OpenBLAS fixed in Xcode 11.2

Jack Howarth howarth.at.macports at gmail.com
Sun Dec 15 17:58:33 UTC 2019


But shouldn't the gmp package get a revision bump now once Xcode 11.3 has
been installed on the 10.15 buildbot so that it gets built with stack
checking?

On Sun, Dec 15, 2019 at 11:31 AM Marcus Calhoun-Lopez <mcalhoun at macports.org>
wrote:

> The xcode_workaround-1.0.tcl PortGroup was designed so that the workaround
> is automatically dropped when using Xcode 11.3.
>
> -Marcus
>
> > On Dec 15, 2019, at 8:49 AM, Jack Howarth <howarth.at.macports at gmail.com>
> wrote:
> >
> > FYI, the gmp test suite failure in the t-powm test case when using core2
> optimization with the default stack checking has been solved in the release
> Xcode 11.3. So the current -fno-stack-checking workaround in the MacPorts
> gmp package can be dropped in favor of just blacklisting Xcode 11.0 through
> 11.2.
> >                 Jack
> >
> > On Sat, Nov 9, 2019 at 10:57 AM Marcus Calhoun-Lopez <
> marcuscalhounlopez at gmail.com> wrote:
> > This is an interesting idea, but there may be other factors at play.
> >
> > GMP has definite ideas about which compiler flags to use.
> > When CFLAGS is set, GMP completely ignores its own and assumes the user
> just wants what is in CFLAGS.
> >
> > Without CFLAGS, GMP uses, for example, -O2 -pedantic
> -fomit-frame-pointer -m64 -mtune=core2 -march=core2.
> > When CFLAGS is set, none of these flags are used.
> >
> > Following your lead, I was able to pass all of the tests just by setting
> CFLAGS (*without* -fcheck-stack).
> >
> > -Marcus
> >
> > > On Nov 9, 2019, at 8:19 AM, Jack Howarth <
> howarth.at.macports at gmail.com> wrote:
> > >
> > >
> > >
> > > On Fri, Nov 8, 2019 at 10:17 PM Ryan Schmidt <ryandesign at macports.org>
> wrote:
> > > On Nov 8, 2019, at 20:03, Joshua Root wrote:
> > >
> > > > On 2019-11-9 10:28 , Jack Howarth wrote:
> > > >> I just noticed a really bizarre aspect to this Xcode 11 bug. If you
> > > >> build gmp manually from source with the default compiler flags in
> use...
> > > >>
> > > >> ./configure
> > > >> make
> > > >> make check
> > > >>
> > > >> the t-powm test case fails. However if you add -fstack-check
> > > >> explicitly on CFLAGS and CXXFLAGS, the t-powm test case failure is
> > > >> suppressed.
> > > >>
> > > >> ./configure  CFLAGS='-O2 -fcheck-stack' CXXFLAGS='-O2 -fcheck-stack'
> > > >> make
> > > >> make check
> > > >
> > > > Are the default compiler flags '-O2'?
> > >
> > > MacPorts default optimization flag is -Os.
> > >
> > >
> > > Same behavior...
> > >
> > > ./configure  CFLAGS='-Os -fstack-check' CXXFLAGS='-Os -fstack-check'
> > > make
> > > make check
> > >
> > > suppresses the failure
> > >
> > > PASS: t-powm
> > >
> > > with Xcode 11.2 in a manual build. Wonder how many of the remaining
> packages that are requiring -fno-check-stack could actually be 'fixed' by
> explicitly passing -fcheck-stack instead. Obviously that option disabled
> some of the problematic stack checking code that Apple's clang is
> automatically generating in the absence of the flag.
> > >
> >
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.macports.org/pipermail/macports-dev/attachments/20191215/4f693253/attachment.html>


More information about the macports-dev mailing list