https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94070
Bug ID: 94070
Summary: Assumed-rank arrays – bounds mishandled,
SIZE/SHAPE/UBOUND/LBOUND
Product: gcc
Version: 10.0
Status: UNCONFIRMED
Keywords: wrong-code
Severity: normal
Priority: P3
Component: fortran
Assignee: unassigned at gcc dot gnu.org
Reporter: burnus at gcc dot gnu.org
CC: jrfsousa at gmail dot com
Target Milestone: ---
Created attachment 47987
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=47987&action=edit
assumed_rank_19_aux.c – to be used with …
As found by José and reported at
https://gcc.gnu.org/ml/fortran/2020-03/msg00016.html
Testing shows that assumed-rank arrays are mishandled in several ways
* When required by the standard, the lower bound is not always shifted to "1"
– this causes misreported bounds.
* Likewise, for BIND(C), the bounds are not always the expected 0.
* When allocatable/pointer, the argument cannot be assumed-rank; hence,
"-1" is a perfectly valid upper and lower bound and size()/shape() shall
report >= 0 and not a negative value!
Likewise for dim[rank-1].extent with bind(C): the value shall not be
negative.
See attached test case. All code which works uses STOP and assert(). All code
which doesn't uses (__builtin_)print(f) diagnostic.
Caveat: I hope I haven't made any mistake when writing the test code.
* * *
Compile with:
gfortran assumed_rank_19.f90 assumed_rank_19_aux.c
When using DejaGNU/before committing: use the commented ISO_Fortran_binding.h
#include as otherwise the testsuite may fail to find that file.
