https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83904
Bug ID: 83904
Summary: gfortran.dg allocatable_function_1 failures
Product: gcc
Version: 7.2.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: testsuite
Assignee: unassigned at gcc dot gnu.org
Reporter: wilson at gcc dot gnu.org
Target Milestone: ---
Some of my build trees are getting allocatable_function_1 failures. Some of
them are not. Looking at this, I see in the failing trees:
_gfortran_runtime_error_at (&"At line 30 of file
/scratch/jimw/freedom-u-sdk/lra-bug-test/riscv-gnu-toolchain/riscv-gcc/gcc/testsuite/gfortran.dg/
allocatable_function_1.f90"[1]{lb: 1 sz: 1}, &"Attempting to allocate already
allocated variable \'%s\'"[1]{lb: 1 sz: 1}, &"foo3"[1]{lb: 1 sz: 1});
The testcase is counting the number of times that the string "free" appears in
the dump file, and gets the count wrong when the pathname of the file contains
the string "free" because of these gfortran_runtime_error_at calls.
I managed to find other examples this problem. For instance:
https://gcc.gnu.org/ml/gcc-testresults/2011-12/msg00997.html
This is a freebsd testsuite run which happens to contain freebsd in the
pathname.
Looking for a more unique string would help, e.g. searching for __builtin_free
solves my problem, but it would be better if the gfortran_runtime_error_at call
was not scanned for strings. If there a gfortran compiler option that avoids
the call the testsuite should use it. Otherwise, it might make sense to fix
the scanner to ignore lines that include gfortran_runtime_error_at_call.
I also noticed that there is a typo in the testcase. There is a comment that
says 9 free calls are expected, but the scan-tree-dump-times expects 10. I see
that the number changed from 9 to 10 in a 2006-12-05 patch from Paul Thomas,
and he fixed the scan-tree-dump-times line, but missed the comment.
In the short term, we should fix the comment, and change the search string from
free to __builtin_free. In the longer term, we probably need to fix the
testsuite to handle this better.
