| Message ID | 20220323125604.3296543-1-jason@redhat.com |
|---|---|
| State | New |
| Headers | show |
| Series | [pushed] c++: using from enclosing class template [PR105006] | expand |
On Wed, 23 Mar 2022, Jason Merrill via Gcc-patches wrote: > Here, DECL_DEPENDENT_P was false for the second using because Row<eT> is > "the current instantiation", so lookup succeeds. But since Row itself has a > dependent using-decl for operator(), the set of functions imported by the > second using is dependent, so we should set the flag. > > Tested x86_64-pc-linux-gnu, applying to trunk. > > PR c++/105006 > > gcc/cp/ChangeLog: > > * name-lookup.cc (lookup_using_decl): Set DECL_DEPENDENT_P if lookup > finds a dependent using. > > gcc/testsuite/ChangeLog: > > * g++.dg/template/using30.C: New test. > --- > gcc/cp/name-lookup.cc | 15 +++++++++++++++ > gcc/testsuite/g++.dg/template/using30.C | 13 +++++++++++++ > 2 files changed, 28 insertions(+) > create mode 100644 gcc/testsuite/g++.dg/template/using30.C > > diff --git a/gcc/cp/name-lookup.cc b/gcc/cp/name-lookup.cc > index 323f96bcd24..ea947fabb7e 100644 > --- a/gcc/cp/name-lookup.cc > +++ b/gcc/cp/name-lookup.cc > @@ -5665,6 +5665,21 @@ lookup_using_decl (tree scope, name_lookup &lookup) > lookup.value = lookup_member (binfo, lookup.name, /*protect=*/2, > /*want_type=*/false, tf_none); > > + /* If the lookup in the base contains a dependent using, this > + using is also dependent. */ > + if (!dependent_p && lookup.value) I wonder if it'd be worthwhile to also test dependent_type_p (scope) here here to avoid iterating over the lookup set when it can't possibly contain a dependent using-decl. > + { > + tree val = lookup.value; > + if (tree fns = maybe_get_fns (val)) > + val = fns; > + for (tree f: lkp_range (val)) > + if (TREE_CODE (f) == USING_DECL && DECL_DEPENDENT_P (f)) > + { > + dependent_p = true; > + break; > + } > + } > + > if (!depscope && b_kind < bk_proper_base) > { > if (cxx_dialect >= cxx20 && lookup.value > diff --git a/gcc/testsuite/g++.dg/template/using30.C b/gcc/testsuite/g++.dg/template/using30.C > new file mode 100644 > index 00000000000..914252dd14c > --- /dev/null > +++ b/gcc/testsuite/g++.dg/template/using30.C > @@ -0,0 +1,13 @@ > +// PR c++/105006 > + > +template<class eT> > +class Row { > + using eT::operator(); > + void operator()(); > + class fixed; > +}; > + > +template<class eT> > +class Row<eT>::fixed : Row { > + using Row::operator(); > +}; > > base-commit: 4a9e92164a547afcf8cd3fc593c7660238ad2d59 > -- > 2.27.0 > >
On 3/23/22 10:29, Patrick Palka wrote: > I wonder if it'd be worthwhile to also test dependent_type_p (scope) here > here to avoid iterating over the lookup set when it can't possibly contain > a dependent using-decl. Good thought:
diff --git a/gcc/cp/name-lookup.cc b/gcc/cp/name-lookup.cc index 323f96bcd24..ea947fabb7e 100644 --- a/gcc/cp/name-lookup.cc +++ b/gcc/cp/name-lookup.cc @@ -5665,6 +5665,21 @@ lookup_using_decl (tree scope, name_lookup &lookup) lookup.value = lookup_member (binfo, lookup.name, /*protect=*/2, /*want_type=*/false, tf_none); + /* If the lookup in the base contains a dependent using, this + using is also dependent. */ + if (!dependent_p && lookup.value) + { + tree val = lookup.value; + if (tree fns = maybe_get_fns (val)) + val = fns; + for (tree f: lkp_range (val)) + if (TREE_CODE (f) == USING_DECL && DECL_DEPENDENT_P (f)) + { + dependent_p = true; + break; + } + } + if (!depscope && b_kind < bk_proper_base) { if (cxx_dialect >= cxx20 && lookup.value diff --git a/gcc/testsuite/g++.dg/template/using30.C b/gcc/testsuite/g++.dg/template/using30.C new file mode 100644 index 00000000000..914252dd14c --- /dev/null +++ b/gcc/testsuite/g++.dg/template/using30.C @@ -0,0 +1,13 @@ +// PR c++/105006 + +template<class eT> +class Row { + using eT::operator(); + void operator()(); + class fixed; +}; + +template<class eT> +class Row<eT>::fixed : Row { + using Row::operator(); +};