PR middle-end/179 gcc -O2 -Wuninitialized missing warning with &var

[Manuel López-Ibáñez]

2008/8/20 Richard Guenther <richard.guenther@gmail.com>:
> On Wed, Aug 20, 2008 at 11:51 AM, Manuel López-Ibáñez
> <lopezibanez@gmail.com> wrote:
>> A different approach.
>>
>> Bootstrapped and regression tested on x86_64-unknown-linux-gnu with
>> --enable-languages=all,ada
>>
>> I am not sure if there is a PR open for the XFAIL in uninit-6.c
>>
>> OK for trunk?
>
> This looks much better!  A few nits:
>
> +  /* We do not care about LHS.  */
> +  if (wi->is_lhs)
> +    return NULL_TREE;
> +
>
> I do not see this is_lhs member set or defined anywhere...  the test
> also looks superfluous.

If I understand the description of struct walk_stmt_info in gimple.h:

  /* True if we are currently walking the LHS of an assignment.  */
  bool is_lhs;

And walk_tree updates it automatically. See count_ptr_derefs in
tree-ssa-alias.c for example of usage.

> +       /* If there is not gimple stmt, then we cannot check VUSE ops.  */
> +       if (data->stmt == NULL)
> +         return NULL_TREE;
>
> In addition I would add
>
>            || !gimple_aliases_computed_p (cfun)
>            || !ZERO_SSA_OPERANDS (data->stmt, SSA_OP_VIRTUAL_DEFS))
>
> which should catch the wi->is_lhs case as well as calls and struct
> copies.

I can test that. Could you give me an explanation of why we want to
test the second? I do not get it and I would like to add a comment
explaining it.

Cheers,

Manuel.