Even though id-expressions is introduced inside [expr], we use it in situations where it is not actually an expression. For example, for declarator-id in [dcl.decl.general].

I agree that id-expression should be moved outside of [expr], and only the parts relevent for [expr] should stay there (value category, result).

For example, for declarator-id in [dcl.decl.general].

Yes, since the declarator-id is an id-expression in grammar, which is reasonable that name lookup can interpret the name as an id-expression that is able to appear in an expression when that name appears in an expression. That is why I said id-expression may limit the template parameter as a non-type template parameter because that parameter's declaration has a declarator-id that is an id-expression. However, any type template parameter is introduced by an identifier, the lookup rules apply to the name shouldn't interpret that name as an id-expression, as we discussed in #5345 (comment) and #5341 (comment)

Anyway, typedef-name and template-name are not comprised by the grammar of an id-expression.

For template argument corresponding to template template-parameter, we have [temp.arg.template] p1 that specifies that the template-argument is expressed as id-expression. So, it could be arguably correct that we say AA is an id-expression when PP is a template template-parameter

However, we do not have a similar special phrase that specifies the type template-argument is expressed as an id-expression. [temp.arg.type] p1 says such a template-argument shall be a type-id.

A template-argument for a template-parameter which is a type shall be a type-id.

type-id is not an id-expression and vice versa.

BTW, [temp.arg.template] p1 requires that the template-argument for template template-paramter shall be the name of a class template or an alias template. It is unclear whether a name introduced by a template template-parameter can be considered to be a name of a class template or of an alias template or not? We do not have a relevant phrase.

For the last point, I think we want [temp.param] p3 to say "... template-name that designates a template whose specializations are types".

I think we want [temp.param] p3 to say "... template-name that designates a template whose specializations are types".

Is it [temp.arg.template] p1 instead of [temp.param] p3? I think [temp.arg.template] p1 might be changed to

A template-argument for a template template-parameter shall be a template-name that designates a template whose specializations are types, expressed as id-expression.

How about the point that a type-id(especially, typedef-name) is not an id-expression? I think it also should be fixed.

[temp.arg.template] is concerned with template arguments, i.e. the use of templates. [temp.param] is concerned with template parameters, i.e. the declaration and definition of templates. In order to properly syntax-check a template definition, I think we need to know that the identifier introduced via a template template parameter is actually (similar to) a class/alias template. Maybe we want to even say that those are alias templates.

I think [temp.arg.template] is fine as-is; this is just another place where id-expression is used with a non-expression meaning (which ought to be fixed, but not like this).

@jensmaurer

a template template parameter is actually (similar to) a class/alias template. Maybe we want to even say that those are alias templates.

I also think so. So, there are two remaining issues, [temp.param] p3 might be fixed by:

• A type-parameter whose identifier does not follow an ellipsis declares its identifier to be an alias template that is a template-name(if declared with template)

Another issue is we should give a special rule to specify the name of a type template parameter (declared without template) whose name becomes a typedef-name is interpreted as an id-expression whenever it is used as a template-argument.