Patch lacking changes in tools/libclang/libclang.exports

Some suggestions in D27384 from @compnerd also apply to your patch (at least about using auto), so you should probably fix it too.

Please, integrate test cases from D27384 into this patch and make sure that tests still pass.

Turns out we already have another patch that fixes the same problem: D26907

Ah, sorry I didn't notice that review, let's keep to your version of the patch then.

This breaks functionality of clang_getNumTemplateArguments/clang_getTemplateArgumentAsType in some case.

This patch breaks python binding tests (which I fixes just now).
Please, next time launch python tests too

LGTM!

Ok, If there are no objections, I'll commit this revision next week then.

It reminded me http://reviews.llvm.org/D17278

I see. There were some changes recently in the indexing functionality. I'm not sure whether this change is intended or not, but since it's not your fault and we already have a lot of tests confirming that clang_CXXMethod_isDeleted is working as expected (e.g. c-index-test -test-print-type) I think it's ok to remove this part of the test then

Looking a bit more at the recent commits in libclang, it seems like all recently added functions follow this naming convention:
clang_Cursor_*
clang_Type_*
e.t.c.

Yes, I've seen that python tests segfaulted too, never had time to investigate it though.

Generally looks pretty good to me. Maybe it's worth to factor out duplicating code, but I guess it's not so important.

Ok, after some more testing I found out that elaborated types can have cv specifiers attached, but after using ET->getNamedType() we lose that information. So obviously the previous solution was wrong. Therefore I'm falling back to the original idea here once again: exposing the elaborated type. Also extending tests to account for the found bug.

I've seen this bug too. Generally this patch looks good to me.

Added tests

Yes, I have commit acces now (didn't test it yet though). I'll try to submit this patch soon.

Well, I think it's ok then. Still there is no test for visibility set from command line case.
Also adding comment to clang_getCursorVisibility explaining that it also works with visibility set from command line would be very useful.

Isn't r246931 what you're looking for?

AFAIUI the canonical type could never be an elaborated type, but just type could.

Yes, I've noticed that inconsistency too. But, seems like, this is an unrelated issue: TemplateSpecializationType type for some reasons uses id instead of qid (See also all other tests with templates in print-type.cpp, e.g. FieldDecl=baz). To get the type with nested name specifiers for the TemplateSpecializationType c-index-test should use the clang_getCanonicalType. But it doesn't use it for pointers. (Compare c-index-test.c:1289 and c-index-test.c:1335)

Yes, now it's ready.

It just occured to me, why do we need to expose the ElaboratedType if we can retrieve the original type and return it to user. Especially when we already do that for Attributed and Decayed types. So here is a much simplier solution.

Yes, please submit it.

This is actually an unrelated issue (clang_getTypeDeclaration doesn't handle the auto type)

Ping? Can someone commit it?

Address raised issues.

Fixed minor formatting issues.