Bugzilla Bug 141206

Hi,

here is an ebuild for compiling pari-2.3.0. The ebuild is copied from
pari-2.1.*, but I had to change a few changes to make pari compile on my
system. I don't know if that will compile for all systems or not. I will add
more use flags to the ebuild soon. (For the record, I'm using gcc 4.1.0, on an
amd64 computer. I like others to test this ebuild on other systems, and send me
feedback.)

Here is a description of pari, copied from their website: 
PARI/GP is a widely used computer algebra system designed for fast computations
in number theory (factorizations, algebraic number theory, elliptic curves...),
but also contains a large number of other useful functions to compute with
mathematical entities such as matrices, polynomials, power series, algebraic
numbers, etc., and a lot of transcendental functions.

Created an attachment (id=92342) [edit]
sci-mathematics/pari-2.3.0.ebuild

*** Bug 121466 has been marked as a duplicate of this bug. ***

Hi Soroosh,

Thank you very much for your ebuild. I've just committed a 
pari-2.3.0 ebuild to portage. It is currently package.mask'ed
pending further testing. Please give it a try and report back.

Thanks,
Markus

Works for me (x86, gcc-3.4.6-r1). Please don't gzip *.pdf documentation.
Andrey

Could this ebuild adopt the 'gmp' USE flag from bug 121466? The GMP kernel is
generally quicker in terms of raw computation throughput, ref:
http://pari.math.u-bordeaux.fr/benchs/timings-int.html

Hi folks,

I just removed pari-2.3.0 from package.mask. Sorry it took so long
but it somehow disappeared from my radar screen. 

@ Marti: As soon as I have some time I will have a look at the gmp 
kernel stuff.

Thanks,
Markus

Reopening ---
On sparc, this version of pari seems to build fine, but at install, we get:
========================================

ln -s Olinux-sparc64/gp-dyn ../gp
make: `../gp' is up to date.
>>> Source compiled.
 * Testing pari kernel ...
make: *** [test-kernel] Error 1                                                
                                  [ !! ]

>>> Install pari-2.3.0 into /var/tmp/portage/pari-2.3.0/image/ category sci-mathematics
Making install in Olinux-sparcv9
Please run Configure first!
make: *** [install] Error 1

!!! ERROR: sci-mathematics/pari-2.3.0 failed.
Call stack:
  ebuild.sh, line 1543:   Called dyn_install
  ebuild.sh, line 1017:   Called src_install
  pari-2.3.0.ebuild, line 73:   Called die

!!! (no error message)
!!! If you need support, post the topmost build error, and the call stack if
relevant.
===============================================
This is with FEATURES=test, but that is not relevant to the failure:
pari has built in Olinux-sparc64, but wants to install from Olinux-sparcv9, and
FEATURES=test fails because of this mismatch.

Comparing with pari-2.1.7-r1 suggests that the correct build directory is
Olinux-sparc64, and the Olinux-sparcv9 is a regression to an earlier failure
which, I believe, Jason Wever fixed once upon a time.

Note that with pari-2.1.7-r1, FEATURES=test doesn't test anything, but if we
make test-compat
make bench
they both run without error on sparc+pari-2.1.7-r1

With pari-2.3.0,  by playing with the files in config so that arch-osname
returns sparc64 instead of sparcv9, things work better, but either the tests
(make bench & make test-compat) nor run for a VERY long time or the never
terminate.  Specifically, for make test-bench, we see
* Testing nfields       for gp-sta..
forever.
Otherwise, benchmark tests with 2.3.0 are significantly faster than with
2.1.7-r1.

The fix for getting 2.3.0 to install on sparc/linux is a matter of getting
everyone to agree on what to call the build directory;  I don't know about the
problems with the tests.  If I get a chance, I'll pursue it, but no promises
unless all tests are known to run successfully on everything else.

And for what it's worth, on amd64, the build is into Olinux-x86_64, which
suggests that the Olinux-sparcv9 is (1) a regression, and (2) incorrect.
On amd64, both 'make bench' and 'make test-compat' run fine (with the nfields
test taking about 4 or 5 times longer than anything else.  If sparc should show
a similar profile, that means nfields should take perhaps 2 or 3 seconds on
sparc, but if it indeed will ever terminate, the run time for is must be
measured in hours.

This ebuild doesn't install the documentation properly. It doesn't set GPDOCDIR
- not sure if that's some new requirement in pari-gp - but it doesn't install
the *.dvi files anyway.

(16:46) gp > ??
Could not find translation file, docdir='/usr/bin' at /usr/bin/gphelp line 152.

(16:46) gp > 

Hi Ferris,

Do these build problems on sparc persists with pari-2.3.2? I would
like to move pari-2.3.2 into stable soon if at all possible.

Thanks much,
Markus

(In reply to comment #10)
> Hi Ferris,
> 
> Do these build problems on sparc persists with pari-2.3.2? I would
> like to move pari-2.3.2 into stable soon if at all possible.

No, nothing has changed between 2.3.0 and 2.3.2. Same error messages.

According to [1] it's the ebuild that needs to be adapted.


[1] http://www.math.u-bordeaux.fr/~belabas/pari/doc/faq.html#sparc64

(In reply to comment #12)
> According to [1] it's the ebuild that needs to be adapted.
> 
> 
> [1] http://www.math.u-bordeaux.fr/~belabas/pari/doc/faq.html#sparc64
> 

If I understand the solution in [1], in the ebuild, in the sparc-specific
portion you would somehow want to force the portable native kernel?

the 'make test-kernel' is trying to test Olinux-sparcv9 but pari has built in
Olinux-sparc64, so of course the test can't find the build directory.

As a test, if by hand I 'ln -s Olinux-sparc64 Olinux-sparcv9' then make
test-kernel prints out OK.

On sparc/linux, you always get 32-bin userland, so perhaps you want it to build
Olinux-sparcv9 --- pari does not seem to contain any assembler code, so that
should be OK.

And indeed, if in the ebuild I force
myhost="sparcv9-linux"
when the ebuild detects sparc, everything seems good.  

That sounds great! Would you be able to whip up a small patch
for sparc that I could apply to the 2.3.2 ebuild?

Thanks,
Markus

Created an attachment (id=130911) [edit]
Allow FEATURES=test to work on sparc for pari-2.3.2

One-line patch to pari-2.3.2.ebuild which builds into directory Olinux-sparcv9
instead of Olinux-sparc64.  For me, at least, this allows a build of pari to
succeed with FEATURES=test.  So far as I can tell, what gets built remains the
same because a pari build uses CFLAGS, and sparc/linux supports only 32-bit
userland.  Olinux-sparc64 suggests an actual 64-bit version of pari.

Thanks much, Ferris! This is now in portage.

Seems like we're done with this bug. Should there be additional
problems with sparc and pari please open a new bug.

Thanks,
Markus