On Fri, 23 Apr 2010, Satish Balay wrote: > On Fri, 23 Apr 2010, Sean Farley wrote: > > > > > > > Essentially, yes. The advantage of actually using pkgconfig is that > > > autoconf, cmake, and others already have support for it. > > > > > > True. > > > > Actually, I'm almost done (once DESTDIR gets supported) with implementing a > > Portfile for PETSc in MacPorts (which has the ability to have multiple > > versions of packages installed), if anyone uses a mac and is interested in > > that. > > One Issue [which might be mac specific] with DESTDIR is: 'make > install' does 'make shared' at the dest dir.
For now I pushed a priliminary DESTDIR support to petsc-dev. Currently it skips sharedlibraries. Since the conf files in DESTDIR are unuseable [because they point to INSTALL_DIR] we cannot invoke make targets with PETSC_DIR=DESTDIR So we'll have to move 'make shared petsc4py' [and other stuff] from install.py back to 'make all' [and get it working perhaps with 'install_name_tool' on mac] Th no-DESTDIR mode should continue to work as before [wrt 'make shared petsc4py' etc.. Satish > > I'm assuming - one of the reasons for this is: .dylib on Macs hardcode > some location info. If so - DESTDIR won't work on Mac [with > sharedlibs?]. If not - there is no need to do 'make shared' during > 'make install'? [Barry?] > > Also --download-petsc4py does the petsc4py build during 'make install' > - we just have to make sure this continues to work with DESTDIR.. > > Satish >
