On 2/13/19 12:36 PM, Alexandre Oliva wrote:
On Feb 8, 2019, Jason Merrill <ja...@redhat.com> wrote:
On 2/8/19 4:07 AM, Alexandre Oliva wrote:
On Feb 7, 2019, Jason Merrill <ja...@redhat.com> wrote:
In protected_accessible_p and shared_member_p, if we're left with a
USING_DECL after strip_using_decl, we can't give a meaningful answer,
and should probably abort; we shouldn't get here with a dependent
expression.
In g++.dg/lookup/using39.C, shared_member_p is called by
cp_parser_primary_expression -> finish_id_expression ->
finish_id_expression_1 -> finish_qualified_id_expr when parsing 'using
A<T>::f' in B<T>.
Then I think finish_qualified_id_expr should check
type_dependent_expression_p first.
That works, as long as we take dependent as non-shared, which, no
surprise, is what the previous code would accidentally do. Doing
otherwise regresses cpp1y/lambda-generic-this1a.C, and I have test
results that prove it ;-)
(I wonder if that won't malfunction if the name ends up resolving to a
static member, or even to a type... Shall I give that a try myself, or
do you know we have tests in place for that already?)
Anyway, here's the patch that passed regstrapping without regressions on
x86_64- and i686-linux-gnu. Ok to install?
OK.
Jason
