Re: [patch fortran]: Fix warnings in resolve.c

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

On 06/08/2010 11:10 AM, Kai Tietz wrote:
> this patch fixes two warnings about "might be used uninitialized" 
> warnings.


> +++ gcc/gcc/fortran/resolve.c   2010-06-08 11:06:40.458536300 +0200
> @@ -6002,7 +6002,7 @@ resolve_deallocate_expr (gfc_expr *e)
> -  gfc_symbol *sym;
> +  gfc_symbol *sym = NULL;

I do not understand this warning. The source is as follows:

------------- < cut > ----------------
  gfc_symbol *sym;
  gfc_component *c;

  /* Check INTENT(IN), unless the object is a sub-component of a
pointer.  */
  check_intent_in = 1;

  if (gfc_resolve_expr (e) == FAILURE)
    return FAILURE;

  if (e->expr_type != EXPR_VARIABLE)
    goto bad;

  sym = e->symtree->n.sym;
  [...]
------------- </cut > ----------------

Thus, I do not see why the compiler warns. Any idea?



> @@ -6196,7 +6196,7 @@ resolve_allocate_expr (gfc_expr *e, gfc_
> -  gfc_symbol *sym;
> +  gfc_symbol *sym = NULL;

Here, I can understand the warning:

  if (e->symtree)
    sym = e->symtree->n.sym;

  if (e->expr_type != EXPR_VARIABLE)
[...]
  else
    {
      if (sym->ts.type == BT_CLASS)

as the compiler cannot know that the variable is indeed non-initialized.
However, I thought that file is build with -Werror thus I am surprised
that we do not get build failures. Namely, it is build here with:

-W -Wall -Wwrite-strings -Wcast-qual -Wstrict-prototypes
-Wmissing-prototypes -Wmissing-format-attribute -pedantic -Wno-long-long
-Wno-variadic-macros -Wno-overlength-strings -Werror -Wold-
style-definition -Wc++-compat

and -Wall turns on -Wuninitialized. However, I do not see a warning in
the logs.

Thus, can you dig a bit more (a) why you see the warning while I don't.
(b) why -Werror does not stop the build? If (a) is solved, the next
question is why it warns for sym in resolve_deallocate_expr.

Tobias