[cfe-commits] [cfe-dev] [PATCH] Source location of "template" keyword in TemplateSpecializationTypeLoc.

[Enea Zaffanella]

On 01/31/2012 10:25 PM, Richard Smith wrote:
> Hi,
>
> On Tue, Jan 31, 2012 at 5:04 AM, Enea Zaffanella <zaffanella at cs.unipr.it
> <mailto:zaffanella at cs.unipr.it>> wrote:
>
>     Please find attached for review a patch that adds the source
>     location of the optional "template" keyword to
>     TemplateSpecializationTypeLoc, thereby allowing for a more faithful
>     AST representation of constructs such as the following:
>
>     namespace N {
>     template <typename T>
>     struct S;
>     }
>
>     template <typename T>
>     void foo() {
>       N::S<T> s;          // TSTLoc without template keyword
>       N::template S<T> s; // TSTLoc with template keyword
>     }
>
>     NOTE: we can not exploit the pretty printer to provide a testcase
>     for the patch, because (as far as we can tell) the pretty printer is
>     visiting the Type hierarchy (i.e., it disregards the TypeLoc hierarchy).
>
>     While at it, also renamed a few members in class
>       DependentTemplateSpecializatio__nTypeLoc
>     so as to have a more consistent and clearer naming scheme:
>       getKeywordLoc --> getElaboratedKeywordLoc
>       getNameLoc    --> getTemplateNameLoc
>
>     The patch is adding 3 new FIXME's in TreeTransform.h: these will be
>     addressed in a forthcoming patch adding the template keyword
>     location to DependentTemplateSpecializatio__nTypeLoc.
>
>
> Patches should be sent to cfe-commits for review; dropping cfe-dev via bcc.

OK, thanks.

> The patch itself is missing tests. Can the fix to TST's source range be
> tested with c-index-test?

Given the very little knowledge we have of the c-index-test tool, we can 
not produce such a test unless some CIndex expert provides some useful 
hints (e.g., pointing us to another testcase where c-index-test is used 
to check non-trivial source location info from a TypeLoc; I say 
"non-trivial" because in the relevant examples the TSTLoc is not the 
outermost type, but it is wrapped by an ElaboratedTypeLoc).

Enea.

> Thanks,
> Richard