On 14 Dez., 05:59, David Kirkby <david.kir...@onetel.net> wrote: > If anyone said including our own compiler with Sage is taking the > "batteries included" approach too far, I would NOT disagree with them, > > But there do appear to be issues which may be difficult / impossible > to resolve with Sage upgrades using gcc that uses Sage's GMP, MPFR and > MPC (See Leif's comments on this issue). > > We could avoid that by shipping our own gcc. I calculate it would add > only 1.6 MB to the bulk of Sage if we used the latest (4.5.1) gcc and > could *reduce* the size of Sage if we used an earlier version. > > Given gcc 4.5.1 is 63 MB in size, you might wonder why I calculate > shipping gcc would add <= 1.6 MB. > > First we would not need the Java or Objective C front ends, so there's > no need to include the full gcc release. So we would add a package > which includes only: > > * gcc core (27 MB) > * C++ front end (6.2 MB) > * Fortran front end (1.4 MB) > > Then we REMOVE the Fortran binaries, which are only needed on OS X. > These take up 33 MB. > > sage: 27+6.2+1.4-33 > 1.60000000000000 > sage: > > Pros: > * No worries about gcc finding the wrong libraries during upgrades, as > it would use those in $SAGE_LOCAL/lib > * We would know what version of gcc people would using, so would not > have to contend with gcc-specific bugs. > * Building on Solaris would be a lot easier, as nobody would need to > install a recent gcc. > * We could remove the code in bits of Sage that checks if g95 or gcc > is used, since we know gcc would be used. > * We could remove the code in ATLAS (and perhaps elsewhere) which > tests if g95 or gfortran are used. > * We could forget about setting SAGE_FORTRAN and SAGE_FORTRAN_LIB > variables as they would be unneeded. > * If instead of using the latest gcc, we used an earlier version, we > would undoubtedly make the compiler package smaller than the Fortran > binaries we remove. > > Cons: > * It would make Sage 1.6 MB larger > * It would slow builds on any system where the Sage gcc was used in > preference to a system one. > > It would be sensible to add a bit of code that allows one to use the > system compilers.
;D :D :D An absolutely great idea. We could also remove targets not supported by Sage from the GCC source tree. One problem remains: If we ship a more recent GCC, we still need to build GMP/MPIR, perhaps also MPFR and MPC, with the system's C compiler, preferably some GCC version, at least for bootstrapping our compiler. (Btw., I failed in trying to compile some older versions of GCC with newer ones, e.g. 4.0.x with 4.3.3, 4.4.3 and 4.5.1.) Thanks for that post, you made my day. Perhaps people *now* get more convinced of removing the Fortran compiler binaries from the Sage source tarball. :) -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
