On Thu, Jan 15, 2026 at 12:06:10PM +0800, Jason Merrill wrote:
> On 1/15/26 4:20 AM, Marek Polacek wrote:
> > On Wed, Jan 14, 2026 at 03:11:26PM -0500, Marek Polacek wrote:
> > > Another month of work, so it's time for v3 of the Reflection patch
> > > series. Since v2, we checked in over 75 commits, mostly cleanups
> > > and bug fixes for reflect.cc.
> > >
> > > The whole branch can be found at
> > > <https://forge.sourceware.org/marek/gcc/commits/branch/reflection>
> > >
> > > I will post v2/v3 diff as well.
>
> A few small comments below.
>
> Note that I'm not going to repeat comments from my last reply to v2 3/9,
> please also keep them in mind.
>
> At this point I think we should go ahead and merge, subject to release
> manager approval; the remaining minor issues can be resolved incrementally.
Thanks again!
> > --- a/gcc/cp/mangle.cc
> > +++ b/gcc/cp/mangle.cc
> > @@ -4183,7 +4183,10 @@ write_reflection (tree refl)
> > write_type (arg);
> > else if (strcmp (prefix, "dm") == 0)
> > {
> > - write_prefix (decl_mangling_context (arg));
> > + tree ctx = decl_mangling_context (arg);
> > + while (ctx && ANON_UNION_TYPE_P (ctx))
> > + ctx = decl_mangling_context (TYPE_NAME (ctx));
>
> Please add a FIXME referring to the ABI discussion.
Not done as per the other email.
> > @@ -133,6 +143,14 @@ get_reflection (location_t loc, tree t, reflect_kind
> > kind/*=REFLECT_UNDEF*/)
> > "parameter %qD", t);
> > return error_mark_node;
> > }
> > + /* If the id-expression denotes a variable declared by an init-capture,
> > + R is ill-formed. */
> > + else if (is_capture_proxy (t))
> > + {
> > + error_at (loc, "%<^^%> cannot be applied to a local entity declared "
> > + "by init-capture");
>
> This diagnostic needs adjustment now that the condition includes all
> captures.
Patch posted:
<https://gcc.gnu.org/pipermail/gcc-patches/2026-January/706070.html>
> > +/* For direct base class relationship R return the binfo related
> > + to the derived type. */
> > +
> > +static tree
> > +direct_base_parent_binfo (tree r)
> > +{
> > + /* Looping needed for multiple virtual inheritance. */
> > + while (BINFO_INHERITANCE_CHAIN (r))
> > + r = BINFO_INHERITANCE_CHAIN (r);
> > + return r;
> > +}
> > +
> > +/* For direct base class relationship R return the derived type
> > + (i.e. when R is (D, B) it returns D). */
>
> Let's use "derived" instead of "parent" in the name of this function; the
> latter is specific to parent_of semantics.
Patch posted:
<https://gcc.gnu.org/pipermail/gcc-patches/2026-January/706071.html>
> > @@ -3064,6 +3035,8 @@ eval_variable_of (location_t loc, const constexpr_ctx
> > *ctx, tree r,
> > tree fun)
> > {
> > if (eval_is_function_parameter (r, kind) == boolean_false_node
> > + /* This doesn't consider the points corresponding to injected
> > + declarations. */
>
> This reads like a fixme, please add that it doesn't seem needed.
Done in r16-6849-g00f7efe1e5cc30.
> > @@ -8552,24 +8373,32 @@ check_splice_expr (location_t loc, location_t
> > start_loc, tree t,
> > foo.[: ^^S::bar :]. */
> > if (!address_p
> > && !member_access_p
> > - && ((DECL_P (t) && DECL_NONSTATIC_MEMBER_P (t))
> > - || (VAR_P (t) && DECL_ANON_UNION_VAR_P (t))))
> > + && DECL_P (t)
> > + && DECL_NONSTATIC_MEMBER_P (t))
> > {
> > if (complain_p)
> > - error_at (loc, "cannot implicitly reference a class member through "
> > - "a splice");
> > + error_at (loc, "cannot implicitly reference a class member %qD "
> > + "through a splice", t);
> > return false;
> > }
> > /* [expr.unary.op]/3.1 "If the operand [of unary &] is a qualified-id or
> > splice-expression designating a non-static member m, other than an
> > explicit object member function, m shall be a direct member of some
> > class C that is not an anonymous union." */
> > - if (address_p && VAR_P (t) && DECL_ANON_UNION_VAR_P (t))
> > - {
> > - if (complain_p)
> > - error_at (loc, "unary %<&%> applied to an anonymous union member %qD "
> > - "that is not a direct member of a named class", t);
> > - return false;
> > + if (address_p
> > + && TREE_CODE (t) == FIELD_DECL
> > + && ANON_UNION_TYPE_P (DECL_CONTEXT (t)))
> > + {
> > + tree c = CP_TYPE_CONTEXT (DECL_CONTEXT (t));
> > + while (ANON_UNION_TYPE_P (c))
> > + c = CP_TYPE_CONTEXT (c);
>
> This could be context_for_name_lookup (t)
Patch posted:
<https://gcc.gnu.org/pipermail/gcc-patches/2026-January/706072.html>
I'll continue addressing your other comments next week.
Marek
