http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40756
--- Comment #3 from Dominique d'Humieres <dominiq at lps dot ens.fr> --- > Let's reopen. I want to add a TREE_BLOCK for (module) global variables > for namelists, cf. PR37132. Those would be used by the BE/LTO. I don't this is related to this PR. > Still, that's the opportunity to check whether we add TREE_BLOCK > where it doesn't make sense. Looking at the code, we seem might > add pointless ones with poplevel in f95-lang.c. Although, Diego stated: > "They are also reachable from EXPRs embedded in some types emitted > by Fortran." - hence, there might be additional places where they get added. This PR has roten for nearly four years without any example of such pointless TREE_BLOCK nor a way to find such instances. If there is really a "memory-hog" associated with it, a pointer to it is needed.
