2011/3/9 Francois Bissey <[email protected]>:
>>   This may be of interest
>>
>>
>> https://www.theopensourceway.org/wiki/How_to_tell_if_a_FLOSS_project_is_doo
>> med_to_FAIL
>>
>> and should explain why sage is not included and/or built from
>> sources in any major distro. Well, I make rpms for Mandriva,
>> and there is the gentoo port...
>>
>
> My turn? :)

:-)

>>   My last work on a sagemath 4.6.2 rpm ended with these
>> main issues:
>>
>> o I am still preferring to use system python. But making
>>   sage build/work with python 2.7.1 has pitfalls as sage
>>   expects python 2.6
>
> I don't have time to work on python-2.7.1 but yes I am expecting issues.
> By the way folks 2.6.6 with pickling patch works for us.

  I am still doing an extra python (2.7.1) build during sage build,
and then install the patched cPickle.so and pickle.py in
$PYTHONPATH (/usr/share/sage/site-packages).

>> o Cannot build with with cython 0.14, so the does a cython
>>   0.13 build and puts it in the path first, and also installs
>>   a cython binary in $SAGE_LOCAL/bin and runtime at
>>   $PYTHONPATH
>
> The 4.6.2 I ship uses cython-0.14.1, then again I know why I wanted to
> review the relevant ticket. I also use ecl-11.1.1/maxima-5.23.2 in advance
> of sage-4.7 - I also was involved in the relevant tickets. Admitely it is a
> lot of work. I also use mpmath-0.17 but I believe you do things differently
> by using mpmath from sympy - or have you changed that?

  I preferred to have a build of cython 0.13 in $SAGE_LOCAL/bin
and $PYTHONPATH because sage itself does build, but
fails most doctests if using cython 0.14. This way, cython
invoked from within sage or sage -sh will (should) use
cython 0.13.

  Also using ecl-11.1.1 but right now maxima 5.23.0.
I usually use maxima with other lisp backends, i.e. I
test with clisp, gcl, ecl, and sbcl (never know which
one is active :-)

>> o Since sagemath 4.6.1, there are crashes at exit. I
>>   reworked a good share of dependencies, but it still
>>   fails. glibc/valgrind shows a double mpz_clear call
>>   when python exits, so, I now set MALLOC_CHECK_=1
>>   so that it usually should just print a warning about
>>   double free, instead of aborting at exit.
>>
>
> Didn't see that. But we have a horror story from glibc:
> https://github.com/cschwan/sage-on-gentoo/issues#issue/40
> My presumption is that vanilla sage is protected from the problem
> thanks to its antique python or some of its configuration options.

  Wow, very nice catch! I just opened a bug report for
mandriva, requesting that patch on our glibc, as the test
case also triggers the failure. Probably my sagemath
package is not having the same symptom because I
disabled tls support in pari, as that looked like a source
of major problems when I updated to pari-2.4.3.alpha, but,
the glibc/tls issue may be the cause of the problem I
noticed on exit of some doctests.

  Using MALLOC_CHECK_=1 now I see something like:
-%<-
sage -t -force_lib "devel/sage/sage/homology/chain_complex.py"
*** glibc detected *** python: free(): invalid pointer: 0x0000000003ca3960 ***
*** glibc detected *** python: free(): invalid pointer: 0x0000000003ca1270 ***
*** glibc detected *** python: free(): invalid pointer: 0x0000000003e09c20 ***
         [3.7 s]

----------------------------------------------------------------------
All tests passed!
Total time for all tests: 3.7 seconds
-%<-
instead of a crash and a doctest failure...

>>   BTW, I tried to be impartial, but still accounted sage at
>> 280 points.
>>
>
> As has been said before it depends of the objectives of the project.
> Currently packager friendliness is not part of it. You, me, the
> few other people working with me and possibly with you, are really
> going beyond the call of duty.
>
> What would be nice: better recognition that we are doing some
> hard work that may an impact on the quality of sage in the long term
> much like portability work. At present time you can use a gentoo
> prefix to build sage on other linuxes and OS X, windows is the next
> frontier (but that require some kind of windows box, I may have
> something in the near future).

  I have been considering for some time to work on some support,
or at least better understanding gentoo prefix... Hope to get
something like a rpm to setup anything needed in Mandriva at
some point.

> Francois
>

Paulo

-- 
To post to this group, send an email to [email protected]
To unsubscribe from this group, send an email to 
[email protected]
For more options, visit this group at http://groups.google.com/group/sage-devel
URL: http://www.sagemath.org

Reply via email to