http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55069
--- Comment #3 from Sean Santos <quantheory at gmail dot com> 2012-10-25 17:49:44 UTC --- Ah, it seems I was quite tired yesterday and made a couple mistakes: - I mixed up a test using an older gfortran version with a test with optimization off. I suppose that this is not a regression after all. - At some point, I seem to have changed my test case to make "blah" a parameter. If "blah" is a parameter, the behavior I described actually is what happens. Sorry about the mixups. As for updating compilers, unfortunately this is not really a decision I can make. My group is required to maintain this software for on several specific machines (not under our direct control) that can go >1 year between major compiler updates. I'm only even using 4.7 because I expect that my current work won't get merged to the trunk for several months anyway. If 4.8 is released in March, I can't expect to release anything that depends on it until 2014. My current workaround is to simply concatenate the empty string onto every affected assignment. Luckily there are not many, and this appears to work for every test case in my limited needs.
