windows/general-package-analysis

Clear message

This page is part the [:windows:Sage on Windows port].

Package Analysis

Below you will find a list of packages in Sage as well as remarks on issues that will potentially come up during the Windows port. As we progress we will add detailed problem reports and how we resolved those issues at individual pages.

NOTE: [http://www.apcocoa.org/wiki/ApCoCoALib:CompilationInstructions Some other useful information]

• dir-0.1 - requires POSIX shell
• prereq-0.3 - requires POSIX shell - small build system issues

• bzip2-1.0.4 - port exists: http://gnuwin32.sourceforge.net/packages/bzip2.htm. Since that port uses MinGW we need to either use MinGW 64 bit or do an MSVC port ourselves. a sage without bzip2 would be possible...; saved objects under unix/osx wouldn't load though.

• sage_scripts-2.10.2.alpha0 - requires POSIX shell

• conway_polynomials-0.2 - platform independent, i.e. no known Windows specific issues trivial -- just a pickle

• gmp-4.2.1.p12 - port by Brian Gladman exists - http://fp.gladman.plus.com/computing/gmp4win.htm. The main disadvantage is that the build is done via project file. A fallback option is to use the pure C fall back code via MSVC or Intel's C compiler since MSVC doesn't do the right inline assembly. Brian's gmp works perfectly in Win64 mode. good!

• termcap-1.3.1 - port exists: http://gnuwin32.sourceforge.net/packages/termcap.htm. Since that port uses MinGW we need to either use MinGW 64 bit or do an MSVC port ourselves.

• readline-5.2.p0 - port exists: http://gnuwin32.sourceforge.net/packages/readline.htm. Since that port uses MinGW we need to either use MinGW 64 bit or do an MSVC port ourselves.

• pari-2.3.3: Some build support exists in tree, but it is currently broken. Support from the maintainers exists and they are willing to integrate fixes upstream. The main issue is to get the build system play nice for MinGW+MSVC.

• ntl-5.4.1.p11: works fine with MSVC. Some functions like the time functions use old and crappy Win95+ interfaces. That can be easily fixed. It is very difficult to get fixes merged upstream, but the potential diff is rather small. The build requires Perl for a tuned build upstream merging probably isn't as hard as you think. The NTL author is a number theorist that many of "us", e.g,. John Cremona, know well.

• eclib-20080127.p0: Works fine on Cygwin, but needs a port to MSVC. Port is probably of medium complexity. A merge of fixes upstream is welcome, but maintainer no longer uses Windows. Maintainer = John Cremona = has never used Windows, as far as I know. Porting is reasonable.

• graphs-20070722 - graph database, so no porting issues
• elliptic_curves-0.1 - elliptic curve database, so no porting issues
• extcode-XXX: various code in interpreted languages, i.e. javascript, java, Magma, pari, singular, no porting issues

• flint-1.06.p1: C99 based code, build system easy enough, Windows specific fixes welcome upstream. Hooked deeply into GMP. Main developer -- Bill Hart -- uses Windows all the time.

• zlib-1.2.3.p3: port exists: http://gnuwin32.sourceforge.net/packages/zlib.htm Microsoft ships a copy under some name, so we might be able to "just" use the system one.

• sqlite-3.5.3.p1 - pure Windows port exists (see http://www.sqlite.org/download.html), but we need ???

• libgpg_error-1.6.p0, libgcrypt-1.4.0.p0, opencdk-0.6.6, gnutls-2.2.1.p1: Shortly the spkgs for libgpg_error, libgcrypt and opencdk will disappear since gnutls now includes copies of them. gnutls has build support for 32 bit MSVC, the feasability of the 64 bit build needs to be investigated. Problems might crop up with inline assembly. maybe there is a native windows library that provides similar functionality. This is only needed so Twisted works with Python, and I'm *sure* that is already very well supported.

• libpng-1.2.22.p5: port exists: http://gnuwin32.sourceforge.net/packages/libpng.htm. Since that port uses MinGW we need to either use MinGW 64 bit or do an MSVC port ourselves. In this case it seems likely that somebody else has done a proper Windows port. We might also be able to uses the system's library, but there might be issues with featurs, i.e. IE's libpng is supposed to be fairly bad. yep, something builtin to windows might be useful.

• python-2.5.1.p13: excellent support for MSVC, 32 & 64 bit mode. yep, very very good support on windows

• freetype-2.3.5: unclear if it is even required on Windows.

• gd-2.0.33.p5: port exists: http://gnuwin32.sourceforge.net/packages/gd.htm. Since that port uses MinGW we need to either use MinGW 64 bit or do an MSVC port ourselves.

• gdmodule-0.56.p4: should be pretty easy to do for MSVC Boothby told me this swig-generated "crap" will be removed from sage soon, he hopes.

• fortran-20071120.p3: Use Intel's Fortran compiler. An alternative for 32 bit is g77.
• lapack-20071123.p0: Works out of the box with GNU make.

• atlas-3.8.p11: Works find fine, Intel C for better performance preferred

• gsl-1.10.p0: port exists: http://gnuwin32.sourceforge.net/packages/gsl.htm. Brian Gladman also maintains a native MSVC port, 32 & 64 bit.

• iml-1.0.1.p9: no port exists, but it should be pretty straight forward. this won't be hard, since iml is just clean C code, mostly concerned with mat

• ipython-0.8.1.p1: should work fine with MSVC *definitely* will work fine.

• givaro-3.2.6.p6: no port exists, C++ code uses templates and it migt. be a little tricky to get right. Upstream welcomes patches could be hard.

• f2c-20070816.p0: part of scikits, should work with MSVC
• blas-20070724: works fine with GNU make and Intel Fortran compiler

• matplotlib-0.91.1.p3: should work with MSVC, more investigation is needed I'm sure this will work -- it is also part of what the enthought people distribute for windows.

• mercurial-0.9.5.p0 - some C extensions, maybe some portability issues, might need some TLC for various config issues
• mpfr-2.3.1: pure C, Windows make file support welcome upstream, depends on gmp.

• pexpect-2.0.p1: This is the big, difficult one. pexpect uses pseudo ttys and a whole bunch of POSIX infrastructure. It is essential and central to Sage and no one has ported pexpect to Windows. While the API is very different it is still possible to map all functionality to native Windows API calls. But it will probably take a lot of energy to get it stable and clean. The ultimate goal would be to get this merged upstream since it would be quite a lot of work to go at this alone yes, this is the biggie. note that pexpect *does* sort of work under cygwin, which might be relevant. Also, instead of porting pexpect, it might be possible to simply do something completely different than accomplishes much the same thing. pexpect was the way to go with Python and Unix -- in windows perhaps there is a totally different approach to interprocess communication that would work better? It's possibly OK if this means rewriting SAGE_ROOT/devel/sage/sage/interfaces/* for this new approach. This definitely needs to be investigated. Some of the programs Sage interfaces with have native "client server" archicture, and that could perhaps be used instead. E.g., Maple and Mathematica do. Maxima almost certainly doesn't...

• mpfi-1.3.4-cvs20071125.p5: depends on mpfr and gmp, pure C library. Port is simple and will probably be merged upstream

• pycrypto-2.0.1.p1: unclear how hard the port will be, but shouldn't be too hard. maybe trivial??

• cython-0.9.6.12: pure python, no portability issues, might needs some TLC for various config issues my impression from cython-devel is that there are already numerous windows users of cython

• zodb3-3.7.0.p0: mostly python, unknown difficulty to port probably easy

• quaddouble-2.2.p7: MSVC project exists, need to investigate 64 bit build
• python_gnutls-1.1.4.p2: Should work, uses dlopen based mechanism. Is essential to the notebook, but can be dropped in favour of system's OpenSSL compatible library
• twistedweb2-20070619.p0 - python, is supported, relevant to notebook
• twisted-2.5.0.p9 - python, is supported, relevant to notebook

• singular-3-0-4-1-20071209.p4: Cygwin only, but needs even work there to run properly (libSingular) Needs to be ported to MSVC, upstream very willing to help out, but lacks expertise. Various issues like memory managment (omalloc) need to be sovled (malloc fallback), but build system issues and stuff like signals make this a bigger job. I would assume this is one of the two heavy jobs together with pexpect. This is probably quite hard. And interesting. At least singular builds under cygwin, which is a good sign, I guess.

• symmetrica-2.0.p1: no Windows port yet, but macro heavy. worrisome.

• libfplll-2.1.6-20071129.p1: no Windows port, needs some specialized math functions, but should be a relatively easy port. Upstream will likely merge Windows patches worry?

• polybori-0.1-r7: no Windows port, could be tricky, but upstream is willing to cooperate. worry??

• rpy-1.0.1.p1: Support for MSVC in 32 bit mode with MSVC. No Cygwin support, but patch by Abshoff exists

• r-2.6.1.p14: Existing MinGW port, not too difficult to build. Port to MSVC desired. They don't already have a native MSVC port? Wow. That's amazing.

• rubiks-20070912.p1: no Windows port, but simple C/C++ code without any POSIX dependencies. Console applications easy?

• libm4ri-20071224.p1: no Windows port, maintainer will accept Windows port [windows port done by Martin Albrecht - w00t]

• ecm-6.1.3: no Windows port, but C, potentially issues with threading worrisome.

• examples-2.10.2.alpha0: no compiled code, i.e. no problem

• clisp-2.41.p12: MinGW port exists. Maintainer is receptive for build issues and usually fixes them quickly. maxima is often used under windows, so we should figure out their prefered lisp. We only use clisp for maxima.

• maxima-5.13.0.p2: Should build fine with Clisp, Windows binary package exists.

• genus2reduction-0.3.p1: unclear, don't really see any problem easy, probably; it's just math

• lcalc-20070107.p1: somewhat crafty C code, ugly hacks in the user interface, but should be portable to MSVC. Builds fine under Cygwin
• sympow-1.018.1.p4: tricky code that needs the right FPU flags set. Otherwise no big hurdles to overcome
• cddlib-094b.p1: unclear how hard the port will be, but shouldn't be too hard.

• gfan-0.2.2.p2: unclear how hard the port will be, but shouldn't be too hard. we should upgrade to 0.2.3 first. Some more remarks: gfan requires argv[0] to switch to the right computation. I.e. during installation the following is executed: "ln -s gfan gfan_weightvector". This should be portable, but symlinks can be problematic on Windows.

• flintqs-20070817.p2: is obsolete, but should build fine once FLINT works, C99 mode required
• palp-1.1.p1: unclear how hard the port will be, but shouldn't be too hard.
• moin-1.5.7.p2: python based, should work fine on Windows
• ipython1-20070130: Windows port exists
• cvxopt-0.9.p5: Has Windows build support, provides binary packages
• jmol-11.5.2.p1: pure java, no portability issues, large community of users on Windows
• linbox-20070915.p5: this will be some heavy lifting, mostly due to some heavy templated code. Upstream is willing to merge and help out.

• gap-4.4.10.p2: Depends on sbrk(), currently builds only via Cygwin. a cygwin build could be fine for us, since there is no binary linking between sage and gap, and almost certainly there never will be any.

Packages with upstream MSVC support

• numpy-20080104-1.0.4.p2: support by enthought, builds out of the box with MSVC + ATLAS
• scipy-20071020-0.6.p3: support by enthought, builds out of the box with MSVC + ATLAS
• scipy_sandbox-20071020.p2: support by enthought, builds out of the box with MSVC + ATLAS
• scons-0.97: works fine on Windows, python based.
• tachyon-0.98beta.p4: Windows port exists, unsure about MSVC + threading.
• weave-0.4.9: now part of scikits, works fine with MSVC

Pure Python Packages

Packages without code

• doc-2.10.2.alpha0: no compiled code, so no portability issues.
• sympy-0.5.7: pure python, no portability issues.
• networkx-0.36.p1: pure python, no portability issues