, we worked aroung bug 61144 with gratuitous use of
> volatile before 4.9.2 was released. This workaround might be reverted
> at some point in the future if the affected versions are deemed too
> buggy for any distros to keep them around.

I have another bug in 4.9 that is really nasty and for which the
minimal code to produce is very lengthy and contrived. So much that I
didn't even yet reported it because I wanted to cook it down a bit,
still. For the moment it involves at least 15 levels of nested for
loops and setjmp/longjmp. The bug then has the compiler crash, but who
knows what it produces if it doesn't crash for code that is more
normal.

There are other issues (but perhaps related) that are really
suspicious with that line of gcc. The static analyser seems to be
screwed and tells me about lack of initialization for variables that
are initialized.

So I wouldn't trust 4.9 one bit for anything related to production.

Jens

-- 
:: INRIA Nancy Grand Est ::: Camus ::::::: ICube/ICPS :::
:: ::::::::::::::: office Strasbourg : +33 368854536   ::
:: :::::::::::::::::::::: gsm France : +33 651400183   ::
:: ::::::::::::::: gsm international : +49 15737185122 ::
:: http://icube-icps.unistra.fr/index.php/Jens_Gustedt ::




[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 181 bytes --]