Floating point overflow in gsw_poly_check test

[DJDavies2]

Building and running the ctests gives this failure:

[ 97%] Built target gsw_objects
[ 98%] Built target gsw_shared
[ 99%] Built target gsw_static
[ 99%] Building Fortran object test/CMakeFiles/gsw_poly_check.x.dir/gsw_poly_check.f90.o
[100%] Linking Fortran executable gsw_poly_check.x
[100%] Built target gsw_poly_check.x
[100%] Building Fortran object test/CMakeFiles/gsw_check_functions.x.dir/gsw_check_functions.f90.o
[100%] Linking Fortran executable gsw_check_functions.x
[100%] Built target gsw_check_functions.x
Test project /home/users/david.davies/cylc-run/mi-bg854/work/1/build_gsw_fortran_teos_spicenag700/build
Start 1: gsw_poly_check
1/2 Test #1: gsw_poly_check ...................Subprocess aborted***Exception: 0.15 sec
Runtime Error: *** Arithmetic exception: Floating overflow - aborting
/home/users/david.davies/cylc-run/mi-bg854/work/1/get_source_gsw_fortran_teos/GSW-Fortran/toolbox/gsw_t_freezing_poly.f90, line 2: Error occurred in GSW_T_FREEZING_POLY
/home/users/david.davies/cylc-run/mi-bg854/work/1/get_source_gsw_fortran_teos/GSW-Fortran/toolbox/gsw_t_freezing_exact.f90, line 43: Called by GSW_T_FREEZING_EXACT
/home/users/david.davies/cylc-run/mi-bg854/work/1/get_source_gsw_fortran_teos/GSW-Fortran/toolbox/gsw_ct_freezing_exact.f90, line 37: Called by GSW_CT_FREEZING_EXACT
/home/users/david.davies/cylc-run/mi-bg854/work/1/get_source_gsw_fortran_teos/GSW-Fortran/test/gsw_poly_check.f90, line 56: Called by GSW_POLY_CHECK

Adding some print statements in gsw_t_freezing_poly.f90 (and removing the elemental) gives these values in the lead up to the calculation:

in gsw_t_freezing_poly, sa = 8.9999999999999998E+90
in gsw_t_freezing_poly, sa_r = 8.9999999999999998E+88
in gsw_t_freezing_poly, x = 2.9999999999999999E+44
in gsw_t_freezing_poly, p_r = 8.9999999999999999E+86
Runtime Error: *** Arithmetic exception: Floating overflow - aborting

I am compiling with nagfor 7.0 via cmake.

[bolding]

Just to note - using gfortran 13 - I don't have an issue:

[bolding]

just looked a bit more - and the logs from the tests actually show some issues - but not enough to give e.g. gsw_check_funnctions.x a return code different from 0

[DJDavies2]

I think for some compilers where the test passes the answer does overflow but the overflow doesn't signal so there isn't a failure.

[bolding]

can you upload your modified file and I'll try with that?

[DJDavies2]

I've uploaded my prints in this branch: https://github.com/DJDavies2/GSW-Fortran/tree/prints-for-issue-29

For NAG I get:

in gsw_t_freezing_poly, sa = 8.9999999999999998E+90
in gsw_t_freezing_poly, sa_r = 8.9999999999999998E+88
in gsw_t_freezing_poly, x = 2.9999999999999999E+44
in gsw_t_freezing_poly, p_r = 8.9999999999999999E+86
Runtime Error: *** Arithmetic exception: Floating overflow - aborting
/home/users/david.davies/cylc-run/mi-bg854/work/1/get_source_gsw_fortran_teos/GSW-Fortran/toolbox/gsw_t_freezing_poly.f90, line 71: Error occurred in GSW_T_FREEZING_POLY
/home/users/david.davies/cylc-run/mi-bg854/work/1/get_source_gsw_fortran_teos/GSW-Fortran/toolbox/gsw_t_freezing_exact.f90, line 43: Called by GSW_T_FREEZING_EXACT
/home/users/david.davies/cylc-run/mi-bg854/work/1/get_source_gsw_fortran_teos/GSW-Fortran/toolbox/gsw_ct_freezing_exact.f90, line 37: Called by GSW_CT_FREEZING_EXACT
/home/users/david.davies/cylc-run/mi-bg854/work/1/get_source_gsw_fortran_teos/GSW-Fortran/test/gsw_poly_check.f90, line 56: Called by GSW_POLY_CHECK

For e.g. gnu I get:

1: in gsw_t_freezing_poly, sa = 8.9999999999999998E+090
1: in gsw_t_freezing_poly, sa_r = 8.9999999999999998E+088
1: in gsw_t_freezing_poly, x = 2.9999999999999999E+044
1: in gsw_t_freezing_poly, p_r = 8.9999999999999999E+086
1: in gsw_t_freezing_poly, gsw_t_freezing_poly = Infinity

and then it carries on and passes the test.