Hi Steve,
That certainly does the trick! OK to commit as far back as you have
the intestinal fortitude for.
Thanks
Paul
On Tue, 6 Aug 2019 at 19:24, Steve Kargl
<s...@troutmask.apl.washington.edu> wrote:
>
> The spaghetti code in PR fortran/91359 has a few gotos
> to jump to different places. In doing so, gfortran
> with generate a function that has a naked 'return'.
> Namely,
>
> foo ()
> {
> logical(kind=4) __result_foo;
>
> goto __label_000002;
> __label_000001:;
> return; /* <-- THIS IS BAD. */
> __label_000002:;
> __result_foo = 0;
> if (!__result_foo) goto __label_000001;
> L.1:;
> return __result_foo;
> return __result_foo;
> }
>
> The attached patch avoids the naked 'return' by
> adding the variable that is constructed from the
> the function name. Namely,
>
> foo ()
> {
> logical(kind=4) __result_foo;
>
> goto __label_000002;
> __label_000001:;
> return __result_foo; /* <-- THIS IS GOOD. */
> __label_000002:;
> __result_foo = 0;
> if (!__result_foo) goto __label_000001;
> L.1:;
> return __result_foo;
> return __result_foo;
> }
>
> Regression tested on x86_64-*-freebsd. OK to commit?
>
> 2019-08-06 Steven G. Kargl <ka...@gcc.gnu.org>
>
> PR fortran/91359
> * trans-decl.c (gfc_generate_return): Ensure something is returned
> from a function.
>
> 2019-08-06 Steven G. Kargl <ka...@gcc.gnu.org>
>
> PR fortran/91359
> * gfortran.dg/pr91359_1.f: New test.
> * gfortran.dg/pr91359_2.f: Ditto.
>
> --
> Steve
--
"If you can't explain it simply, you don't understand it well enough"
- Albert Einstein
