On 09/07/2012 05:07 PM, Paolo Carlini wrote:
the regression of template/crash83.C because of excess errors. The latter happens
because we have code in cp_parser_class_head which tries to improve error recovery
for cases of missing 'template <>' in explicit specializations, and, for that
rather broken testcase, it confuses the code I'm tentatively adding, resulting in the
additional error message below:
crash83.C:5:27: error: an explicit specialization must be preceded by ‘template
<>’
template<typename = class A<0>: > struct B {}; // { dg-error "explicit
specialization|expected" }
^
crash83.C:5:27: error: too many template-parameter-lists for A<0> (should be 1)
So we're adding a second irrelevant diagnostic to go with the first one.
The error in this testcase is a simple stray ':' causing a syntax
error, it has nothing to do with explicit specialization. It would be
nice to improve this, but given that we're already not giving the right
error, adding a second wrong error doesn't seem like a big problem.
More importantly, I'm having troubles removing the early
cp_parser_check_template_parameters check in cp_parser_class_head: one
issue is that it also checks for too few template-parameter-lists not
just too many, and thus catches rather broken testcases like, eg,
template/error7.C or template/class2.C:
template <class T>
struct A {
struct B;
struct A::B { }; // { dg-error "" }
};
This testcase is indeed ill-formed, but its ill-formedness has nothing
to do with too few template-parameter-lists. The problem is that clause
9 says
If a class-head-name contains a nested-name-specifier, the
class-specifier shall refer to a class that was previously declared
directly in the class or namespace to which the nested-name-specifier
refers, ... and the class-specifier shall appear in a namespace
enclosing the previous declaration.
The definition of A::B does not appear at namespace scope, so it is
ill-formed. Removing the incorrect error is an improvement, but then we
need to add the correct error rather than accept the testcase.
but there are definitely more problems with just checking
template_header_count < wanted too in
maybe_process_partial_specialization and removing the early check, I'm
seeing many, many fails if I just try to extend "p4" like that...
Like what?
Jason
