https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69423
--- Comment #10 from Paul Thomas <pault at gcc dot gnu.org> --- Created attachment 37686 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=37686&action=edit Provisional patch for the PR I have reworked the handling of deferred string function results to fix this PR. Clearly, using the indirect reference to the passed string length is unsafe; whether an optimization bug or not. It's a pity because it was quite elegant! Instead, I have used the original approach of an in-scope character length, which is nulled on entry and assigned to the result length on exit. Catching all the various cases took quite a lot of rummaging around in trans-decl.c(gfc_trans_deferred_vars)! The patch is attached. It needs a LOT of cleaning up before submission, which i hope to accomplish in the coming couple of days. However, it regtests OK on the current trunk as it stands. The corresponding deja-gnu-ified testcase is copied below. Paul ! { dg-do run } ! ! Test the fix for PR69423. ! ! Contributed by Antony Lewis <ant...@cosmologist.info> ! program tester character(LEN=:), allocatable :: S S= test(2) if (len(S) .ne. 4) call abort if (S .ne. "test") call abort if (allocated (S)) deallocate (S) S= test2(2) if (len(S) .ne. 4) call abort if (S .ne. "test") call abort if (allocated (S)) deallocate (S) contains function test(alen) character(LEN=:), allocatable :: test integer alen, i do i = alen, 1, -1 test = 'test' exit end do ! This line would print nothing when compiled with -O1 and higher. ! print *, len(test),test if (len(test) .ne. 4) call abort if (test .ne. "test") call abort end function test function test2(alen) result (test) character(LEN=:), allocatable :: test integer alen, i do i = alen, 1, -1 test = 'test' exit end do ! This worked before the fix. ! print *, len(test),test if (len(test) .ne. 4) call abort if (test .ne. "test") call abort end function test2 end program tester
