Hi Paul, > The attached patch fixes PR86328 and PR86760. The regression was > caused by my commit r252949. > > The parts of the patch that fix the PRs are in trans.c and > trans-array.c. The problem was caused by fixing the expressions that > would provide the 'span' in gfc_build_array_ref, since the latter > expected a variable expression. A number of evaluations of component > array elements were producing pre blocks that were not added and so > the temporaries were not being evaluated. > > The fix is to pass the COMPONENT_REF and extract the 'span' directly from it. > > The rest of the patch arises from PR86328 comment #12. In fact, this > took most of the time that I have spent on these PRs :-( Having done > this, I felt that I had to include this part of the patch in the > submission. However, I have found a host of related bugs, which I will > put together in one PR. > > My inclination is to commit the patch without the parts in resolve.c, > trans-expr.c and pr86328_12.f90, especially for 8-branch. I am open to > suggestions for 9-branch. > > Bootstraps and regtests on FC28/x68_64 - OK for 8- and 9-branches?
the patch is ok for trunk from my side. I also agree that it makes sense to backport those parts that address the regression to 8-branch. Thanks for the fix! Cheers, Janus
