On 1/2/26 4:59 PM, Jakub Jelinek wrote:
On Thu, Jan 01, 2026 at 01:43:50PM -0800, Jerry D wrote:
On 1/1/26 1:33 PM, Thomas Koenig wrote:
This will avoid potential undefined behavior on various systems. I
will make this change and test. If passes, OK for mainline?
Good thinking.
Your revised version is OK if it passes.
Thanks a lot for the patch!
Best regards
Thomas
Pushed:
commit 489423763d3c8b84d3409f4b200fb6b19ad96db3 (HEAD -> master,
origin/master, origin/HEAD)
Author: Jerry DeLisle <[email protected]>
Date: Tue Dec 30 14:46:35 2025 -0800
Fortran: Generate a runtime error on recursive I/O
PR libfortran/119136
This broke a lot of libgomp tests on all major targets.
Compare e.g.
https://gcc.gnu.org/pipermail/gcc-testresults/2026-January/866363.html
and
https://gcc.gnu.org/pipermail/gcc-testresults/2026-January/866375.html
FAIL: libgomp.fortran/appendix-a/a.2.1.f90 -O3 -fomit-frame-pointer
-funroll-loops -fpeel-loops -ftracer -finline-functions execution test
FAIL: libgomp.fortran/appendix-a/a.22.8.f90 -O0 execution test
FAIL: libgomp.fortran/appendix-a/a.22.8.f90 -O1 execution test
FAIL: libgomp.fortran/appendix-a/a.22.8.f90 -O2 execution test
FAIL: libgomp.fortran/appendix-a/a.22.8.f90 -O3 -fomit-frame-pointer
-funroll-loops -fpeel-loops -ftracer -finline-functions execution test
FAIL: libgomp.fortran/appendix-a/a.28.1.f90 -O0 execution test
FAIL: libgomp.fortran/appendix-a/a.28.1.f90 -O2 execution test
FAIL: libgomp.fortran/appendix-a/a.28.1.f90 -O3 -fomit-frame-pointer
-funroll-loops -fpeel-loops -ftracer -finline-functions execution test
FAIL: libgomp.fortran/appendix-a/a.28.1.f90 -O3 -g execution test
FAIL: libgomp.fortran/appendix-a/a.28.1.f90 -Os execution test
FAIL: libgomp.fortran/appendix-a/a.28.2.f90 -O0 execution test
FAIL: libgomp.fortran/appendix-a/a.28.2.f90 -O1 execution test
FAIL: libgomp.fortran/appendix-a/a.28.2.f90 -O2 execution test
FAIL: libgomp.fortran/appendix-a/a.28.2.f90 -O3 -fomit-frame-pointer
-funroll-loops -fpeel-loops -ftracer -finline-functions execution test
FAIL: libgomp.fortran/appendix-a/a.28.2.f90 -Os execution test
FAIL: libgomp.fortran/appendix-a/a.28.3.f90 -O0 execution test
FAIL: libgomp.fortran/appendix-a/a.28.3.f90 -O1 execution test
FAIL: libgomp.fortran/appendix-a/a.28.3.f90 -O2 execution test
FAIL: libgomp.fortran/appendix-a/a.28.3.f90 -O3 -fomit-frame-pointer
-funroll-loops -fpeel-loops -ftracer -finline-functions execution test
FAIL: libgomp.fortran/appendix-a/a.28.3.f90 -O3 -g execution test
FAIL: libgomp.fortran/appendix-a/a.28.3.f90 -Os execution test
FAIL: libgomp.fortran/depend-4.f90 -O execution test
FAIL: libgomp.fortran/omp_hello.f -O execution test
FAIL: libgomp.fortran/omp_orphan.f -O execution test
FAIL: libgomp.fortran/omp_workshare1.f -O execution test
FAIL: libgomp.fortran/omp_workshare2.f -O execution test
FAIL: libgomp.fortran/recursion1.f90 -O0 execution test
FAIL: libgomp.fortran/recursion1.f90 -O1 execution test
FAIL: libgomp.fortran/recursion1.f90 -O2 execution test
FAIL: libgomp.fortran/recursion1.f90 -O3 -fomit-frame-pointer -funroll-loops
-fpeel-loops -ftracer -finline-functions execution test
FAIL: libgomp.fortran/recursion1.f90 -O3 -g execution test
FAIL: libgomp.fortran/recursion1.f90 -Os execution test
FAIL: libgomp.fortran/rwlock_2.f90 -O0 execution test
FAIL: libgomp.fortran/rwlock_2.f90 -O1 execution test
FAIL: libgomp.fortran/rwlock_2.f90 -O2 execution test
FAIL: libgomp.fortran/rwlock_2.f90 -O3 -fomit-frame-pointer -funroll-loops
-fpeel-loops -ftracer -finline-functions execution test
FAIL: libgomp.fortran/rwlock_2.f90 -O3 -g execution test
FAIL: libgomp.fortran/rwlock_2.f90 -Os execution test
all the errors in the logs look like
1: THREAD# 1 X = 5
2: THREAD# 0 X = 5
At line 19 of file
/home/jakub/src/gcc/libgomp/testsuite/libgomp.fortran/appendix-a/a.2.1.f90
(unit = 6, file = 'stdout')
Fortran runtime error: Recursive I/O not allowed
Error termination. Backtrace:
#0 0xf7e464b0 in data_transfer_init
at ../../../libgfortran/io/transfer.c:3132
#1 0x8048457 in ???
#2 0xf7a7da32 in gomp_thread_start
at ../../../libgomp/team.c:129
#3 0xf78a51df in ???
#4 0xf792b2d7 in ???
#5 0xffffffff in ???
Recursive I/O shouldn't indeed be allowed, but multiple threads
trying to transfer concurrently should be fine (of course, locking
should be used for that).
Jakub
I am looking for a way to check if libgomp is active and if so skip the check.
If there is no such mechanism than I will revert the patch and we will need to
find some other way to detect the recursive I/O.
Jerry
