Very strange (possibly compiler) bug with date2num

[nichannah]

The patch below introduces one passing and one failing test. This problem occurred with intel 17 but not 15, 16 or 18. It did not happen with gfortran.

diff --git a/src/tests/datetime_tests.f90 b/src/tests/datetime_tests.f90
index 0abb98d..799fadb 100644
--- a/src/tests/datetime_tests.f90
+++ b/src/tests/datetime_tests.f90
@@ -1026,6 +1026,16 @@ SUBROUTINE test_datetime
                     'datetime % utc() == num2date(date2num(datetime)) (now)')
   n = n+1

+  a = datetime(1901,05,1)
+  tests(n) = assert(a == num2date(date2num(a)),&
+                    'datetime == num2date(date2num(datetime)) (weird bug passes)')
+  n = n+1
+
+  a = datetime(1901,05,1,3)
+  tests(n) = assert(a == num2date(date2num(a)),&
+                    'datetime == num2date(date2num(datetime)) (weird bug fails)')
+  n = n+1
+
   ! Test for overflowing month
   a = datetime(2014,11,30,1)
   tests(n) = assert(a == num2date(date2num(a)),&

Intel 14 gives a compiler error

Scanning dependencies of target datetime_tests
[ 88%] Building Fortran object CMakeFiles/datetime_tests.dir/src/tests/datetime_tests.f90.o
/home/599/nah599/more_home/Download/datetime-fortran/src/tests/datetime_tests.f90(1): catastrophic error: **Internal compiler error: internal abort** Please report this error along with the circumstances in which it occurred in a Software Problem Report.  Note: File and line given may not be explicit cause of this error.
compilation aborted for /home/599/nah599/more_home/Download/datetime-fortran/src/tests/datetime_tests.f90 (code 1)
make[2]: *** [CMakeFiles/datetime_tests.dir/src/tests/datetime_tests.f90.o] Error 1
make[1]: *** [CMakeFiles/datetime_tests.dir/all] Error 2
make: *** [all] Error 2

Intel 13 gives a segmentation fault:

test datetime == num2date(date2num(datetime)) (weird bug passes) : PASS
test datetime == num2date(date2num(datetime)) (weird bug fails)  : PASS
test datetime == num2date(date2num(datetime)) (overflowing month): PASS
test datetime == num2date(date2num(datetime)) (overflowing year) : PASS
-----------------------------------------------------------------------
forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image              PC                Routine            Line        Source
datetime_tests     0000000000452820  mod_datetime_mp_e         998  mod_datetime.f90
datetime_tests     000000000043D63D  datetime_tests_mp        1066  datetime_tests.f90
datetime_tests     0000000000448916  MAIN__                   1284  datetime_tests.f90
datetime_tests     000000000040300C  Unknown               Unknown  Unknown
libc.so.6          00002B896DBB1D1D  Unknown               Unknown  Unknown
datetime_tests     0000000000402F09  Unknown               Unknown  Unknown

[milancurcic]

Good catch and I agree, very strange bug. I only have ifort 18 with me right now and both test pass as expected. Maybe filing this on the Intel forums would be a good idea if this issue becomes a showstopper for you?

[milancurcic]

date2num and num2date use real64 from iso_fortran_env for some literal constants, I'm wondering if its implementation varies between these versions of ifort.

[nichannah]

I'm no longer able to reproduce this error with out latest compiler versions:

intel-compiler/2019.5.281 intel-compiler/2020.0.166