On Fri, 2008-04-04 at 12:17 -0400, Adam C Powell IV wrote: > On Fri, 2008-04-04 at 17:18 +0200, Domenico Andreoli wrote: > > retitle 474249 petsc: FTBFS: wrongly guess the MPI directory > > tags 474249 +patch > > thanks > > > > On Fri, Apr 04, 2008 at 04:58:26PM +0200, Domenico Andreoli wrote: > > > > > > It looks like definition of PETSC_MPI in debian/rules is somewhat pesky, > > > probably it is broken by localization issues.. don't know, but on my > > > hppa it comes to "udelam" instead of "lam", which is wrong. > > > > > > I suggest you to rewrite such guessing using readlink command. > > > > > > > --- rules.orig 2008-04-04 17:02:38.000000000 +0200 > > +++ rules 2008-04-04 17:07:00.000000000 +0200 > > @@ -26,7 +26,7 @@ > > # Note that as of PETSc 2.3.0 this only specifies the name of the PETSc > > # packages, the implementation is chosen by BuildSystem according to what's > > # available at build time. > > -PETSC_MPI=$(shell ls -l /etc/alternatives/mpi | cut -b 74- | sed > > s/include//g | sed s/lib//g | sed s/\\///g) > > +PETSC_MPI=$(shell readlink /etc/alternatives/mpi | xargs basename) > > PETSC_MPI_DIR=/usr/lib/$(PETSC_MPI) > > > > # Overriding this with "contrib" attempts to link with hypre and parmetis. > > Thank you! I saw that error and had no clue about how to proceed.
Ah, trouble in paradise. The lam link is to /usr/include/lam where this works, but the openmpi link is to /usr/lib/openmpi/include so this returns "include" for PETSC_MPI. I'm fixing it using a kludgey sed string to get rid of usr, lib, include, then / which should work for both. > Does the package build on HPPA with this patch? On a bunch of other > arches, it's getting stuck in petsc's python-based build system, > somewhat after the point where it broke on HPPA. That one issue in > petsc is holding up a massive transition of lots of packages... Found a solution to this problem. BTW, thanks for building and uploading spooles on HPPA! (That was you, wasn't it?) -Adam -- GPG fingerprint: D54D 1AEE B11C CE9B A02B C5DD 526F 01E8 564E E4B6 Engineering consulting with open source tools http://www.opennovation.com/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
