Re: [Patch, fortran] PR78293 - - [5/6/7 Regression] SIGABRT with function result used as argument

[Andre Vehreschild <vehre at gmx dot de>]

Hi Thomas,

> Andre put me to shame with a devastatingly simple replacement for a
> horribly complicated and wrong patch that I was getting into.

I did not mean to. I happened to work in the same area and the PR's description
rang a whole chorus of bells what might have been going wrong.

> The part of the patch in trans-expr.c fixes the PR and the part in
> trans-stmt.c fixes a memory leak in function 'tt'. This latter fixes

I propose to change the comment in the second chunk (trans-stmt.c) from:

-         /* Deallocate any allocatable components after all the allocations
-            and assignments of expr3 have been completed.  */

to

+ 	/* Deallocate any allocatable components in expressions that use a
+          temporary, i.e. are not of expr-type EXPR_VARIABLE or force the
+          use of a temporary, after the assignment of expr3 is completed.  */

Mind that the indentation is corrupted by my mailer. Please polish the comment
a bit. It feels awkward.

> half of the memory leaks in class_array_15.f03. I have noted the rest
> of this problem in PR38319 with which it is associated.
> 
> Bootstraps and regtests on FC21/x86_64 - OK for trunk and, later
> 5-branch and 6-branch?

Ok'ing a patch I participated in writing does not feel correct. So not doing it.
When committing please add yourself also to the Changelog in gcc/fortran to
prevent confusion when this patch causes regressions. You are as much
responsible for fixing it as me.

Regards,
	Andre

> Cheers
> 
> Paul
> 
> 2016-11-24  Andre Vehreschild  <vehre@gcc.gnu.org>
> 
>     PR fortran/78293
>     * trans-expr.c (gfc_conv_procedure_call): Prepend deallocation
>     of alloctable components to post, rather than adding to
>     se->post.
>     * trans-stmt.c (gfc_trans_allocate): Move deallocation of expr3
>     allocatable components so that all expr3s are visited.
> 
> 2016-11-24  Paul Thomas  <pault@gcc.gnu.org>
> 
>     PR fortran/78293
>     * gfortran.dg/allocatable_function_10.f90: New test.
>     * gfortran.dg/class_array_15.f03: Increase builtin_free count
>     from 11 to 12.


-- 
Andre Vehreschild * Email: vehre ad gmx dot de