Am 05.01.26 um 10:06 PM schrieb Jerry D:
On 1/5/26 12:32 PM, Harald Anlauf wrote:
Dear All,
when a type-spec was provided with a character array constructor,
it seemed to get "lost" when passing the array constructor to
a procedure, even to the LEN() intrinsic.
The attached patch fixes this, enables simplification of LEN()
when passed an array constructor with type-spec, and also adds
the type-spec to the fortran dump.
Regtested on x86_64-pc-linux-gnu. OK for mainline?
Thanks,
Harald
Hi Harald,
I just finished testing this. Looks OK to go.
I did not see anything obviously wrong with this.
Commit at your discretion.
Jerry
Hi Jerry,
thanks for checking. Unfortunately, I found a last-minute issue...
The following code was previously miscompiled and is now rejected:
program p
integer :: n = 2
call foo ([character(n) :: 'abc'])
contains
subroutine foo (x)
character(*), intent(in) :: x(:)
print *, len(x)
end subroutine foo
end
gcc-15 prints 3 instead of 2; gcc-16 with my patch does:
pr47425-xx.f90:3:36:
3 | call foo ([character(n) :: 'abc'])
| ^
Error: size of variable 'A.5' is too large
This is triggered by the hunk from trans-array.cc.
I have to think a little bit longer how to proceed...
