https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82934
--- Comment #3 from Paul Thomas <pault at gcc dot gnu.org> --- (In reply to Dominique d'Humieres from comment #1) > The test gives the expected result with gfortran 4.8 and 4.9. The change in > behavior occurred between revisions r221412 + 3 patches (2015-03-13, works) > and r2214164 + 3 patches (2015-03-16, ICE). Oddly, the fortran ChangeLog shows nothing between these dates that one could implicate. As you will see, the fix was trivial. I am just about to post it on the list and will apply it to 6- and 7-branches over the weekend. Even if not a regression, it does prevent an ICE in a particularly robust way :-) Thanks for investigating. Paul