Differences between revisions 30 and 31
Revision 30 as of 2008-11-14 13:41:56
Size: 4913
Editor: anonymous
Comment: converted to 1.6 markup
Revision 31 as of 2009-01-08 08:07:48
Size: 4836
Comment: move build pages to remove the extra "solaris" directory, add 3.2.3 page
Deletions are marked like this. Additions are marked like this.
Line 18: Line 18:
 * [[/solaris/sage-3.0.5|Sage 3.0.5 build issues and doctest failures]]
 * [[/solaris/sage-3.1.3|Sage 3.1.3 build issues and doctest failures]]
 * When switching to the Sun Forte cc make sure to disable the use of {{{-fast}}} for GMP since it leads to test failures with MPFR for example.
 * [[solaris/sage-3.2.3|Sage 3.2.3 build issues and doctest failures]]
 * [[solaris/sage-3.1.3|Sage 3.1.3 build issues and doctest failures]]
 * [[solaris/sage-3.0.5|Sage 3.0.5 build issues and doctest failures]]

The Solaris Port of Sage

This page tracks the Solaris port of Sage.

Goal

  • Get Sage compiling and running with recent Sun Forte Compiler on Solaris 10 and higher on x86/x86-64/sparc

Status

  • mabshoff is currently working on a 32 bit build on Solaris 9/Sparc. It should run on Solaris 10 and non-Sparc CPUs supported by Solaris, i.e. x86 and x86-64. There are a few build fixes and package issues to work around. But due to some issues in the coercion code/libSingular many doctests that involve mv polynomials fail.

Short Term Plan

  • build Sage 3.1.3 with gcc 4.2.4 + custom binutils in 32 bit mode
  • add the needed spkg-install fixes to build Sage 3.2 in 64 bit mode with gcc 4.2.4

Build Fixes to Integrate

Open Issues

  • clisp fails to compile, but clisp 2.44.1 builds fine with gcc 3.4.6. It segfaults during "make check". This has been reported to clisp-devel and we are debugging the problem [clips will be replaced by ecl in the future. If clisp cannot be fixed and Maxima will not be ported to ecl this will likely mean that the replacement of Maxima will be accelerated]
  • libfplll's Solaris workaround for round() is wrong for negative numbers. This was pointed out by Paul Zimmermann. The other two patches have been merged upstream and will be in the next libfplll release [The workaround is only needed for Solaris 9. Since our goal is now Solaris 10 and above fixing this is not a high priority]
  • Symmetrica seems to segfault on Sparc [this might be related to coercion, so until proven Symmetrica is innocent. We have requested the alleged Symmetrica test suite to test this out, but nothing so far]
  • givaro-3.2.10.rc3.spkg includes fenv.h in src/src/kernel/zpz/givgfq.inl - this is an issue on Solaris 9, but works fine on Solaris 10
  • log2 isn't defined in math.h, but sun's math.h

History of efforts

  • 3.1.3 builds on SkyNet/Fulvia with custom toolchain
  • 3.0.1-3.0.5: misc fixes
  • 3.0:
    • fix lcalc build on Solaris
  • 2.11:
    • add Solaris 9 specific stdint.h workaround since Solaris 9 only supports a draft standard of C99. Solaris 10 has no such problem - 32 bits only so far

  • 2.10.4:
    • no significant changes
  • 2.10.3
    • start of integrating 64 bit build support - this will take a while to get in
  • 2.10.2:
    • same as 2.10.1 since the emphasis wasn't on this port this release.
  • 2.10.1:
    • drop in of stdint.h in $SAGE_LOCAL/include makes everything compile up to libfplll
    • fix libfplll build issue [spkg at 1957, merged in 2.10.1.rc2]
    • known issue: ATLAS screws up the combined Lapack+ATLAS dynamic library build. Maybe this is ld related?
    • because of the above issue numpy cannot be imported properly. Creating an liblapack.so manually from liblapack.a and rebuilding numpy makes numpy work

    • rpy doesn't build: error: conflicting types for 'uintptr_t' - probable cause is 32 bit vs 64 bit issues

    • fix rubiks build issue [spkg at 1959, merged in 2.10.1.rc2]
    • apply sage-lib build fixes from #1328 and #1329. Make sure to copy solaris_fix.h also into sage/rings [merged in 2.10.1.rc3]
  • 2.8.15:
    • The release has many Solaris specific fixes, but we need to fix #1325 before we will have something that can be used as a beta build.
  • 2.8.14:
    • new bug in matrix code that provides a special class for multivariate polynomials as entries causes segfaults all over the map.
    • int_fast64_t needs to be 4 bytes large, otherwise Sage throws an exception at startup. This needs fixing and actually causes doctest failures with the modp code
    • libfplll miscompiles, fix clear, but needs to be merged upstream
    • FLINT miscompiles, fixed upstream
    • lcalc miscompiles, fixes need to be integrated
    • clisp compiles, but segfaults. Solution: use 2.39 binary until issue is solved
    • cvxopt doesn't work (complex.h missing)
    • 2-3 patches are needed to sage.spkg - easy fixes, need to be merged upstream
    • failed doctest: number of partitions broken - maybe fpu precision issues
    • failed doctest: Schubert polynomial triggers error in symmetrica
  • 2.8.3:
    • worked reasonably well after workarounds, about 10-12 doctest failures
    • work arounds needed: clisp, lcalc
    • missing/failed spkgs: cvxopt
  • pre 2.8:
    • various people (David Kirkby) worked on this, but I am not sure if Sage ever passed 100% of doctests

ToDo

  • make doctests pass 100% :)

  • open track tickets for all known issue
  • make building toolchain automatic
  • find way to make Sage aware of the toolchain