------- Comment #4 from janus at gcc dot gnu dot org 2010-08-27 09:06 ------- (In reply to comment #3) > First, the patch did not apply cleanly, the first hunk was rejected. I applied > it by hand, and I got a problem down the road in my library: > =============================================================== > gfortran -ggdb -I.. -I../modules -I. -c psb_srwextd.f90 > psb_srwextd.f90:76.13: > > call aa%mv_to_coo(actmp,info) > 1 > Error: Actual argument at (1) must be definable as the dummy argument 'a' is > INTENT = OUT/INOUT > psb_srwextd.f90:84.39: > > if (info == psb_success_) call aa%mv_from_coo(actmp,info) > 1 > Error: Actual argument at (1) must be definable as the dummy argument 'a' is > INTENT = OUT/INOUT
I tried to reproduce this by modifying your original test case, but did not succeed so far. Can you provide a standalone test case for this problem? My feeling is that this is another bug uncovered by the fix for the previous one. The patch in comment #2 is free of testsuite regressions and certainly does the right thing. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45420