#7095: os x 10.6 port -- numerous mysterious errors caused by weird "abort trap"
issue
-----------------------+----------------------------------------------------
Reporter: was | Owner: tbd
Type: defect | Status: needs_review
Priority: blocker | Milestone: sage-4.3.1
Component: porting | Keywords:
Work_issues: | Author: Craig Citro, John Palmieri, Francis
Clarke, William Stein
Upstream: N/A | Reviewer:
Merged: |
-----------------------+----------------------------------------------------
Changes (by newvalueoldvalue):
* priority: critical => blocker
* status: new => needs_review
* author: => Craig Citro, John Palmieri, Francis Clarke, William Stein
Comment:
Ok, I think I've got this one nailed down. To test, install the new spkgs
here:
http://sage.math.washington.edu/home/craigcitro/sage/SPKGs/python-2.6.4.p2.spkg
http://sage.math.washington.edu/home/craigcitro/sage/SPKGs/matplotlib-0.99.1.p4.spkg
I'm also attaching a patch for the main sage repository which does a
handful of minor cleanup (removes the big 10.6 warning at startup, fixes a
few doctest failures and typos). To test, install python, then matplotlib,
and do a `sage -br` or `sage -ba` (it will need to rebuild everything
regardless, unfortunately). I've done this on my laptop and bsd.math, and
all of the doctests mentioned in this ticket pass. I did a full `-long`
test on my machine, and everything passes with this patch.
Here's what was going on: we built Python with `MACOSX_DEPLOYMENT_TARGET`
set to `10.3`. However, the OSX 10.3 headers are no longer available on
the system, and this leads to Python and its dependencies get compiled
with a mishmash of 10.4 and 10.6 headers. For most files, this isn't a
problem, but Apple does some clever trickery with its system libraries to
maintain multiple copies of system calls with different behavior. For
instance, there are multiple copies of `fopen(1)` floating around, and
which one you're using depends on which copy of `stdio.h` you include. So
in the case of the matplotlib errors above, we ended up with some
confusion between `_fopen` and `_fopen$DARWIN_EXTSN`, where we compiled
`ft2font.o` looking for the first one, but we actually needed the
functionality provided by the second one. The reason that William's patch
for matplotlib was working is that he forced a recompile of `ft2font.o`
(and `ft2font.so`) in an environment where the `MACOSX_DEPLOYMENT_TARGET`
was set differently (i.e. not set at all), which led to the right symbols
getting generated at compile time.
The fix I added was to just set `MACOSX_DEPLOYMENT_TARGET` to the
architecture we're building on at compile-time. This means that we can't
use 10.6 binaries on 10.5, which I don't think anyone would feel safe
about anyway. In theory, we could set it to 10.4, but I don't see why we
would -- it doesn't really make sense for us, since we have a monolithic
distribution.
Many thanks to everyone who commented on this ticket -- in particular,
John Palmieri's simplified matplotlib issue above (the one in pure python)
made debugging this a snap (for some definition of "snap," I guess). I'm
listing John and William as authors, since they were key, and Francis too,
because he actually suggested the right fix above!
--
Ticket URL: <http://trac.sagemath.org/sage_trac/ticket/7095#comment:36>
Sage <http://www.sagemath.org>
Sage: Creating a Viable Open Source Alternative to Magma, Maple, Mathematica,
and MATLAB
--
You received this message because you are subscribed to the Google Groups
"sage-trac" group.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to
[email protected].
For more options, visit this group at
http://groups.google.com/group/sage-trac?hl=en.
