https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104972
Bug ID: 104972
Summary: Class dummy argument for array of custom types stuck
on -fcheck=bounds
Product: gcc
Version: 11.2.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: fortran
Assignee: unassigned at gcc dot gnu.org
Reporter: albandil at atlas dot cz
Target Milestone: ---
When the following program is compiled with `-fcheck=bounds` using a recent
version of gfortran, it triggers a bogus (?) error:
program test
type vec
integer :: x(3)
end type
type(vec) :: v(2)
call sub(v)
contains
subroutine sub (v)
class(vec), intent(in) :: v(:)
integer :: k, q(3)
q = [ (v(1)%x(k), k = 1, 3) ] ! <-- fails here
end subroutine
end program
The error message is
At line 19 of file test.f90
Fortran runtime error: Index '3' of dimension 1 of array 'v%_data%x' above
upper bound of 2
When the lengths of the arrays `x` and `v` in the code are modified, the
message changes as well, but it always appears as if the compiler interchanged
the dimension of `v` and of `x`. I observe this behaviour with the compiler
version on openSUSE Tumbleweeed 20220316
$ gfortran --version
GNU Fortran (SUSE Linux) 11.2.1 20220103 [revision
d4a1d3c4b377f1d4acb34fe1b55b5088a3f293f6]
I also got the same run-time errors when pulling and compiling a recent Git
version of GCC,
$ gfortran --version
GNU Fortran (GCC) 12.0.1 20220310 (experimental)
However, in older versions of the compiler (8, 9 and 10) using the
`-fcheck=bounds` flag with this program works fine. Provided that the code is
syntactically valid as it is, this suggests a compiler regression.
The error goes away also when I replace the keyword `class` with `type`.
