Re: (Re)allocation of allocatable arrays on assignment - F2003

[Tobias Burnus <burnus at net-b dot de>]

Dear Paul,

thanks for the patch.

On 11/04/2010 07:18 AM, Paul Richard Thomas wrote:
> In addition to correcting bugs, the main addition to the patch is the
> option -f(no-)realloc_lhs.  Default behaviour for -std>  f95 is to
> reallocate on assignment.

I think a '-' instead of a '_' is more in line with most GCC and 
gfortran options; essentially only Darwin and SH use some '_' in the 
flags. While most of the patch uses realloc_lhs, there is at least one 
place where realloc-lhs is used. (As Intel uses "-assume realloc_lhs", 
the '_' has also some merits.) Thus, please make the use of '-' vs. '_' 
consistent and consider using a hyphen instead of an underscore.

Additionally, the gfortran manpage (as the gcc one) states:

       In some cases, options have positive and negative forms; the 
negative
       form of -ffoo would be -fno-foo.  This manual documents only one of
       these two forms, whichever one is not the default.

As -std=gnu is the default, which implies -frealloc-lhs/-frealloc_lhs, 
we should document
-fno-realloc-lhs/-fno-realloc_lhs instead. Sorry for missing this when I 
wrote the patch.

> Please note that allocatable assignments of the kind:
>         x = transformational_array_intrinisic (...)
> now produce a temporary unless -fno-realloc_lhs or -std=f95 are
> deployed.  This could be fixed relatively easily by detecting the
> option in the library and repeating the logic of
> gfc_alloc_allocatable_for_assignment there.

I think that would be useful - also for cases such as
  a = abs(a)
where the RHS is a elemental* function as one knows that the LHS and RHS 
must have the same shape. However, such a patch can be deferred.

> Bootstrapped and regtested on FC9/x86_64 - OK for trunk?

The patch looks OK.

@@ -3676,6 +3700,11 @@ gfc_conv_loop_setup (gfc_loopinfo * loop, locus * 
where)
              continue;
            }

+         /* Avoid using an allocatable lhs in an assignment, since
+            there might be a reallocation coming.  */
+         if (loopspec[n] && ss->is_alloc_lhs)
+           continue;
+

I was wondering whether one should also add a 
"gfc_option.flag_realloc_lhs &&" to the if.


+}
+
+/* Allocate the lhs of an assignment to an allocatable array, otherwise

Add another empty line before the comment.

Tobias