[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On 28 Okt., 03:52, kcrisman kcris...@gmail.com wrote: The Sage packages that directly depend on Fortran are: R, f2c, lapack, blas, numpy, scipy, cvxopt: Note that numpy can be built and used *without* fortran, e.g., when I removed fortran from psage I kept numpy... So a non-Fortran Sage would lose a lot of numerical ability (mpmath would stay, but it looks like GSL depends on blas/atlas/something with Fortran) GSL needs /some/ BLAS implementation, and (therefore) comes with CBLAS, a C implementation of BLAS (which might of course be inferior to other implementations). -Leif -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On Wed, Oct 27, 2010 at 6:01 PM, kcrisman kcris...@gmail.com wrote: Regarding fortran, for a Microsoft Visual C++ version of Sage, I will just get rid of Fortran (and Lisp) entirely, and not bother with building anything currently in Sage that depends on them... My thoughts about fortran oscillate as well. If I have the option, I try to just use Python + Cython. For current Sage needs, though, the reason for Fortran is Scipy and R, correct? (Maybe PolyBoRi?) The Sage packages that directly depend on Fortran are: R, f2c, lapack, blas, numpy, scipy, cvxopt: deep:sage-4.5.3 wstein$ grep FORTRAN spkg/standard/deps $(INST)/$(FORTRAN) \ $(INST)/$(R): $(BASE) $(INST)/$(PYTHON) $(INST)/$(ATLAS) $(INST)/$(ICONV) $(INST)/$(FORTRAN) $(INST)/$(FORTRAN): $(BASE) $(INST)/$(PYTHON) $(INSTALL) $(SAGE_SPKG) $(FORTRAN) 21 tee -a $(SAGE_LOGS)/$(FORTRAN).log $(INST)/$(F2C): $(BASE) $(INST)/$(FORTRAN) $(INST)/$(LAPACK): $(BASE) $(INST)/$(FORTRAN) $(INST)/$(BLAS): $(BASE) $(INST)/$(FORTRAN) $(INST)/$(NUMPY): $(BASE) $(INST)/$(PYTHON) $(INST)/$(FORTRAN) $(INST)/$(F2C) \ $(INST)/$(SCIPY): $(BASE) $(INST)/$(FORTRAN) $(INST)/$(F2C) \ $(INST)/$(CVXOPT): $(BASE) $(INST)/$(FORTRAN) $(INST)/$(F2C) Note that numpy can be built and used *without* fortran, e.g., when I removed fortran from psage I kept numpy... I can be more precise than that. numpy needs fortran if you include lapack support, lapack is a pure fortran library. Support for lapack in numpy is optional. So, if I am not mistaken, it just means you have lapack-less numpy. Francois -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On 10/26/10 08:32 PM, William Stein wrote: On Tue, Oct 26, 2010 at 11:15 AM, leifnot.rea...@online.de wrote: On 26 Okt., 19:36, William Steinwst...@gmail.com wrote: Regarding fortran, for a Microsoft Visual C++ version of Sage, I will just get rid of Fortran (and Lisp) entirely, and not bother with building anything currently in Sage that depends on them... Feed f2c with all Fortran sources and ship the results... ;-) I wish, but that does not work. Believe me, we've tried. Have you tried the latest version of f2c? The version we have seems to be rather old. version.c has in it: char F2C_version[] = 19991025; but the latest from the netlib site has: char F2C_version[] = 20061008; so it looks like our version is 7 years older than the latest. Dave -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
Hi Francois, Well thanks for the plug for our work. We are quite happy to develop the prefix :-) part of sage-on-gentoo, so far the effort has been limited but if you are testing it on an arch we are quite happy to try to get as much as possible keyworded. Is the keywording the most prominent issue, or getting dozens of ebuilds $EPREFIX-ified? (Or is this the same? My understanding of Gentoo internals and the Gentoo ecosystem is still rather limited.) By the way did you try sage-4.5.3 or 4.6 (alphaX/rc0)? I don't think it really mattered, but I targeted sage-4.5.3 (while some 4.6 alpha had been the youngest ebuild then). Indeed gentoo do not offer you the possibility to have relocatable binaries as such. It's not the point in a way. However gentoo can create binary packages (on an individual package basis) so it would be possible to install binaries from a base gentoo prefix install. As far as I understand, (re-)installing such a binary package would only be possible in exactly the same $EPREFIX path, where the binary package was originally installed in (although possibly on another computer, of course). That is too tight a restriction in many use cases (e.g. if you're not the administrator and want/need to install Sage in your home directory). Francois How to continue? E.g. what would you propose, as how I should make known the two gcc-apple patches I mentioned (nobody else seems to use Gentoo Prefix on Mac OS X 10.4 Tiger, so who possibly would/could review say a patch on Gentoo Bugzilla)? Another point is that both Gentoo Prefix, as well as the sage-on- gentoo overlay, use patched versions of Python. Maybe it's possible to join these? Cheers, Georg -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
Hi Georg Hi Francois, Well thanks for the plug for our work. We are quite happy to develop the prefix : :-) : part of sage-on-gentoo, so far the effort has been limited but if you are testing it on an arch we are quite happy to try to get as much as possible keyworded. Is the keywording the most prominent issue, or getting dozens of ebuilds $EPREFIX-ified? (Or is this the same? My understanding of Gentoo internals and the Gentoo ecosystem is still rather limited.) The two things goes together it has to be $EPREFIX-ified first. Once it is done it is of course another matter to a particular arch keyworded. Keywording makes things easier but is not an obstacle for a seasoned user (I use plenty of ebuilds that are not keyworded ppc to install sage-on-gentoo on a ppc machine. I could probably spend an hour or two filling keyword requests for ppc). By the way did you try sage-4.5.3 or 4.6 (alphaX/rc0)? I don't think it really mattered, but I targeted sage-4.5.3 (while some 4.6 alpha had been the youngest ebuild then). Indeed gentoo do not offer you the possibility to have relocatable binaries as such. It's not the point in a way. However gentoo can create binary packages (on an individual package basis) so it would be possible to install binaries from a base gentoo prefix install. As far as I understand, (re-)installing such a binary package would only be possible in exactly the same $EPREFIX path, where the binary package was originally installed in (although possibly on another computer, of course). That is too tight a restriction in many use cases (e.g. if you're not the administrator and want/need to install Sage in your home directory). Yes it would be better if it was more flexible. I think it is technically doable but it is not easy. Francois How to continue? E.g. what would you propose, as how I should make known the two gcc-apple patches I mentioned (nobody else seems to use Gentoo Prefix on Mac OS X 10.4 Tiger, so who possibly would/could review say a patch on Gentoo Bugzilla)? Not sure. It may very well be that no one wants/has time to support 10.4. The only thing to do is really to fill a bug, it will be assigned to the prefix people. I guess we could pick the patch ourselve in sage-on-prefix (see below). Another point is that both Gentoo Prefix, as well as the sage-on- gentoo overlay, use patched versions of Python. Maybe it's possible to join these? You should be aware that Christopher made a second overlay called sage-on-prefix to address issues which are specifically related to prefix: http://github.com/cschwan/sage-on-prefix Right now the focus has been x86-linux and amd64-linux because it is easy to self host them on a regular Gentoo install. Christopher also has a windows machine and is looking at using sage-on-gentoo for windows as well. We should definitely host a merged python ebuild on the sage-on-prefix overlay. It is not however a big deal at the moment. The only reason we have our own python in the overlay is pickling: http://bugs.gentoo.org/show_bug.cgi?id=301691 http://bugs.python.org/issue7689 Hopefully there will be some move to merge this upstream. Cheers, Francois -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On Tue, Oct 26, 2010 at 10:36 AM, William Stein wst...@gmail.com wrote: On Tue, Oct 26, 2010 at 10:11 AM, Volker Braun vbraun.n...@gmail.com wrote: downloading Xcode or installing it is a little annoying, but fairly straightforward even for newbies, because Apple wants to make it easy for them. But fortran is another matter. I just tried http://gcc.gnu.org/wiki/GFortranBinariesMacOS on OSX 10.6 and all I had to do was click on the link and then click through the standard installer. Its easier than installing Xcode... Please also try on PPC OS X 10.4, 10.5, and 10.6, and also Intel OS X 10.4, 10.5, 10.6, both 32 and 64-bit, then report back. That's what the Sage fortran spkg handles. (1) Install development tools that are completely standard to install on your OS. These days, on Linux, this includes GCC = Gnu Compiler Collection, which includes Fortran, no problem. On OS X, installing Fortran is far from standard, since Apple doesn't care about it, and there are several variants out there. How about Windows? There you'll always have to download cygwin and a bunch of cygwin packages to get off the ground. I oscillate in my hopes about cygwin... There are other approaches to Sage on Windows, e.g.,: http://windows.sagemath.org/ Regarding fortran, for a Microsoft Visual C++ version of Sage, I will just get rid of Fortran (and Lisp) entirely, and not bother with building anything currently in Sage that depends on them... My thoughts about fortran oscillate as well. If I have the option, I try to just use Python + Cython. But if I had to choose between C/C++ and fortran for *numeric* projects, I would probably choose fortran these days, + fwrap or f2py + Python. Especially after reading benchmarks like these: http://www.oonumerics.org/blitz/benchmarks/acou3d.html where one needs to use templates and lots of expertize in C++ to even beat fortran code written by pretty much anybody... Ondrej -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
Regarding fortran, for a Microsoft Visual C++ version of Sage, I will just get rid of Fortran (and Lisp) entirely, and not bother with building anything currently in Sage that depends on them... My thoughts about fortran oscillate as well. If I have the option, I try to just use Python + Cython. For current Sage needs, though, the reason for Fortran is Scipy and R, correct? (Maybe PolyBoRi?) Though I also note http://www.sagemath.org/doc/numerical_sage/f2py.html refers to another cool %magic thing. Just checking. - kcrisman -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
Sorry OT On Oct 28, 12:50 am, Ondrej Certik ond...@certik.cz wrote: Especially after reading benchmarks like these: http://www.oonumerics.org/blitz/benchmarks/acou3d.html where one needs to use templates and lots of expertize in C++ to even beat fortran code written by pretty much anybody... Well the benchmark I see requires the Fortran programmer to implement a tiled traversal of a 3-d array and by hand unrolling 3 iterations to avoid copying arrays. I don't think that pretty much anybody has those tricks up his sleeve. Blitz++ transparently does the Hilbert- curve array traversal and lets you swap array handles instead of array contents, so you don't have to manually unroll the iteration. I wouldn't want to write my own expression template library from scratch, but _using_ blitz++ is much easier than writing the analogous code in Fortran. Volker -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On Wed, Oct 27, 2010 at 6:01 PM, kcrisman kcris...@gmail.com wrote: Regarding fortran, for a Microsoft Visual C++ version of Sage, I will just get rid of Fortran (and Lisp) entirely, and not bother with building anything currently in Sage that depends on them... My thoughts about fortran oscillate as well. If I have the option, I try to just use Python + Cython. For current Sage needs, though, the reason for Fortran is Scipy and R, correct? (Maybe PolyBoRi?) The Sage packages that directly depend on Fortran are: R, f2c, lapack, blas, numpy, scipy, cvxopt: deep:sage-4.5.3 wstein$ grep FORTRAN spkg/standard/deps $(INST)/$(FORTRAN) \ $(INST)/$(R): $(BASE) $(INST)/$(PYTHON) $(INST)/$(ATLAS) $(INST)/$(ICONV) $(INST)/$(FORTRAN) $(INST)/$(FORTRAN): $(BASE) $(INST)/$(PYTHON) $(INSTALL) $(SAGE_SPKG) $(FORTRAN) 21 tee -a $(SAGE_LOGS)/$(FORTRAN).log $(INST)/$(F2C): $(BASE) $(INST)/$(FORTRAN) $(INST)/$(LAPACK): $(BASE) $(INST)/$(FORTRAN) $(INST)/$(BLAS): $(BASE) $(INST)/$(FORTRAN) $(INST)/$(NUMPY): $(BASE) $(INST)/$(PYTHON) $(INST)/$(FORTRAN) $(INST)/$(F2C) \ $(INST)/$(SCIPY): $(BASE) $(INST)/$(FORTRAN) $(INST)/$(F2C) \ $(INST)/$(CVXOPT): $(BASE) $(INST)/$(FORTRAN) $(INST)/$(F2C) Note that numpy can be built and used *without* fortran, e.g., when I removed fortran from psage I kept numpy... William Though I also note http://www.sagemath.org/doc/numerical_sage/f2py.html refers to another cool %magic thing. Just checking. - kcrisman -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org -- William Stein Professor of Mathematics University of Washington http://wstein.org -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
The Sage packages that directly depend on Fortran are: R, f2c, lapack, blas, numpy, scipy, cvxopt: Note that numpy can be built and used *without* fortran, e.g., when I removed fortran from psage I kept numpy... So a non-Fortran Sage would lose a lot of numerical ability (mpmath would stay, but it looks like GSL depends on blas/atlas/something with Fortran) and a non-Lisp Sage would lose Maxima (for us, primarily symbolic ODE/sum/integrals, since we don't use sympy for these + second option for numerical integration and maybe some other stuff). Good to know, thanks. - kcrisman -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On Wed, Oct 27, 2010 at 7:52 PM, kcrisman kcris...@gmail.com wrote: The Sage packages that directly depend on Fortran are: R, f2c, lapack, blas, numpy, scipy, cvxopt: Note that numpy can be built and used *without* fortran, e.g., when I removed fortran from psage I kept numpy... So a non-Fortran Sage would lose a lot of numerical ability (mpmath would stay, but it looks like GSL depends on blas/atlas/something with Fortran) numpy and GSL *BOTH* stay. The entire point of GSL is that it is numerical code written in C instead of Fortran. In fact, GSL is in psage. You might even try downloading and building the latest version of psage: http://purple.sagemath.org/source/?C=N;O=D and a non-Lisp Sage would lose Maxima (for us, primarily Yes. symbolic ODE/sum/integrals, since we don't use sympy for these + second option for numerical integration and maybe some other stuff). Yes. Good to know, thanks. - kcrisman -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org -- William Stein Professor of Mathematics University of Washington http://wstein.org -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On Oct 27, 11:16 pm, William Stein wst...@gmail.com wrote: On Wed, Oct 27, 2010 at 7:52 PM, kcrisman kcris...@gmail.com wrote: The Sage packages that directly depend on Fortran are: R, f2c, lapack, blas, numpy, scipy, cvxopt: Note that numpy can be built and used *without* fortran, e.g., when I removed fortran from psage I kept numpy... So a non-Fortran Sage would lose a lot of numerical ability (mpmath would stay, but it looks like GSL depends on blas/atlas/something with Fortran) numpy and GSL *BOTH* stay. The entire point of GSL is that it is numerical code written in C instead of Fortran. Okay - the website seemed to indicate dependence on various things that have fortran, but it did say weak dependence so maybe the core functionality doesn't need that? GSL requires a BLAS library for vector and matrix operations but maybe that doesn't have to be fortran, or maybe that's irrelevant to the main numerical stuff one would care about. Thanks for clarifying that, anyway. In fact, GSL is in psage. You might even try downloading and building the latest version of psage: http://purple.sagemath.org/source/?C=N;O=D Not a bad idea, of course - I'll try this on my Tiger PPC Mac tomorrow to make sure it works fine :) though it certainly should. - kcrisman -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On Wed, Oct 27, 2010 at 8:56 PM, kcrisman kcris...@gmail.com wrote: On Oct 27, 11:16 pm, William Stein wst...@gmail.com wrote: On Wed, Oct 27, 2010 at 7:52 PM, kcrisman kcris...@gmail.com wrote: The Sage packages that directly depend on Fortran are: R, f2c, lapack, blas, numpy, scipy, cvxopt: Note that numpy can be built and used *without* fortran, e.g., when I removed fortran from psage I kept numpy... So a non-Fortran Sage would lose a lot of numerical ability (mpmath would stay, but it looks like GSL depends on blas/atlas/something with Fortran) numpy and GSL *BOTH* stay. The entire point of GSL is that it is numerical code written in C instead of Fortran. Okay - the website seemed to indicate dependence on various things that have fortran, but it did say weak dependence so maybe the core functionality doesn't need that? GSL requires a BLAS library for vector and matrix operations but maybe that doesn't have to be fortran, or maybe that's irrelevant to the main numerical stuff one would care about. Thanks for clarifying that, anyway. In fact, GSL is in psage. You might even try downloading and building the latest version of psage: http://purple.sagemath.org/source/?C=N;O=D Not a bad idea, of course - I'll try this on my Tiger PPC Mac tomorrow to make sure it works fine :) though it certainly should. It probably will, since it's basically just sage. That said, I list the sorted platforms at http://purple.sagemath.org/ as Supported platforms: 64-bit Ubuntu Linux and OS X 10.6 -- William - kcrisman -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org -- William Stein Professor of Mathematics University of Washington http://wstein.org -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
I think the Sage policy should be to never ship part of the toolchain. In the particular case of gfortran OSX, it seems to be easy enough to install the .dmg file. Ideally, Sage should have 1) easy-to-follow documentation for setting up the toolchain. Preferably from some standard repository for the given OS. Install gcc doesn't quite cut it :-) For example, on Solaris the instructions could be: Set up Blastwave, install the following packages (./pgkadd package1 package2). I still haven't been able to successfully build Sage on my OpenSolaris installation... 2) a configure-like test script that tests the requirements and gives useful hints on how to install missing pieces of the toolchain and/or required libraries. Volker On Oct 26, 8:53 am, Dr. David Kirkby david.kir...@onetel.net wrote: The fortran package in Sage is 33 MB in size. It contains a binary fortran compiler for OS X only. So users of Linux, Solaris and those porting to other platforms are downloading 33 MB of totally useless code. Given the size and the fact it's only needed on one platform, would it not be more sensible if this was downloaded only if needed? Every time I download a new alpha to test, I'm downloading 33 MB of unnecessary code. I expect many people that run OS X also download it unnecessarily, since many will have fortran compilers anyway. A list of binaries for gfortran can be found at http://gcc.gnu.org/wiki/GFortranBinaries#MacOS I can't think of any other open-source application I've come across where I download a binary of the compiler every time I download the code. There are a few packages that are not needed on all platforms (cephes, iconv are two I can think of), but neither of those are particularly large, so are less of an issue. Dave -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On 26 Okt., 12:39, Volker Braun vbraun.n...@gmail.com wrote: 2) a configure-like test script that tests the requirements and gives useful hints on how to install missing pieces of the toolchain and/or required libraries. Yep. Especially odd is that Sage ships four MacOS X *binaries* in a so- called *source* distribution. (The Linux binaries were removed a while ago.) One could consider the bunch of fonts shipped also binaries, i.e. these should IMHO be moved to separate packages, since they don't change that frequently, at least not as often as the packages that currently contain them. (Some of these and even obsolete ones are in the package's Mercurial repository, too, btw.) I'd also prefer a concept of platform-specific spkgs (perhaps also tarballs specific to e.g. Darwin and Cygwin), cf. http://groups.google.com/group/sage-release/msg/6564b50668274ae7 -Leif -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
I agree that it is a bad policy to ship binaries in a source distribution. I would suggest a separate package for MAC OS with the binaries of the compiler if this is reallly needed On Tue, Oct 26, 2010 at 9:30 AM, leif not.rea...@online.de wrote: On 26 Okt., 12:39, Volker Braun vbraun.n...@gmail.com wrote: 2) a configure-like test script that tests the requirements and gives useful hints on how to install missing pieces of the toolchain and/or required libraries. Yep. Especially odd is that Sage ships four MacOS X *binaries* in a so- called *source* distribution. (The Linux binaries were removed a while ago.) One could consider the bunch of fonts shipped also binaries, i.e. these should IMHO be moved to separate packages, since they don't change that frequently, at least not as often as the packages that currently contain them. (Some of these and even obsolete ones are in the package's Mercurial repository, too, btw.) I'd also prefer a concept of platform-specific spkgs (perhaps also tarballs specific to e.g. Darwin and Cygwin), cf. http://groups.google.com/group/sage-release/msg/6564b50668274ae7 -Leif -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
I don't like having different source distributions for different OS'es. Thats unnecessary added complexity. If you want to save a few megabytes you can always rsync the sage.tar archive, this will avoid retransmitting the unchanged spkgs. Volker On Oct 26, 1:30 pm, leif not.rea...@online.de wrote: On 26 Okt., 12:39, Volker Braun vbraun.n...@gmail.com wrote: 2) a configure-like test script that tests the requirements and gives useful hints on how to install missing pieces of the toolchain and/or required libraries. Yep. Especially odd is that Sage ships four MacOS X *binaries* in a so- called *source* distribution. (The Linux binaries were removed a while ago.) One could consider the bunch of fonts shipped also binaries, i.e. these should IMHO be moved to separate packages, since they don't change that frequently, at least not as often as the packages that currently contain them. (Some of these and even obsolete ones are in the package's Mercurial repository, too, btw.) I'd also prefer a concept of platform-specific spkgs (perhaps also tarballs specific to e.g. Darwin and Cygwin), cf.http://groups.google.com/group/sage-release/msg/6564b50668274ae7 -Leif -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
No offense, but everyone who has written so far in this thread is speaking only to people who know what the word toolchain means in this context. Unfortunately, Apple doesn't provide a fortran compiler, and setting one up for those who don't know that word is nontrivial and goes against the (in my opinion, more important) Sage philosophy of batteries included. It's particularly annoying that on older Macs gfortran isn't even available so we need g95, but that is also how it is, and we definitely still get people asking about this platform as well. Basically, someone who would like to have a brand spankin' new Sage they can call their own (as opposed to a binary download) should not have needless hurdles placed in front of them. Should we provide gcc? No - downloading Xcode or installing it is a little annoying, but fairly straightforward even for newbies, because Apple wants to make it easy for them. But fortran is another matter. Pablo, are you suggesting that sage -sdist would make two different types of source distributions, one for Mac, one for non-Mac? That could be confusing, but I guess as long as they were (automatically, in the sdist process) named VERY clearly, and it was VERY obvious on the download site which was which (and why there is a difference), that could be a solution (though it would add some responsibilities to someone, either web person or release manager). But certainly 33 MB is nothing to sniff at, particularly in situations with low bandwidth. What does that (piece of Sage) compress to in the .tgz? - kcrisman -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On 2010-10-26 15:15, kcrisman wrote: Basically, someone who would like to have a brand spankin' new Sage they can call their own (as opposed to a binary download) should not have needless hurdles placed in front of them. Should we provide gcc? No - downloading Xcode or installing it is a little annoying, but fairly straightforward even for newbies, because Apple wants to make it easy for them. But fortran is another matter. Could we instead ship the *source code* of the fortran compiler? That would almost certainly be smaller. Jeroen. -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
Could we instead ship the *source code* of the fortran compiler? That would almost certainly be smaller. No idea how easy that would be. I'm sure it could be done, but actually implementing it might get a little tricky. G95: Instructions at http://www.g95.org/source.shtml seem straightforward for people who know how to configure properly (sadly, I am not at that point), and seems like it could be automated spkg- style. This would only need to happen for OS X 10.4, so checking in the spkg for uname with Darwin 8.x would be sufficient; we do things like that elsewhere, presumably including the current fortran spkg. Unfortunately, although it is possible to do Gfortran on OS X 10.4, it's not so easy because you need a newer gcc which isn't included, and if we are including batteries we probably don't want to include a different gcc; Xcode makes things pretty easy. Gfortran: The page http://gcc.gnu.org/wiki/GFortranSource has more spartan instructions. It also isn't easy to disentangle from gcc itself (I mean, there is stuff in there, but I don't know whether it would play nicely with Apple's gcc). We might in that case need to do it version-by-version - we have a few different gfortran binaries, so we'd likely have to include all of the appropriate sources. Still, each of these are 2 MB in bz2 format, so I would estimate we could get in under 10 MB total with just source if we could figure out how to do it right, maybe more like 3-4 MB if we only needed one version of gfortran. That would be a nice compromise to the issues raised here, and I'd be very happy to test these (as I'm sure some others would be as well); unfortunately, I don't have the skills needed to actually create such an spkg. - kcrisman -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On 26 Okt., 15:15, Volker Braun vbraun.n...@gmail.com wrote: I don't like having different source distributions for different OS'es. Me either, but if we keep shipping binaries in a source distributions that's an easy way to go: Create a fortran-x.y.z-darwin.spkg that's only included in sage-u.v.w- darwin.tar, trivial to implement in sage -sdist. Even better would be to (also) have it in e.g. $SAGE_ROOT/spkg/ darwin/, which would be easier to handle in the sage-upgrade script. Thats unnecessary added complexity. If you want to save a few megabytes you can always rsync the sage.tar archive, this will avoid retransmitting the unchanged spkgs. rsync is not an option for ordinary users, nor when upgrading Sage (which already lowers the amount of necessary downloads, but not if somebody decides to change a few characters in e.g. the Fortran spkg's spkg-install; also [unnecessarily] many packages will get rebuilt due to such.) Replying to Karl-Dieter: I wouldn't mind keeping an spkg with binaries for Darwin (and perhaps other binary spkgs for other platforms), but people using different platforms shouldn't be bothered by that. The question would be if the users are to download it manually, download a different tarball, or just get instructions on how to install after 'configure' realized some part of the toolchain is missing. As said on sage-release, we currently don't want automatic downloads during the (real) build process; we could of course offer that as an option, though other people again don't like interactive scripts... A simple get_prerequisites script would be an alternative, also adding some option to enable quering to or automatically getting them (i.e., platform-specific spkgs as required). A lot of what's currently made in individual spkg-installs should IMHO be moved to Sage's 'configure', e.g. detecting a capable libreadline (and its header files) and the decision to eventually install Sage's readline spkg (cf. #9530, #9523). Also, I guess there are people using other platforms that would say installing prerequisite xy on my system/platform isn't easy either. ;-) -Leif -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
Leif - Any thoughts on the source suggestion? If there really would be little confusion with a separate Darwin source distribution that included fortran (where it could be verified this was the only difference), that might also be a way to go, as you say. Also, I guess there are people using other platforms that would say installing prerequisite xy on my system/platform isn't easy either. ;-) Perhaps, but in that case they shouldn't be using such user-unfriendly systems in the first place ;) - kcrisman -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
No, my point is that gfortran is not distributed with sage in source form. I think that we should either '- include gfortran in source form in sage distribution - split the binary files in a separated file otherwise it is not a source distribution. )but a mix of a source and a binary one. This could even violate the GPL (under which gfortran is distributed) siince we are distributing the binaries but not the sources. I see that spliting it would mean more complexity, so perhaps we could put gfortran sources and compile them only on Mac OS None of these solutions seems to be nice, though. Pablo Pablo, are you suggesting that sage -sdist would make two different types of source distributions, one for Mac, one for non-Mac? That could be confusing, but I guess as long as they were (automatically, in the sdist process) named VERY clearly, and it was VERY obvious on the download site which was which (and why there is a difference), that could be a solution (though it would add some responsibilities to someone, either web person or release manager). But certainly 33 MB is nothing to sniff at, particularly in situations with low bandwidth. What does that (piece of Sage) compress to in the .tgz? - kcrisman -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On Tue, Oct 26, 2010 at 6:15 AM, kcrisman kcris...@gmail.com wrote: No offense, but everyone who has written so far in this thread is speaking only to people who know what the word toolchain means in this context. Unfortunately, Apple doesn't provide a fortran compiler, and setting one up for those who don't know that word is nontrivial and goes against the (in my opinion, more important) Sage philosophy of batteries included. It's particularly annoying that on older Macs gfortran isn't even available so we need g95, but that is also how it is, and we definitely still get people asking about this platform as well. +1 Basically, someone who would like to have a brand spankin' new Sage they can call their own (as opposed to a binary download) should not have needless hurdles placed in front of them. Should we provide gcc? No - downloading Xcode or installing it is a little annoying, but fairly straightforward even for newbies, because Apple wants to make it easy for them. But fortran is another matter. Indeed. Pablo, are you suggesting that sage -sdist would make two different types of source distributions, one for Mac, one for non-Mac? That could be confusing, but I guess as long as they were (automatically, in the sdist process) named VERY clearly, and it was VERY obvious on the download site which was which (and why there is a difference), that could be a solution (though it would add some responsibilities to someone, either web person or release manager). But certainly 33 MB is nothing to sniff at, particularly in situations with low bandwidth. What does that (piece of Sage) compress to in the .tgz? It this it is already compressed, so it does not compress any further. Since day 1 Sage has had a certain list of requirements to build which -- as Karl remarks above -- was developed for people who don't know what toolchain means. (1) Install development tools that are completely standard to install on your OS. These days, on Linux, this includes GCC = Gnu Compiler Collection, which includes Fortran, no problem. On OS X, installing Fortran is far from standard, since Apple doesn't care about it, and there are several variants out there. (2) Type make. Removing the Fortran-for-OSX binary from Sage totally breaks this, and I'm completely against breaking (1) and (2) above. Your motivation is: So users of Linux, Solaris and those porting to other platforms are downloading 33 MB of totally useless code. (1) You could remove fortran-*.spkg from your tarball before downloading it; this could be automated with your own script. (2) sage -upgrade could be changed to never grab a new fortran spkg if the operating system isn't OS X. -- William P.S. I care more about (1)-(2) above than about fortran per se. With psage, I removed the fortran spkg entirely, and removed everything that depended on it. I did the same with lisp. This keeps (1)-(2), while moving toward a smaller and more maintainable kernel that I can then build other things on top of. -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On Tue, Oct 26, 2010 at 8:18 AM, Pablo De Napoli pden...@gmail.com wrote: No, my point is that gfortran is not distributed with sage in source form. I think that we should either '- include gfortran in source form in sage distribution Including gfortran or g95 in source form is not realistic, as you'll discover if you spend a few weeks trying... - split the binary files in a separated file otherwise it is not a source distribution. )but a mix of a source and a binary one. This could even violate the GPL (under which gfortran is distributed) siince we are distributing the binaries but not the sources. Seriously? I see that spliting it would mean more complexity, so perhaps we could put gfortran sources and compile them only on Mac OS None of these solutions seems to be nice, though. Building from source would be nice. Unfortunately, it is not a solution, because it is far too difficult. Confusing users needlessly to save 30MB is not an option. And people who seriously care about whether or not sage-x.y.z.tar is a source distribution in some pure sense can (and do!) just delete fortran.spkg. Most users just care that whatever sage-x.y.z.tar is, it just works. Pablo Pablo, are you suggesting that sage -sdist would make two different types of source distributions, one for Mac, one for non-Mac? That could be confusing, but I guess as long as they were (automatically, in the sdist process) named VERY clearly, and it was VERY obvious on the download site which was which (and why there is a difference), that could be a solution (though it would add some responsibilities to someone, either web person or release manager). But certainly 33 MB is nothing to sniff at, particularly in situations with low bandwidth. What does that (piece of Sage) compress to in the .tgz? - kcrisman -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org -- William Stein Professor of Mathematics University of Washington http://wstein.org -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On 10/26/10 04:24 PM, William Stein wrote:
Confusing users needlessly to save 30MB is not an option.
And people who seriously care about whether or not sage-x.y.z.tar is a
source distribution in some pure sense can (and do!) just delete
fortran.spkg. Most users just care that whatever sage-x.y.z.tar is,
it just works.
There is one other possible option, which should not confuse users, and would
save at least 11.3 MB.
1) Add the compression program 'p7zip' as a standard package.
http://www.7-zip.org/
The source for this is 3.6 MB, so lets assume with the SPKG.txt, repository,
spkg-install that becomes a 3.7 MB .spkg file.
(In fact, I suspect a lot of the source could be removed, as a lot of it is for
a GUI we would not want).
2) Use p7zip to compress the binaries in the fortran package, rather than bzip2
which is what is currently used. (If you look in the package, you will see
several tar .tar.bz2 files)
Using that compressor, the fortran package would then be 18 MB not 33 MB, as
p7zip is a much better compressor.
33-18-3.7=11.3 MB
So we could save 11.3 MB.
Of course, with p7zip we could compress all the sources with a better compressor
too, though the benefit of p7zip over bzip2 is much better with binaries.
As far as I can see, the spkg-install package has this line in it to decompress
file which will be some .tar.bz2 file.
if os.system('tar xvjf %s/%s'%(os.path.abspath(os.curdir), file)):
If that was changed, to a pipe with p7zip and tar, we could save about 11.3 MB.
Of course, with a better compressor in Sage, we could actually use it on every
.spkg, though the savings would not be as substantial on packages containing
just source. I'd guess we could probably save a few tens of MB though.
Dave
--
To post to this group, send an email to sage-devel@googlegroups.com
To unsubscribe from this group, send an email to
sage-devel+unsubscr...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/sage-devel
URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On Tue, Oct 26, 2010 at 12:24 PM, William Stein wst...@gmail.com wrote: On Tue, Oct 26, 2010 at 8:18 AM, Pablo De Napoli pden...@gmail.com wrote: No, my point is that gfortran is not distributed with sage in source form. I think that we should either '- include gfortran in source form in sage distribution Including gfortran or g95 in source form is not realistic, as you'll discover if you spend a few weeks trying... - split the binary files in a separated file otherwise it is not a source distribution. )but a mix of a source and a binary one. This could even violate the GPL (under which gfortran is distributed) siince we are distributing the binaries but not the sources. Seriously? well not . gfortran sources are available on the internet. But yes if you read the text of the GPL literally. (nobody will do this, I hope) -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
downloading Xcode or installing it is a little annoying, but fairly straightforward even for newbies, because Apple wants to make it easy for them. But fortran is another matter. I just tried http://gcc.gnu.org/wiki/GFortranBinariesMacOS on OSX 10.6 and all I had to do was click on the link and then click through the standard installer. Its easier than installing Xcode... (1) Install development tools that are completely standard to install on your OS. These days, on Linux, this includes GCC = Gnu Compiler Collection, which includes Fortran, no problem. On OS X, installing Fortran is far from standard, since Apple doesn't care about it, and there are several variants out there. How about Windows? There you'll always have to download cygwin and a bunch of cygwin packages to get off the ground. And Sage definitely must not ship the cygwin.dll as it needs to be unique on the system... Volker -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On Oct 26, 7:11 am, kcrisman kcris...@gmail.com wrote: Leif - Any thoughts on the source suggestion? If there really would be little confusion with a separate Darwin source distribution that included fortran (where it could be verified this was the only difference), that might also be a way to go, as you say. If you really want separate source distributions, it shouldn't be Darwin and everything else: it should be everything and optional smaller version for linux (or linux solaris, or systems which already include fortran, or whatever). That is, there should be one source distribution that works *everywhere* so that non-computer- savvy users can download it and just type make. Then for people who want to read the fine print, they can get the smaller version if their system meets the requirements. But to save 30 MB, it doesn't seem worth the possible confusion. If we want to do anything along these lines, I like Dave's idea of including a different compression tool. -- John -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On 26 Okt., 16:11, kcrisman kcris...@gmail.com wrote: Leif - Any thoughts on the source suggestion? If there really would be little confusion with a separate Darwin source distribution that included fortran (where it could be verified this was the only difference), that might also be a way to go, as you say. Rather than having two (or more!) source tarballs, which would (as is) require different upgrade paths as well, and probably annoy people mirroring Sage distros, what about having a separate sage-x.y.z-darwin- prerequisites.tar, to be extracted into the same location as the source tarball (i.e., having the same directory structure to simplify things)? This would require some little changes to the spkg/install and sage- upgrade scripts (and sage-sdist). As William says, I think shipping g95/gfortran sources to be built on Darwin is non-trivial and hardly worth doing (feel free to try it, to work on a wide range of Darwin installations...) If we do the above, i.e. provide an additional Darwin prerequisites tarball, we could even include an XCode spkg, as long as Apple's copyright lets us. (I haven't looked at that, but since it is based on / includes GNU software, it possibly does.) Similar might be convenient for Cygwin I guess, as AFAIR Sage requires Cygwin packages not present / installed by default as well. (In principle, we could move more packages into such a - not necessarily platform-dependent - category, to be only [downloaded and] installed if desired, e.g. when there's no suitable system-wide copy installed. Mercurial is just one example of such. But that's another roughly related debate.) Also, I guess there are people using other platforms that would say installing prerequisite xy on my system/platform isn't easy either. ;-) Perhaps, but in that case they shouldn't be using such user-unfriendly systems in the first place ;) I remember having to manually (re)build gfortran / GCC 4.2.1 to build Sage on Ubuntu 7.10 (which I guess isn't older than MacOS X 10.4), because the Linux binary had been removed from the spkg, there was no matching gfortran installed and (updated) packages for 7.10 weren't available. I'd appreciate if only people having user-friendly operating systems had to pay the price of (each time) downloading additional 33 MB. ;-) As with rsync, downloading only parts of a tarball isn't an option for ordinary users (even if they knew they don't need all of it). Also - for historical reasons I guess - currently the Fortran spkg sets up sage_fortran, which is sub-optimal but required *on all platforms*, s.t. one cannot simply omit or delete it. (As Dave suggested IIRC, we could move that elsewhere, e.g. 'configure' or sage-env, and/or simply use FC/F77, FFLAGS etc. where we call the fortran compiler.) ignore W.r.t. the batteries, I don't need them for my solar-powered systems... ;-) Buying a coffee portioner (I don't use anyway) with every pound is just environmental pollution and a waste of resources. /ignore As I understand this, Sage should be an (almost) self-contained distribution of *math* software, also including /some/ bits required for development (and - if necessary - /modified/ versions of basic things like Python), but isn't an operating system distribution or completely stand-alone SDK (software development toolkit). That of course doesn't prevent us offering *separate* (perhaps optional) easy-to-plug-in prerequisites (and add-on) packages to make installing and using Sage as easy as possible for everyone. And there are usually binary distributions for people not (yet) willing or able to build Sage from source themselves. -Leif -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On Tue, Oct 26, 2010 at 10:11 AM, Volker Braun vbraun.n...@gmail.com wrote: downloading Xcode or installing it is a little annoying, but fairly straightforward even for newbies, because Apple wants to make it easy for them. But fortran is another matter. I just tried http://gcc.gnu.org/wiki/GFortranBinariesMacOS on OSX 10.6 and all I had to do was click on the link and then click through the standard installer. Its easier than installing Xcode... Please also try on PPC OS X 10.4, 10.5, and 10.6, and also Intel OS X 10.4, 10.5, 10.6, both 32 and 64-bit, then report back. That's what the Sage fortran spkg handles. (1) Install development tools that are completely standard to install on your OS. These days, on Linux, this includes GCC = Gnu Compiler Collection, which includes Fortran, no problem. On OS X, installing Fortran is far from standard, since Apple doesn't care about it, and there are several variants out there. How about Windows? There you'll always have to download cygwin and a bunch of cygwin packages to get off the ground. I oscillate in my hopes about cygwin... There are other approaches to Sage on Windows, e.g.,: http://windows.sagemath.org/ Regarding fortran, for a Microsoft Visual C++ version of Sage, I will just get rid of Fortran (and Lisp) entirely, and not bother with building anything currently in Sage that depends on them... And Sage definitely must not ship the cygwin.dll as it needs to be unique on the system... Fortunately this is not true anymore. See http://cygwin.com/ml/cygwin-announce/2009-12/msg00027.html where it says - Multiple Cygwin installations can co-exist on a machine, as long as you keep them separate.. -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
Please also try on PPC OS X 10.4, 10.5, and 10.6, and Maybe not the PPC OS X 10.6 ;-) that would be *truly* impressive! I oscillate in my hopes about cygwin... There are other approaches to Oh, but for a double-clickable binary at least! Even if to do development work would be too hard to package. I know I have no moral authority on this as I can't do it, and this has been discussed ad nauseam before, but it would be so awesome, and Mike's binary did eventually work for me... after a bit of not-for-dummies stuff, of course. Sorry, now I'm OT. - kcrisman -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On 26 Okt., 19:36, William Stein wst...@gmail.com wrote: Regarding fortran, for a Microsoft Visual C++ version of Sage, I will just get rid of Fortran (and Lisp) entirely, and not bother with building anything currently in Sage that depends on them... Feed f2c with all Fortran sources and ship the results... ;-) (I'm not sure how to handle Maxima at the moment.) -Leif -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On Tue, Oct 26, 2010 at 11:15 AM, leif not.rea...@online.de wrote: On 26 Okt., 19:36, William Stein wst...@gmail.com wrote: Regarding fortran, for a Microsoft Visual C++ version of Sage, I will just get rid of Fortran (and Lisp) entirely, and not bother with building anything currently in Sage that depends on them... Feed f2c with all Fortran sources and ship the results... ;-) I wish, but that does not work. Believe me, we've tried. (I'm not sure how to handle Maxima at the moment.) -Leif -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org -- William Stein Professor of Mathematics University of Washington http://wstein.org -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
Hi folks, for sime time now, there is a tendency of the Sage distribution to become unmaintanable (only minutes ago, I read the upteenth message thread and trac ticket about the recurrent Suse Linux 11.x/Arch Linux bash/readline issue ...). There are several possibilities/ways to go. One way I explored recently, was to set up Gentoo Prefix on my Mac (see http://www.gentoo.org/proj/en/gentoo-alt/prefix/index.xml), and then try to install gentoo-on-sage onto it (see http://github.com/cschwan/sage-on-gentoo). Although I never had used Gentoo before, and OS X 10.4 is not anymore supported out-of-the-box from Gentoo Prefix (e.g. I had to manually add two more patches to the recent gcc-apple ebuilds), I essentially got through the bootstrapping with success (resulting in a fully working Gentoo Prefix base system of about 100 packages), and got the sage-on-gentoo overlay integrated far enough to see how many more package would be needed (some 150 packages more). That's a total of roughly 250 packages. But remembering that the current Sage distribution has 97 spkg's, I consider this still a very manageable number. And it really means all batteries included, including a gcc, bash, and everything --- except an OS kernel. Conceptually, those 250 packages do work on a Linux kernel, a Darwin kernel (Mac OS), BSD kernels, Solaris/Illumos, and the Windows NT kernel (see the infomation about Gentoo Prefix at the link I gave). In practice, although many of the latter 150 packages do not build yet out-of-the-box in the prefix setting, we already know they essentially do --- since this works for the Sage distribution. And the remaining 50 packages or so do not sound as to be an unsurmountable obstacle (looking closely, the hard ones are either in the base system or in the sage-on-gentoo overlay, so already addressed). As for a better maintanability of the Sage project, the goal would be to start with the 100 packages of the Gentoo Prefix base system, get as many packages as possible into the Gentoo Prefix full system (again, say, another 100 or more of the 250), and maintain only an as slim as possible sage-on-gentoo-prefix overlay, where only the (say, mathematically) really interesting and/or challenging packages remain (three or four dozens in total, not more). In practice however, there are some more obstacles --- e.g. the rolling releases nature of Gentoo (and thus Gentoo Prefix), or the currently less-than-perfect support of the latter to relocate its installation (and thus the currently missing ability to create binary distributions, which Sage can do, this is a must-have). But the general direction seems promising to me, and there are more such generic hosted distibutions (there does not seem to be a common name for this kind of thing) to choose from, e.g. Nixpkg (see http://nixos.org/nixpkgs/), or the CDL-licensed (which is GPL compatible) ATT Advanced Software Technologies Open Source Collection (see http://www2.research.att.com/sw/download/), both of which work under Linux, BSD, OS X --- and Windows (where the former relies on Cygwin, and latter on the UWin environment also provided by ATT under the CDL --- they don't state it on their outdated Webpage, but UWin does work up to and on Windows 7, see their mailing list, and with both MinGW and MS compilers). And since it is such a general approach, it means support of (and ports to), all interesting architectures is essentially only a matter of some (albeight maybe considerable) *initial* efforts, but with very, very low *maintenance* efforts. Anyway, in the Gentoo Prefix case, I did get to compile myself (starting with the Apple gcc v4.01 of the latest available XCode v2.5 for Mac OS X 10.4 Tiger) not only a nicely working gcc v4.2 (for the very first time on this system!), but also the respective gfortran compiler, from source each --- and I bet this way works on OS X 10.4, OS X 10.5, OS X 10.6, and oncoming OS X 10.7, be it Intel or PPC (as far as still supported) systems! Cheers, Georg -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On 10/26/10 06:36 PM, William Stein wrote: I oscillate in my hopes about cygwin... There are other approaches to Sage on Windows, e.g.,: http://windows.sagemath.org/ Is the Cygwin problem just that nobody is working on it, or are there fundamental reasons why it is causing a problem. 6-12 months ago you told me if you and Mike worked at it for a few weeks you could do it. You have a list of tickets that need solving at http://trac.sagemath.org/sage_trac/wiki/CygwinPort I've made comments on several of them, but nobody has replied. One ticket has a patch that has been awaiting review for months. It just seems like nobody is working on it any more. Perhaps you should have finished the Cygwin port before starting PSage, though I can see why the latter is more attractive to you. However, in the long run, getting the Cygwin port finished might attract more research mathematicians which is who PSage is aimed at. So finishing the Cygwin port might ultimately get you PSage developed more quickly. I personally don't see why the Cygwin port should be so hard. From what I understand, there are only a dozen or so doctest issues to resolve. BTW, would it not be better to delete this page? I found it while Googling http://wiki.sagemath.org/windows/cygwin-issues it seems rather out of date. Dave -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On Tue, Oct 26, 2010 at 1:42 PM, Dr. David Kirkby david.kir...@onetel.net wrote: I personally don't see why the Cygwin port should be so hard. From what I understand, there are only a dozen or so doctest issues to resolve. It's not hard -- I've just been busy with other things and haven't worked on it. No one else has worked on it either which is why it is where it is. --Mike -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
[sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
On 26 Okt., 22:42, Dr. David Kirkby david.kir...@onetel.net wrote: BTW, would it not be better to delete this page? I found it while Googling http://wiki.sagemath.org/windows/cygwin-issues it seems rather out of date. Something to keep for software archaeologists. (SCNR) -Leif -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
Re: [sage-devel] Re: Is it *really* necessary for everyone to download fortran.spkg ?
Hi folks, for sime time now, there is a tendency of the Sage distribution to become unmaintanable (only minutes ago, I read the upteenth message thread and trac ticket about the recurrent Suse Linux 11.x/Arch Linux bash/readline issue ...). There are several possibilities/ways to go. One way I explored recently, was to set up Gentoo Prefix on my Mac (see http://www.gentoo.org/proj/en/gentoo-alt/prefix/index.xml), and then try to install gentoo-on-sage onto it (see http://github.com/cschwan/sage-on-gentoo). Although I never had used Gentoo before, and OS X 10.4 is not anymore supported out-of-the-box from Gentoo Prefix (e.g. I had to manually add two more patches to the recent gcc-apple ebuilds), I essentially got through the bootstrapping with success (resulting in a fully working Gentoo Prefix base system of about 100 packages), and got the sage-on-gentoo overlay integrated far enough to see how many more package would be needed (some 150 packages more). That's a total of roughly 250 packages. But remembering that the current Sage distribution has 97 spkg's, I consider this still a very manageable number. And it really means all batteries included, including a gcc, bash, and everything --- except an OS kernel. Conceptually, those 250 packages do work on a Linux kernel, a Darwin kernel (Mac OS), BSD kernels, Solaris/Illumos, and the Windows NT kernel (see the infomation about Gentoo Prefix at the link I gave). In practice, although many of the latter 150 packages do not build yet out-of-the-box in the prefix setting, we already know they essentially do --- since this works for the Sage distribution. And the remaining 50 packages or so do not sound as to be an unsurmountable obstacle (looking closely, the hard ones are either in the base system or in the sage-on-gentoo overlay, so already addressed). As for a better maintanability of the Sage project, the goal would be to start with the 100 packages of the Gentoo Prefix base system, get as many packages as possible into the Gentoo Prefix full system (again, say, another 100 or more of the 250), and maintain only an as slim as possible sage-on-gentoo-prefix overlay, where only the (say, mathematically) really interesting and/or challenging packages remain (three or four dozens in total, not more). In practice however, there are some more obstacles --- e.g. the rolling releases nature of Gentoo (and thus Gentoo Prefix), or the currently less-than-perfect support of the latter to relocate its installation (and thus the currently missing ability to create binary distributions, which Sage can do, this is a must-have). But the general direction seems promising to me, and there are more such generic hosted distibutions (there does not seem to be a common name for this kind of thing) to choose from, e.g. Nixpkg (see http://nixos.org/nixpkgs/), or the CDL-licensed (which is GPL compatible) ATT Advanced Software Technologies Open Source Collection (see http://www2.research.att.com/sw/download/), both of which work under Linux, BSD, OS X --- and Windows (where the former relies on Cygwin, and latter on the UWin environment also provided by ATT under the CDL --- they don't state it on their outdated Webpage, but UWin does work up to and on Windows 7, see their mailing list, and with both MinGW and MS compilers). And since it is such a general approach, it means support of (and ports to), all interesting architectures is essentially only a matter of some (albeight maybe considerable) *initial* efforts, but with very, very low *maintenance* efforts. Anyway, in the Gentoo Prefix case, I did get to compile myself (starting with the Apple gcc v4.01 of the latest available XCode v2.5 for Mac OS X 10.4 Tiger) not only a nicely working gcc v4.2 (for the very first time on this system!), but also the respective gfortran compiler, from source each --- and I bet this way works on OS X 10.4, OS X 10.5, OS X 10.6, and oncoming OS X 10.7, be it Intel or PPC (as far as still supported) systems! Well thanks for the plug for our work. We are quite happy to develop the prefix part of sage-on-gentoo, so far the effort has been limited but if you are testing it on an arch we are quite happy to try to get as much as possible keyworded. By the way did you try sage-4.5.3 or 4.6 (alphaX/rc0)? Indeed gentoo do not offer you the possibility to have relocatable binaries as such. It's not the point in a way. However gentoo can create binary packages (on an individual package basis) so it would be possible to install binaries from a base gentoo prefix install. Francois -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
