[cfe-dev] class template methods

[Peter Stirling]

This does sound promising, but I'm not sure how to make use of it.

At the moment my walker (which is based on clang::ASTConsumer) has a 
handler for class templates that looks like:

void Walker::TraverseClassTemplate(clang::ClassTemplateDecl const* d) {
   TRACE(1, "walk template " + getDeclMetadataString(d));

   bool was = inInstantiation;
   inInstantiation = true;

   for(auto decl : d->specializations()) {
     visitor->RequestCompleteType(decl);
     TraverseCXXRecord(decl);
   }

   inInstantiation = was;
}

Visitor::RequestCompleteType() calls Sema::RequireCompleteType() (which 
IMO has the wrong name, it returns false on what I would consider, from 
the name, success).

So, I don't construct the specializations for methods directly in my 
code, I assume that some of the instantiation is performed by 
RequireCompleteType(), but surely if I checked for SFINAE at that level 
then I would lose EVERY methods for std::pair<int const, int const> and 
not just the ones that involve mutation?

On 14/04/15 18:26, Richard Smith wrote:
> On Tue, Apr 14, 2015 at 8:38 AM, Reid Kleckner <rnk at google.com 
> <mailto:rnk at google.com>> wrote:
>
>     Sounds a lot like a SFINAE test. Assuming you already have an AST
>     with the class template instantiation, then you'd attempt to
>     instantiate the assignment operator inside a SFINAETrap.
>     Instantiation will either pass or fail and you'll get your answer,
>     and you should be able to continue compilation or further queries
>     normally.
>
>
> I think the above is about the best that you can do.
>
> One note: if an error occurs outside of the immediate context of the 
> substitution, a SFINAETrap will not suppress it (which is actually a 
> good thing in this case, because you cannot continue compilation 
> normally after such an error -- any part of the AST that failed may be 
> marked invalid, suppressing further errors in other contexts).
>
>     On Mon, Apr 13, 2015 at 9:57 PM, Peter Stirling
>     <peter at pjstirling.plus.com <mailto:peter at pjstirling.plus.com>> wrote:
>
>         Hi,
>
>         Is there a way to ask clang whether a method for an
>         instantiation of a class template would be an error to call?
>         (I only need yes/no)
>
>         For example:
>
>         std::pair has assignment operators defined, but for the
>         instantiation std::pair<int const, int const> calling the
>         assignment operators is an error, because you can't assign to
>         int const.
>
>         If clang can't tell me this, I'm looking at generating a
>         translation unit with a call to the method, and then checking
>         whether there was an error compiling it. Given that I will
>         need to do this once per method, per instantiation, per
>         template, (which looks to be in the thousands based on a test
>         I just did) what is the best way to do this?
>         _______________________________________________
>         cfe-dev mailing list
>         cfe-dev at cs.uiuc.edu <mailto:cfe-dev at cs.uiuc.edu>
>         http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>
>
>
>     _______________________________________________
>     cfe-dev mailing list
>     cfe-dev at cs.uiuc.edu <mailto:cfe-dev at cs.uiuc.edu>
>     http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20150416/5bf30246/attachment.html>