Bruno Haible wrote:
> I'm also having similar warnings in the *list modules and in GNU clisp.
>
>> Since I require warning free compilation with -Wuninitialized,
>> I am considering this patch. Would you prefer to avoid the
>> warning in a different manner?
>
> Well, I would prefer if the inaccuracy of the warnings be acknowledged by
> the GCC developers. Has it already been reported?
I don't know.
This "new" warning is being exposed because gcc's use analysis
can now span function boundaries.
> We already have an IF_LINT macro in this file. Let's extend it so that it can
> be used in this case as well. Users GCC < 3.0 will now see more warnings, but
> they can upgrade to a newer GCC anyway. This avoids to have #if inside
> functions for such a trivial stuff. Applying this:
Please do not push that, at least not with my name on it...
> 2009-11-21 Jim Meyering <meyer...@redhat.com>
> Bruno Haible <br...@clisp.org>
>
> diffseq: avoid spurious gcc warnings
> * lib/diffseq.h (IF_LINT): Enable only with GCC >= 3.0.
> (compareseq): Initialize two members of "part" to avoid used-
> uninitialized warnings.
>
> --- lib/diffseq.h.orig 2009-11-21 18:39:58.000000000 +0100
> +++ lib/diffseq.h 2009-11-21 18:34:58.000000000 +0100
> @@ -68,9 +68,10 @@
> # define EARLY_ABORT(ctxt) false
> #endif
>
> -/* Use this to suppress gcc's `...may be used before initialized' warnings.
> */
> +/* Use this to suppress gcc's `...may be used before initialized' warnings.
> + The Code argument may contain syntax that assumes GCC 3.0 or newer. */
> #ifndef IF_LINT
> -# ifdef lint
> +# if defined lint && __GNUC__ >= 3
> # define IF_LINT(Code) Code
> # else
> # define IF_LINT(Code) /* empty */
> @@ -464,7 +465,7 @@
> }
> else
> {
> - struct partition part;
> + struct partition part IF_LINT (= { .xmid = 0, .ymid = 0 });
Any code using a one-argument macro like that is rejected,
due to the comma:
../lib/diffseq.h:467:64: error: macro "IF_LINT" passed 2 arguments,\
but takes just 1
FYI, I tried the same thing before writing the #if-adding patch.
