On Wed, Aug 26, 2026 at 6:57 PM Jakub Jelinek <[email protected]> wrote:
> On Wed, Aug 26, 2026 at 11:50:37AM +0200, Tomasz Kaminski wrote:
> > >
> > Any reason not to check &c.c (array) itself>
>
> Added in my copy, incremental
> --- a/gcc/testsuite/g++.dg/cpp26/within-lifetime1.C 2026-08-25
> 16:58:17.000000000 +0200
> +++ b/gcc/testsuite/g++.dg/cpp26/within-lifetime1.C 2026-08-26
> 18:42:14.645233986 +0200
> @@ -335,6 +335,8 @@ foo ()
> return 141;
> if (!std::is_within_lifetime (t + 1))
> return 142;
> + if (!std::is_within_lifetime (&c.c))
> + return 143;
> return 0;
> }
>
> added at the end as I don't want to renumber everything.
> Maybe I should have used
> return __LINE__;
> in all cases. Changed that now too.
>
> > > + {
> > > + int c = 42;
> > > + if (x == 1)
> > > + __builtin_is_within_lifetime (a); // { dg-error
> > > "'__builtin_within_lifetime' on 'c' outside its lifetime is not a
> constant
> > > expression" }
> > >
> > That's message should very confusing, '__builtin_within_lifetime' on 'c'
> > outside its lifetime is not a constant expression", while
> > the purpose of this function is to check if they are within the lifetime.
> > Maybe we should say outside it's storage duration?
>
> But it is really about lifetime of the complete object.
> Perhaps I should add the "complete object " to the message to make it
> clear?
>
That would certainly help.
>
> > > +
> > > +struct A { int a = 42; bool b = std::is_within_lifetime (&a); };
> //
> > > { dg-error "" "" { xfail *-*-* } }
> > >
> > What are the errors here and for next example? The member lifetime seems
> > clear at corresponding points.
>
> There are no errors, that is the problem.
> clang++ reports on https://godbolt.org/z/dKM9q9163
> <source>:4:13: error: constexpr variable 'n' must be initialized by a
> constant expression
> <source>:3:33: note: 'std::is_within_lifetime' cannot be called with a
> pointer to an object whose lifetime has not yet begun
> ...
> <source>:7:13: error: constexpr variable 'o' must be initialized by a
> constant expression
> <source>:6:40: note: 'std::is_within_lifetime' cannot be called with a
> pointer to an object whose lifetime has not yet begun
> ...
> <source>:9:70: error: call to consteval function 'C::C' is not a constant
> expression
> <source>:9:29: note: '__builtin_is_within_lifetime' cannot be called with
> a pointer to an object whose lifetime has not yet begun
> 9 | struct C { consteval C () { __builtin_is_within_lifetime (this); }
> } p; // { dg-error "" "" { xfail *-*-* } }
> |
> ^
>
> In all the 3 cases the complete object lifetime has not started yet
> (because
> it is still being constructed).
>
> > > +constexpr A n;
> > >
> > I.e. a is in lifetime here.
>
> But n is not yet in lifetime.
>
> > > +
> > > +struct B { consteval B () : a (42), b (std::is_within_lifetime (&a))
> {}
> > > int a; bool b; };
> > >
> > And not in lifetime here.
>
> If the complete object would be in lifetime (see the testcase in the
> description rather than patch), I think a would be in lifetime here, but
> not b itself nor some nsdm initialized in a later mem-initializer.
>
> > > + static_assert (&corge (0));
> > > + static_assert (&corge (A {}));
>
> I have actually meant __builtin_is_within_lifetime calls in the
> static_asserts too, added in my copy.
>
> Jakub
>
>
