Differences between revisions 16 and 17
Revision 16 as of 2010-08-24 20:18:47
Size: 4923
Editor: leif
Comment: Added comments regarding #9722. Updated debug instructions to refer to PARI 2.4.3.svn-12577.p4 (from p3).
Revision 17 as of 2010-08-24 20:29:00
Size: 4923
Editor: leif
Comment: SAGE_PARI_tune -> SAGE_TUNE_pari
Deletions are marked like this. Additions are marked like this.
Line 55: Line 55:
 * Tuning (i.e. building with {{{SAGE_PARI_tune=yes}}}) is completely broken.  * Tuning (i.e. building with {{{SAGE_TUNE_pari=yes}}}) is completely broken.

New version of PARI/GP: 2.4.3 svn-12577

Relevant tickets

  • 9343: PARI upgrade

  • 9591: genus2reduction

  • 9592: lcalc (see also 9775 for a later version of lcalc)

  • 9722: PARI/GP build error on Fedora 13 (Merged into Sage 4.5.3.alpha2. I'm currently porting the changes to the PARI 2.4.3.svn-12577.p4 spkg. -Leif)

Files

If you want to upgrade an existing Sage, you need the install 3 spkg's and apply 2 patches. Do this in order as in the list below. In particular, the genus2reduction spkg will not install without the new pari spkg.

For reviewers

  • Download sage-4.6.prealpha3.tar, which is a complete Sage distribution based on sage-4.5.3.alpha1. Just extract it, and type make. In this file, the following tickets have been merged:

    • 9343 (needs_work): PARI upgrade

    • 9591 (needs_review): genus2reduction

    • 9592 (needs_review): lcalc

    • 9750 (positive_review): Document that PARI no longer assumes more than GRH

    • 9636 (needs_review): Catch output from PARI in Sage

    • 9400 (needs_review): modify the NumberField constructor to pass in optional integer B such that all the internal pari routines will replace the discriminant by its gcd with B, making some things massively faster

      • This ticket is not strictly related to the PARI upgrade, but it changes a lot of code using PARI. It also fixes an issue with nfinit() and wraps some additional functions in sage/libs/pari/gen.pyx.

    • 9703 (closed): Doctest failures caused by non-working sympow on 32-bit Solaris x86 and 32-bit OpenSolaris

    • 9735 (closed): Numerical noise on sage/stats/hmm/chmm.pyx on both Solaris x86 and OpenSolaris x86

      • The above two tickets have nothing to with PARI, but they should make Sage work 100% on Solaris and OpenSolaris.

  • The main issue seems to be making sure it actually compiles.

    • Once it compiles, you can run the PARI/GP test suite by doing

      •      env SAGE_CHECK=yes ./sage -f pari-2.4.3.svn-12577.p4.spkg
  • For reference, these are the patches for the spkg's:

Debugging PARI/GP build errors

If the new pari spkg doesn't build, you should try building PARI/GP stand-alone (i.e. outside of Sage). You should download pari-2.4.3.svn-12577.p4.spkg and then type the following in a shell:

tar xjf pari-2.4.3.svn-12577.p4.spkg
cd pari-2.4.3.svn-12577.p4/src
./Configure --graphic=none
make gp

If this works, there is a bug in Sage's build process for PARI/GP. If this gives errors, this indicates an upstream issue with PARI/GP. In any case, any errors should be reported on the Sage trac.

Known issues

  • Tuning (i.e. building with SAGE_TUNE_pari=yes) is completely broken.

  • Build issues on Fedora 13: see 9722. (Not all are limited to Fedora 13, i.e. could in principle occur on other systems as well.)