[PATCH] [Sema] Re-land "Make FunctionType's TSI use unadjusted argument types"

[Reid Kleckner]

On Wed, Jun 12, 2013 at 9:15 AM, Reid Kleckner <rnk at google.com> wrote:

> On Tue, Jun 11, 2013 at 6:24 PM, Eli Friedman <eli.friedman at gmail.com>wrote:
>
>> On Tue, Jun 11, 2013 at 3:09 PM, Reid Kleckner <rnk at google.com> wrote:
>>
>>> For a Type held by an AST node, I completely agree, and I'll try to
>>> address that.  For a type held by a TypeSourceInfo, I disagree, the
>>> FunctionProtoType should really hold the undecayed parameter types.
>>>
>>
>> Hmm... I was thinking more along the lines of keeping around the
>> un-decayed type, but automatically decaying it for the user of the API
>> unless they explicitly request the un-decayed version.  I'm not sure I like
>> keeping different versions of the type in the AST nodes vs. TypeSourceInfo.
>>
>
> Maybe, but it seems pretty heavyweight to add a custom iterator to do the
> decay.  Right now the iterator is just an ArrayRef.  Every dynamic decay
> will trigger a uniquing in the ASTContext, unless we put some kind of cache
> into ArrayType or FunctionType similar to how CanonicalType works.
>

I think I'm going to abandon this change.  I think I see a way forward on
my real problem, which is mangling MS function types.  I didn't realize
that the TypeLoc holds the original ParmVarDecls.  The ParmVarDecl holds
both the decayed and undecayed type.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130612/a6226bc5/attachment.html>