On Thu 2008-12-04 16:58, Lisandro Dalcin wrote: > On Thu, Dec 4, 2008 at 4:42 PM, Barry Smith <bsmith at mcs.anl.gov> wrote: > > > > Make is NOT the problem! (It is just one of several) > > Indeed. However, at some time I'll try to make PETSc build with Scons.
Presumably you have some experience with SCons, but I think it's really not such a good way to go. It's pretty slow for large projects and the caching design seems to generate inconsistent state somewhat regularly. Also, big projects seem to always end up with a fork of SCons or abandon it for alternatives. I've been using CMake for my stuff (linking against PETSc and a few other libs). The syntax is pretty hideous for scripting, but the declarative build definition is very nice. Since it uses the native build system, the IDE-using people on Windows would probably prefer it as well. It would be trivial (like an hour, plus some for tests) to replace the current recursive make with CMake, replacing BuildSystem would obviously require a bit of tedium, if it was even desirable. BuildSystem is a very different beast from the "configuration" tools that are out there since it also fills in as a package manager (extremely nice since lots of optional dependencies have really painful build systems). In any case, having proper dependency analysis would be *really* cool, a do-nothing rebuild of ParaView (includes VTK, >2M LOC in many directories) takes 10 seconds so the usual PETSc few-file recompile after 'hg pull -u' would be a 2-second affair. Anyway, I'd suggest having a look at CMake before implementing an SCons build. Jed -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 197 bytes Desc: not available URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20081204/a06a6165/attachment.pgp>