Re: [Patch, fortran] PR30236 - alternate-return subroutine in generic interface causes ice/segfault

[Brooks Moses <brooks dot moses at codesourcery dot com>]

Paul Thomas wrote:
> *** gcc/testsuite/gfortran.dg/generic_11.f90	(revision 0)
> --- gcc/testsuite/gfortran.dg/generic_11.f90	(revision 0)
> ***************
> *** 0 ****
> --- 1,39 ----
> + ! { dg-do compile }
[...]
> + program test
> +   use arswitch
> +   implicit none
> +   integer :: i = 0
> +  10 continue
> +   call with (i, *20)
> +   print *, " looping with i = ", i
> +   i = 2
> +   goto 10
> +  20 print *, " leaving with i = ", i
> + end

Looking at this, it seems like it might be a good idea for the testcase 
not to risk infinite looping as a failure mode.  Maybe something like this?

(Also, you're still calling the specific "with" name, not the generic 
"gen" name, so this isn't testing the disambiguation; I changed it to 
call the generic name.)

program test
  use arswitch
  implicit none
  integer :: i = 0
  call gen (i, *10)
  i = 2
  call gen (i, *20)
 10 continue
  call abort()
 20 continue
end

Though, actually, I just noticed that that's not a "dg-do run" test, 
even though the main program is rather longer than you'd need just for 
compile testing.  However, it's probably useful to have a test for 
alternate returns that's run -- so far, the only one I know of 
(altreturn_1.f90) is compile-only -- so changing this to dg-do run might 
be a good idea.  For that matter, calling this one altreturn_2.f90 might 
also be a little better than generic_11.f90, since there are a lot fewer 
of the former.


Meanwhile, as an unofficial review -- the rest of the patch looks good 
to me, and matches the sort of thing I was finding to be the cause of PR 
30235.  With the above notes, I'd give the whole thing an OK.

- Brooks


P.S. let me know if you do call this one altreturn_2, and I'll rename my 
pending-review one to altreturn_3.  Or you can name yours altreturn_3. 
Or something.  :)