William Stein wrote: > On Sun, Sep 27, 2009 at 2:12 PM, Dr. David Kirkby > <david.kir...@onetel.net> wrote: >> William Stein wrote: >> >>>> Just to add that even the very latest release of Solaris (Solaris 10 >>>> update 7) on SPARC only ships with gcc 3.4.3. Users using SPARC >>>> hardware are unlikely to have root access as it tends not to be used by >>>> home users too much (yours truly is a bit of an exception). >>>> >>>> I just checked on 'disk' too, which is running the November 2008 version >>>> of OpenSolaris, and that too only appears to have 3.4.3. >>> I'm not sure how much of an argument that is, since I'm not enough of >>> a masochist to try build Sage on Solaris with the system-wide standard >>> compilers. >> I don't follow that argument. >> >> If the standard compiler will build Sage, it gets over a big hurdle. >> Building gcc from source on Solaris is no easy task. Lots of people have >> trouble with it. >> >> Lots of people wont have root access on Solaris machines. Neither >> Micheal nor I have managed to produce a gcc binary that did not require >> LD_LIBRARY_PATH to be set to include the compiler's libraries. >> >> I think Sage binaries produced for distribution would be smaller, as >> there would be no need to include various gcc libraries. >> >> So IMHO, if Sage would build with 3.4.x, it would be a big plus. > > I certainly agree that it would be a huge plus. I just consider it > "too difficult". Moreover, every single person I know but you who has > put work into porting Sage to Solaris has started by building their > own GCC 4.x and toolchain.
I have too. I'm using my own custom toolchain. > These are all people who are extremely > good at porting (not just Michael Abshoff, but other people too). I > think they have a point. Can one build sage-4.1.1 on Solaris right > now and have it pass all tests, even with a custom toolchain? Building gcc on Solaris is not easy. It is very easy to waste huge amounts of time doing this. Plenty of people find building gcc on Solaris a real pain - many simply give up. Virtually anyone (Blastwave etc) producing binaries for gcc on Solaris SPARC use the Sun linker and assembler. The GNU ones are known to be troublesome, especially with 64-bit code. Unfortunately, Sage made extensive use of linker options that were GNU specific, which prevented a gcc built with the superior Sun assembler and linker from working. That I believe is the main reason Michal needed a custom toolchain, as his used the inferior GNU assembler and linker. I believe I've now sorted all the code in Sage that assumes a GNU linker and assembler if built with gcc. (If using Sun Studio, there are still some bits of code that fail because of GCC-specific linker options). As such, there is less need to have a custom gcc toolchain now. Ratpoints might be the only reason. People porting to Solaris probably will use a later gcc. But if one is not a programmer, but just wants to build Sage easily, having it work with the compiler supplied in Solaris would be a huge benefit. You might note the person from comp.unix.solaris that tried to build Sage on OpenSolaris recently used the gcc in /usr/sfw/bin. That compiler works well. It does not however support fortran - it is only built for C and C++. >> How much of an issue is it to sort out ratpoints? > > I suspect it would be easy, but I haven't tried. But above you consider supporting 3.4.x "too difficult". --~--~---------~--~----~------------~-------~--~----~ 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 -~----------~----~----~----~------~----~------~--~---
