https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62270
--- Comment #3 from Tobias Burnus <burnus at gcc dot gnu.org> ---
(In reply to Marek Polacek from comment #0)
> -Wlogical-not-parentheses detected these two issues:
I really missed such a warning before; Coverity also detected such issues, but
it has way too much output to dig through all.
> interface.c:compare_parameter
> 2013 /* F2008, 12.5.2.5; IR F08/0073. */
> 2014 if (formal->ts.type == BT_CLASS && formal->attr.class_ok
> 2015 && actual->expr_type != EXPR_NULL
> 2016 && ((CLASS_DATA (formal)->attr.class_pointer
> 2017 && !formal->attr.intent == INTENT_IN)
Here, it really should be: Print an error when passing a nonpolymophic actual
argument to a polymorphic dummy/formal argument - to avoid modification of the
run-time type. But as that's not possible with INTENT_IN, one can permit it in
that case.
Thus, the obvious patch should be correct:
--- a/gcc/fortran/interface.c
+++ b/gcc/fortran/interface.c
@@ -2016,3 +2016,3 @@ compare_parameter (gfc_symbol *formal, gfc_expr *actual,
&& ((CLASS_DATA (formal)->attr.class_pointer
- && !formal->attr.intent == INTENT_IN)
+ && formal->attr.intent != INTENT_IN)
|| CLASS_DATA (formal)->attr.allocatable))
> trans-expr.c:gfc_conv_procedure_call
> 4445 if (fsym->attr.optional
> 4446 && e->expr_type == EXPR_VARIABLE
> 4447 && (!e->ref
> 4448 || (e->ref->type == REF_ARRAY
> 4449 && !e->ref->u.ar.type != AR_FULL))
In the surrounding code block, one inserts a deallocation call to the data
component of a array descriptor ("free (var->data);") - but if the argument can
be absent ("optional" attribute), one has var == NULL. Thus, one needs to add
an extra check before dereferencing. But that is only needed if the users don't
add array/subcomponent references themselves (e.g."var(:)%subcomp").
Thus, the check is only needed if e->ref == NULL - or one has an (automaically
added) reference to the full array. Again, the obvious patch should be correct:
--- a/gcc/fortran/trans-expr.c
+++ b/gcc/fortran/trans-expr.c
@@ -4592 +4592 @@ gfc_conv_procedure_call (gfc_se * se, gfc_symbol * sym,
- && !e->ref->u.ar.type != AR_FULL))
+ && e->ref->u.ar.type != AR_FULL))
Side remark: I wonder whether BT_CLASS needs some extra care.
> The first one should likely be "&& formal->attr.intent != INTENT_IN)", but
> that regressed gfortran.dg/pointer_intent_7.f90 test.
The problem is that one has two different issues - and now one earlier check
triggers. For
call bar3p (b)
there are two related issues: One passes a pointer-intent(in) variable to a
pointer-intent(inout) dummy-argument variable. And passes a nonpolymorphic
pointer to a polymorphic dummy-argument variable w/o intent(in).
Seemingly, the second issue is checked before in the code - and didn't trigger.
(Which was a bug.) Thus, simply change the dg-error message.
[Depending what the user intended, one or the other message is more helpful -
hence, it doesn't matter which one is shown.]
I won't be able to package and test this code before this evening; feel free to
package, regtest and submit it yourself.
(In reply to Thomas Koenig from comment #2)
> A longer version would be [...] Is this what is intended here?
Well, the code is bogus - for obvious reasons. The problem was more what is
correct and makes code wise sense - re-writing the existing condition doesn't
help.
