[PATCH] Addresses TemplateParameterDepth calculation of Local Member templates

[Richard Smith]

Hi Faisal,

I'd prefer putting the TemplateParameterDepthRAII code next to the calls to
ActOnReenterTemplateScope (unless there's some reason for them to be
separated?). If you increase the depth once for each
ActOnReenterTemplateScope call, there shouldn't ever be a need to bump it
by more than one. That should also handle the cases of a generic lambda in
a default argument in a method or class template, and a generic lambda /
local class w/member template in a non-template function in a class
template, which I think the current patch won't get right.

+  // ok here we need to make sure that the depth of the template parameter
list
+  // of a potential member function template is adjusted for the template
+  // parameter depth - in case we have local classes with member templates

Please start comments with a capital letter and end them with a full stop.
Also drop the "ok" :)

+  TemplateParameterDepthRAII
CurTemplateDepthTracker(TemplateParameterDepth);
+  FunctionTemplateDecl *FTD = 0;
+  if (LM.TemplateScope && (FTD = dyn_cast<FunctionTemplateDecl>(LM.D))) {
+    TemplateParameterList *TPL = FTD->getTemplateParameters();
+    unsigned MemberTemplateDepth = TPL->getDepth();
+    while( TemplateParameterDepth <= MemberTemplateDepth )

Space after "while", no spaces immediately inside parens.

Index: test/CXX/generic-lambdas/fv_print.h

Subdirectories inside test/CXX should correspond to sections of the
standard. The test itself should be cleaned up prior to commit: remove the
printf-based testing mechanism, and don't use lli. If you really want to
test CodeGen rather than just Sema (and I think you can avoid that), you
should FileCheck the -emit-llvm output. Otherwise, try to craft a -verify
test, and put it in test/SemaTemplate.

Thanks!

On Mon, Apr 22, 2013 at 4:40 AM, Faisal Vali <faisalv at gmail.com> wrote:

> Hi Doug,
>     I had mentioned this issue to you briefly in Bristol having to do
> tangentially with generic lambdas: the template parameter depth does not
> get calculated correctly when local member templates are parsed.  This
> patch addresses that issue.  I know local member templates are not
> standard, but some of the machinery I will be using to implement
> incrementing depth appropriately within generic lambdas is included.  This
> only addresses the issue of template argument deduction for local member
> templates - other issues (and i don't remember what they are exactly now -
> i think you had showed me an example in Portland...) will probably get
> addressed as the generic lambda implementation moves along (if we care to
> fix local member templates entirely that is).
>
> Will refine the patch based on feedback.
>
> Thanks!
>
> Faisal Vali
>
>
> _______________________________________________
> cfe-commits mailing list
> cfe-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130422/48c7fe7c/attachment.html>