On Dec 17, 2015, at 9:28 AM, Marius Schamschula < address@hidden> wrote:
Sebastian, I was hoping to get feedback on the mailing list but people do not care much about the Mac. Is 4.0.1 working for you? See below.
Sebastian
I have issues with CMatrix.cc-tst on Mac OS X
Integrated test scripts:
liboctave/array/Array.cc-tst ................................ PASS 18/18 liboctave/array/CMatrix.cc-tst ..............................panic: Segmentation fault: 11 -- stopping myself... Segmentation fault: 11
However, I have no problems with 4.0.0 and 4.1.0+ on the same machine, same build script etc. Any idea?
I just built 4.0.1-rc1, based on MacPorts, and I get a different seg-fault as when running make test:
libinterp/dldfcn/__osmesa_print__.cc-tst ....................panic: Segmentation fault: 11 -- stopping myself...
Somehow, the osmesa_print issue doesn’t seem to want to go away.
I guess this was to be expected. The problem is (was?) also present on Linux if you link to a third-party osmesa. My patch https://savannah.gnu.org/patch/download.php?file_id=35085 only fixes compilation issues; segfault remains. However, why not building with "--without-OSMesa"?
My problem seems to come from veclibfort. What blas library do you use?
Sebastian
I forgot to scroll down: Ben has a patch for osmesa for OS X (which I plan to test shortly).
To allow docs to be built for the release candidate, and in the absence of osmesa, you’ll want to apply the changesets below (already applied to default, but not to stable).
These first two fix problems with gnuplot v5
This one backs-up osmesa with gnuplot
Its not obvious that either of these qualify as regressions, so I’m not planning to merge these into stable. However, if there is sufficient testing and support for doing so, I have no objection. In the meantime, I recommend that homebrew, macports, and Fink apply each of these.
Ben
|