From: Dmitry K. <ka...@mc...> - 2013-01-14 18:30:21
|
Roy, Yes, the problem was with PETSc's makefile. It existed in the early versions of petsc-3.3, but was fixed by one of the early patches: https://bitbucket.org/petsc/petsc-dev/diff/makefile?diff2=a0fe59a5123c&at=default I was running the latest petsc-3.3 and petsc-dev everywhere _except_ the machine in question, which prompted me to blame the machine. Without the fix above make -C $PETSC_DIR -s getincludedirs and similar will fail, if run from anywhere other than $PETSC_DIR. It looks like the problem was short lived (introduced here: https://bitbucket.org/petsc/petsc-dev/diff/makefile?diff2=5f84ba60b4ac&at=default), so libmesh configure probably needn't worry about it. Dmitry. On Fri, Jan 11, 2013 at 1:30 PM, Roy Stogner <roy...@ic...>wrote: > > On Fri, 4 Jan 2013, Roy Stogner wrote: > > On Fri, 4 Jan 2013, John Peterson wrote: >> >> On Fri, Jan 4, 2013 at 11:22 AM, Dmitry Karpeev <ka...@mc...> >>> wrote: >>> >> >> The symptoms are these: configure finds petsc, but leaves >>>> PETSCINCLUDEDIRS >>>> and >>>> PETSCLINKLIBS blank, which results in an >>>> insufficient libmesh_optional_INCLUDES, which >>>> in turn results in a failure to find petsc.h at build time. >>>> >>> >>> I might have broken this with my recent changes to petsc.m4... >>> >> >> Yeah; I now get breakage myself when trying to build on Xubuntu with >> clang, when it fails to include mpi.h. It looks like, on the path I >> hit, we're correctly scarfing PETSC_CC_INCLUDES but never >> incorporating that into PETSCINCLUDEDIRS? >> > > I just pushed a fix for my own problem. It appears: > > The problem wasn't John's fault, just a slight change to PETSc > variables when I upgraded Ubuntu which (in the CXX=clang++ scenario) > our petsc.m4 wasn't prepared for. > > My solution won't fix Dmitry's problem. Even if somehow his > configuration is leading to an empty PETSCINCLUDEDIRS which gets fixed > up by now adding PETSC_CC_INCLUDES, nothing I did will lead to a fixed > PETSCLINKLIBS. > > I'm not even sure what the *explanation* is for Dmitry's problem. > Either his $PETSC_DIR/makefile is giving empty outputs for those > targets, or neither it nor $PETSC_DIR/conf/variables exist as readable > files on his system? > > At least the latter problem we should probably be testing for, but > adding a final "else" to that if/elif test... > --- > Roy > -- Dmitry Karpeev, Ph.D. Assistant Computational Mathematician Mathematics and Computer Science Argonne National Laboratory Argonne, Illinois, USA and Fellow Computation Institute University of Chicago 5735 S. Ellis Avenue Chicago, IL 60637 ----------------------- Phone: 630-252-1229 Fax: 630-252-5986 |