20045
Comment: Put up workaround for OSX vs. gdb issue
|
19940
add clean fix for gdb on OSX issue
|
Deletions are marked like this. | Additions are marked like this. |
Line 4: | Line 4: |
The notebook doesn't work unless ctypes are manually build. This is a bug in Python 2.5.2. Need to check Python 2.5.4 and 2.6.1 as well as 3.0 for workarounds. |
|
Line 23: | Line 21: |
ctypes broken, use the old workaround, but also look at 2.5.4 and 2.6.1 for potential solutions | The notebook doesn't work unless ctypes are manually build. This is a bug in Python 2.5.2. * Python 2.6.1 ships with 64 Darwin support, but it is unclear if that is a drop in replacement for Python 2.5.x * Python 2.5.4 does not fix the issue in 2.5.2 The binary _ctypes.so extension was build manually using an older libffi release. See http://sourceware.org/libffi/ - note that the latest release at this point is libffi-3.0.8, released on December 19, 2008. The issue might be fixable by using an external libffi. |
Line 383: | Line 388: |
The above problem is caused by the IPython import. See also http://mail.python.org/pipermail/python-list/2007-February/427200.html for some similar issue. To monkey patch around this in local/lib/python2.5/commands.py change {{{ def getstatusoutput(cmd): """Return (status, output) of executing cmd in a shell.""" import os pipe = os.popen('{ ' + cmd + '; } 2>&1', 'r') text = pipe.read() sts = pipe.close() if sts is None: sts = 0 if text[-1:] == '\n': text = text[:-1] return sts, text }}} to {{{ def getstatusoutput(cmd): """Return (status, output) of executing cmd in a shell.""" import os pipe = os.popen('{ ' + cmd + '; } 2>&1', 'r') text = "" # pipe.read() sts = "" # pipe.close() if sts is None: sts = 0 if text[-1:] == '\n': text = text[:-1] return sts, text }}} With the above change the IPython import throws an excpetion, but Sage works. The fix is obviously to do something clean in IPython related to the readline import. |
The above problem is caused by the IPython import. The problem goes away if we disable the following libedit import test in rlineimpl.py {{{ uses_libedit = False if sys.platform == 'darwin' and have_readline: import commands (status, result) = commands.getstatusoutput( "otool -L %s | grep libedit" % _rl.__file__ ) if status == 0 and len(result) > 0: # we are bound to libedit - new in Leopard _rl.parse_and_bind("bind ^I rl_complete") print "Leopard libedit detected." uses_libedit = True }}} I will open a ticket for this problem shortly. The issue has been reported to the ipython-dev list. |
Sage 64 bit OSX 10.5 build notes for Sage 3.2.3
Notes
fortran.spkg
Use OSX-gfortran.spkg (coming soon) which is a repackaged Gfortran 4.2.3 by the R people. It is quad universal.
Make local/bin/sage_fortran like
gfortran -m64 "$@"
Make sure to chmod 755 and extend DYLD_LIBRARY_PATH to include the Fortran runtime - maybe in toolchain?
gmp
use eMPIRe
Python
The notebook doesn't work unless ctypes are manually build. This is a bug in Python 2.5.2.
- Python 2.6.1 ships with 64 Darwin support, but it is unclear if that is a drop in replacement for Python 2.5.x
- Python 2.5.4 does not fix the issue in 2.5.2
The binary _ctypes.so extension was build manually using an older libffi release. See http://sourceware.org/libffi/ - note that the latest release at this point is libffi-3.0.8, released on December 19, 2008.
The issue might be fixable by using an external libffi.
atlas
The following error pops up, but it is harmless:
/bin/sh: /Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/bin/which_fortran: No such file or directory
It should still be fixed.
pynac
No OSX64 bit support yet - an unclean version that builds would be:
diff -r 56d7250fb1dc spkg-install --- a/spkg-install Wed Oct 15 10:32:53 2008 +0200 +++ b/spkg-install Tue Jan 06 13:40:13 2009 -0800 @@ -22,8 +22,10 @@ build_pynac() { cd ${PYNACDIR} - CPPFLAGS=-I"$SAGE_LOCAL"/include/python2.5 PKG_CONFIG_PATH=${SAGE_LOCAL}/lib/pkgconfig ./configure --disable-static --prefix=${SAGE_LOCAL} - CPPFLAGS=-I"$SAGE_LOCAL"/include/python2.5 $MAKE + LDFLAGS="-m64"; export LDFLAGS [needs to be 64 bit OSX specific] + PKG_CONFIG_PATH="$SAGE_LOCAL"/lib/pkgconfig/: export PKG_CONFIG_PATH + CPPFLAGS="-m64 -I\"$SAGE_LOCAL\"/include/python2.5" ./configure --disable-static --prefix=${SAGE_LOCAL} + CPPFLAGS="-m64 -I\"$SAGE_LOCAL\"/include/python2.5" $MAKE if [ $? -ne 0 ]; then echo "Error building pynac." exit 1
r
libpng trouble since the build with a dynamic libpng was fixed recently:
Error in dyn.load(file, ...) : unable to load shared library '/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/spkg/build/r-2.6.1.p21/src/library/grDevices/libs/grDevices.so': dlopen(/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/spkg/build/r-2.6.1.p21/src/library/grDevices/libs/grDevices.so, 6): Symbol not found: __cg_png_create_info_struct Referenced from: /System/Library/Frameworks/ApplicationServices.framework/Versions/A/Frameworks/ImageIO.framework/Versions/A/ImageIO Expected in: /Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib//libPng.dylib Calls: <Anonymous> ... namespaceImport -> loadNamespace -> library.dynam -> dyn.load Execution halted
The fix here is to analog to 32 bits to disable aqua support for now.
sagelib
sage -b: devel/sage/c_lib/SConstruct
## We want the debug and optimization flags, since debug symbols are so useful, etc. env.Append( CFLAGS="-O2 -g -m64" ) env.Append( CXXFLAGS="-O2 -g -m64" ) env.Append( LINKFLAGS="-m64 -single_module -flat_namespace -undefined dynamic_lookup" )
The above fix needs to be dependent on SAGE64 being set. With the above flags set libcsage.dylib is build as a 64 bit library. The spkg-install does the linking, so everything should work on OSX 64 bit once a clean version of the above proto-patch is merged.
Another problem is that the memory.pyx extension needs to be linked against libcsage:
--- a/module_list.py Mon Jan 05 23:03:45 2009 -0800 +++ b/module_list.py Tue Jan 06 17:13:03 2009 -0800 @@ -854,7 +854,7 @@ Extension('sage.rings.memory', sources = ['sage/rings/memory.pyx'], - libraries=['gmp','stdc++']), + libraries=['csage','ntl','gmp','stdc++']), Extension('sage.rings.morphism', sources = ['sage/rings/morphism.pyx']),
Otherwise Sage starts up, but doctesting fails always with
sage -t "devel/sage/sage/rings/complex_interval_field.py" Traceback (most recent call last): File "/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/tmp/complex_interval_field.py", line 2, in <module> from sage.all_cmdline import *; File "/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib/python2.5/site-packages/sage/all_cmdline.py", line 14, in <module> from sage.all import * File "/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib/python2.5/site-packages/sage/all.py", line 56, in <module> from sage.rings.memory import pmem_malloc ImportError: dlopen(/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib/python2.5/site-packages/sage/rings/memory.so, 2): Symbol not found: _rand_n Referenced from: /Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib/python2.5/site-packages/sage/rings/memory.so Expected in: dynamic lookup
Similar linker fixes are needed for Cygwin
ghmm
ghmm is build 32 bits only
libghmm.1.0.0.dylib: Mach-O dynamically linked shared library i386 libghmm.1.dylib: Mach-O dynamically linked shared library i386 libghmm.dylib: Mach-O dynamically linked shared library i386
Quick and ugly hack, needs SAGE64 specific cleanup:
Sprocketer:ghmm-20080813 michaelabshoff$ hg diff diff -r 9d3e399ae771 spkg-install --- a/spkg-install Sun Aug 03 18:34:26 2008 -0700 +++ b/spkg-install Tue Jan 06 15:31:08 2009 -0800 @@ -7,6 +7,9 @@ # this does *NOT* work. First the configure.in script is hardcoded to # ignore that option. Second, if you change that then it turns out # that ghmm doesn't even compile. + +export CFLAGS="-O2 -g -m64" +export LDFLAGS="-m64" ./configure --prefix="$SAGE_LOCAL"
boehmgc
Also is build as 32 bit lib at the moment. Should be trivial to fix via something along the lines of
+export CFLAGS="-O2 -g -m64" +export LDFLAGS="-m64"
in spkg-install.
singular
We need to enable system malloc for MacIntel since it is only chosen pre default for MacPPC. The ugly hack below needs to be cleaned up:
diff -r 0e3b296b487a spkg-install --- a/spkg-install Sun Nov 30 23:10:51 2008 -0800 +++ b/spkg-install Tue Jan 06 16:46:38 2009 -0800 @@ -116,7 +116,8 @@ --enable-IntegerProgramming \ --disable-doc \ "$DYNAMIC_KERNEL" \ - "$ENABLE_DEBUG" + "$ENABLE_DEBUG" \ + --with-malloc=system if [ $? -ne 0 ]; then echo "Unable to configure Singular."
But at exit we are having problems with double frees:
Sprocketer:sage-3.2.3-64bit michaelabshoff$ ./sage ---------------------------------------------------------------------- | Sage Version 3.2.3, Release Date: 2009-01-05 | | Type notebook() for the GUI, and license() for information. | ---------------------------------------------------------------------- sage: import sys sage: sys.maxint 9223372036854775807 sage: !uname Darwin sage: Exiting SAGE (CPU time 0m0.07s, Wall time 0m13.65s). sage.bin(162) malloc: *** error for object 0x106592030: Non-aligned pointer being freed (2) *** set a breakpoint in malloc_error_break to debug sage.bin(162) malloc: *** error for object 0x106592038: Non-aligned pointer being freed *** set a breakpoint in malloc_error_break to debug sage.bin(162) malloc: *** error for object 0x106591048: Non-aligned pointer being freed *** set a breakpoint in malloc_error_break to debug sage.bin(162) malloc: *** error for object 0x106591030: Non-aligned pointer being freed (2) *** set a breakpoint in malloc_error_break to debug Sprocketer:sage-3.2.3-64bit michaelabshoff$ ./sage
These also happen on Cygwin, so this needs to be investigated and fixed.
The problem is also reproducible on 32 bit OSX when compiling Singular to use the system's malloc:
Sprocketer:sage-3.2.3.alpha0 michaelabshoff$ ./sage ---------------------------------------------------------------------- | Sage Version 3.2.3.alpha0, Release Date: 2008-12-24 | | Type notebook() for the GUI, and license() for information. | ---------------------------------------------------------------------- sage: Exiting SAGE (CPU time 0m0.07s, Wall time 0m16.28s). sage.bin(90498) malloc: *** error for object 0x5335018: Non-aligned pointer being freed *** set a breakpoint in malloc_error_break to debug sage.bin(90498) malloc: *** error for object 0x5335020: Non-aligned pointer being freed (2) *** set a breakpoint in malloc_error_break to debug sage.bin(90498) malloc: *** error for object 0x5333078: Non-aligned pointer being freed *** set a breakpoint in malloc_error_break to debug sage.bin(90498) malloc: *** error for object 0x5333068: Non-aligned pointer being freed *** set a breakpoint in malloc_error_break to debug Sprocketer:sage-3.2.3.alpha0 michaelabshoff$
The problem might be a similar issue as the scope problem found in Cython 0.11-devel in mid January. Note that the majority of the doctests below fail because minpoly() in libSingular segfaults in conditions where it should not be called in the first place.
tachyon
Fails due to 64 bit build problems - known problem with ticket
flintqs
Fails due to 64 bit build problems - known problem with ticket
cvxopt
Linker trouble - no clue why yet.
gcc -m64 -L/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib -bundle -undefined dynamic_lookup build/temp.macosx-10.3-i386-2.5/C/base.o build/temp.macosx-10.3-i386-2.5/C/dense.o build/temp.macosx-10.3-i386-2.5/C/sparse.o -L/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib -lm -llapack -lblas -lgfortran -o build/lib.macosx-10.3-i386-2.5/cvxopt/base.so ld: library not found for -lgfortran collect2: ld returned 1 exit status
The build issue goes away when installing gfortran directly into $SAGE_LOCAL, but it might also be a failure due to missing gfortran due to env issues. This needs to be verified.
scipy
Import issue with d1mach: Note that d1mach.o manages to be 32 bits:
Sprocketer:scipy-20071020-0.6.p4 michaelabshoff$ file ./src/build/temp.macosx-10.3-i386-2.5/scipy/integrate/mach/d1mach.o ./src/build/temp.macosx-10.3-i386-2.5/scipy/integrate/mach/d1mach.o: Mach-O object i386
This causes import failures all over the place as Scipy's special functions module does fail to import. This seems to be due to scipy using gfortran in some cases when it finds it instead of sage_fortran:
Found executable /Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/bin/gfortran Found executable /usr/bin/ld Found executable /usr/bin/ar Found executable /usr/bin/ranlib compiling Fortran sources Fortran f77 compiler: /Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/bin/gfortran -Wall -ffixed-form -fno-s econd-underscore -fPIC Fortran f90 compiler: /Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/bin/gfortran -Wall -fno-second-undersc ore -fPIC Fortran fix compiler: /Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/bin/gfortran -Wall -ffixed-form -fno-s econd-underscore -Wall -fno-second-underscore -fPIC creating build/temp.macosx-10.3-i386-2.5/scipy/integrate/mach compile options: '-c' gfortran:f77: scipy/integrate/mach/d1mach.f gfortran:f77: scipy/integrate/mach/i1mach.f gfortran:f77: scipy/integrate/mach/r1mach.f gfortran:f77: scipy/integrate/mach/xerror.f
One workaround is to use a dummy gfortran wrapper.
sympow
For now sympow is build in 32 bit mode, but it works:
Sprocketer:sage-3.2.3-64bit michaelabshoff$ file local/lib/sympow/sympow local/lib/sympow/sympow: Mach-O executable i386
Since OSX will always have a 32 bit runtime this is something we should be able to live with.
Doctest Failures
Currently 22 doctest failures:
sage -t "devel/sage/sage/calculus/calculus.py" sage -t "devel/sage/sage/calculus/wester.py" sage -t "devel/sage/sage/crypto/mq/sr.py" sage -t "devel/sage/sage/groups/generic.py" sage -t "devel/sage/sage/interfaces/lisp.py" sage -t "devel/sage/sage/interfaces/qsieve.py" sage -t "devel/sage/sage/libs/pari/gen.pyx" sage -t "devel/sage/sage/rings/polynomial/laurent_polynomial.pyx" sage -t "devel/sage/sage/rings/polynomial/multi_polynomial.pyx" sage -t "devel/sage/sage/rings/polynomial/multi_polynomial_element.py" sage -t "devel/sage/sage/rings/polynomial/multi_polynomial_ideal.py" sage -t "devel/sage/sage/rings/polynomial/multi_polynomial_ideal_libsingular.pyx" sage -t "devel/sage/sage/rings/polynomial/multi_polynomial_libsingular.pyx" sage -t "devel/sage/sage/rings/polynomial/toy_buchberger.py" sage -t "devel/sage/sage/schemes/elliptic_curves/ell_finite_field.py" sage -t "devel/sage/sage/schemes/elliptic_curves/ell_generic.py" sage -t "devel/sage/sage/schemes/elliptic_curves/ell_number_field.py" sage -t "devel/sage/sage/schemes/elliptic_curves/ell_point.py" sage -t "devel/sage/sage/schemes/generic/affine_space.py" sage -t "devel/sage/sage/schemes/hyperelliptic_curves/hyperelliptic_finite_field.py" sage -t "devel/sage/sage/tests/book_stein_ent.py"
A lot of the above are segaults caused by the tentative Singular vs. Givaro collision.
pre-scipy fix and install of R
41 Doctest failures, but many are easily fiaxable:
sage -t devel/sage/sage/calculus/wester.py # 3 doctests failed sage -t devel/sage/sage/calculus/calculus.py # 0 doctests failed sage -t devel/sage/sage/crypto/mq/sr.py # Segfault sage -t devel/sage/sage/finance/time_series.pyx # 5 doctests failed sage -t devel/sage/sage/functions/special.py # 5 doctests failed sage -t devel/sage/sage/functions/transcendental.py # 5 doctests failed sage -t devel/sage/sage/groups/generic.py # Segfault sage -t devel/sage/sage/interfaces/expect.py # 11 doctests failed sage -t devel/sage/sage/interfaces/lisp.py # 4 doctests failed sage -t devel/sage/sage/interfaces/qsieve.py # 7 doctests failed sage -t devel/sage/sage/interfaces/r.py # 176 doctests failed sage -t devel/sage/sage/libs/pari/gen.pyx # 2 doctests failed sage -t devel/sage/sage/modular/congroup.py # 14 doctests failed sage -t devel/sage/sage/modular/cusps.py # 1 doctests failed sage -t devel/sage/sage/modular/modsym/ambient.py # 4 doctests failed sage -t devel/sage/sage/modular/modsym/boundary.py # 2 doctests failed sage -t devel/sage/sage/modules/vector_double_dense.pyx # 12 doctests failed sage -t devel/sage/sage/modules/vector_real_double_dense.pyx # 1 doctests failed sage -t devel/sage/sage/numerical/test.py # 5 doctests failed sage -t devel/sage/sage/numerical/optimize.py # 4 doctests failed sage -t devel/sage/sage/plot/plot3d/list_plot3d.py # 6 doctests failed sage -t devel/sage/sage/rings/arith.py # 3 doctests failed sage -t devel/sage/sage/rings/integer.pyx # 4 doctests failed sage -t devel/sage/sage/rings/polynomial/laurent_polynomial.pyx # Segfault sage -t devel/sage/sage/rings/polynomial/multi_polynomial.pyx # Segfault sage -t devel/sage/sage/rings/polynomial/multi_polynomial_element.py # Segfault sage -t devel/sage/sage/rings/polynomial/multi_polynomial_ideal_libsingular.pyx # Segfault sage -t devel/sage/sage/rings/polynomial/multi_polynomial_ideal.py # 1 doctests failed sage -t devel/sage/sage/rings/polynomial/multi_polynomial_libsingular.pyx # Segfault sage -t devel/sage/sage/rings/polynomial/toy_buchberger.py # Segfault sage -t devel/sage/sage/rings/polynomial/toy_d_basis.py # 2 doctests failed sage -t devel/sage/sage/schemes/elliptic_curves/ell_generic.py # Segfault sage -t devel/sage/sage/schemes/elliptic_curves/ell_finite_field.py # Segfault sage -t devel/sage/sage/schemes/elliptic_curves/ell_number_field.py # Segfault sage -t devel/sage/sage/schemes/elliptic_curves/ell_point.py # Segfault sage -t devel/sage/sage/schemes/generic/affine_space.py # Segfault sage -t devel/sage/sage/schemes/hyperelliptic_curves/hyperelliptic_finite_field.py # Segfault sage -t devel/sage/sage/stats/r.py # 1 doctests failed sage -t devel/sage/sage/stats/test.py # 1 doctests failed sage -t devel/sage/sage/tests/book_stein_modform.py # 4 doctests failed sage -t devel/sage/sage/tests/book_stein_ent.py # 3 doctests failed
Reasons for failures:
- different gmp, i.e. eMPIRe, accounts for 10 failures
- the missing compoenents also cause failures
- Singular does double frees at exit on occasion due to the use of system malloc, a patch to work around those has been done
gdb problems
It seems that ipyhon is causing trouble when starting Sage in 64 bit mode on OSX under gdb:
(gdb) r Starting program: /Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/bin/python /Users/michaelabshoff/Desktop/sage-3.2.3-64bit/tmp/.doctest_ell_finite_field.py warning: posix_spawn failed, trying execvp, error: 86 Traceback (most recent call last): File "/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/tmp/.doctest_ell_finite_field.py", line 2, in <module> from sage.all_cmdline import *; File "/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib/python2.5/site-packages/sage/all_cmdline.py", line 14, in <module> from sage.all import * File "/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib/python2.5/site-packages/sage/all.py", line 64, in <module> from sage.misc.all import * # takes a while File "/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib/python2.5/site-packages/sage/misc/all.py", line 16, in <module> from sage_timeit_class import timeit File "sage_timeit_class.pyx", line 3, in sage.misc.sage_timeit_class (sage/misc/sage_timeit_class.c:603) File "/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib/python2.5/site-packages/sage/misc/sage_timeit.py", line 12, in <module> import timeit as timeit_, time, math, preparser, interpreter File "/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib/python2.5/site-packages/sage/misc/interpreter.py", line 95, in <module> import IPython.ipapi File "/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib/python2.5/site-packages/IPython/__init__.py", line 57, in <module> __import__(name,glob,loc,[]) File "/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib/python2.5/site-packages/IPython/ipstruct.py", line 22, in <module> from IPython.genutils import list2dict2 File "/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib/python2.5/site-packages/IPython/genutils.py", line 118, in <module> import IPython.rlineimpl as readline File "/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib/python2.5/site-packages/IPython/rlineimpl.py", line 37, in <module> (status, result) = commands.getstatusoutput( "otool -L %s | grep libedit" % _rl.__file__ ) File "/Users/michaelabshoff/Desktop/sage-3.2.3-64bit/local/lib/python2.5/commands.py", line 54, in getstatusoutput text = pipe.read() IOError: [Errno 4] Interrupted system call Program exited with code 01.
The above problem is caused by the IPython import. The problem goes away if we disable the following libedit import test in rlineimpl.py
uses_libedit = False if sys.platform == 'darwin' and have_readline: import commands (status, result) = commands.getstatusoutput( "otool -L %s | grep libedit" % _rl.__file__ ) if status == 0 and len(result) > 0: # we are bound to libedit - new in Leopard _rl.parse_and_bind("bind ^I rl_complete") print "Leopard libedit detected." uses_libedit = True
I will open a ticket for this problem shortly. The issue has been reported to the ipython-dev list.