[cfe-dev] libclang: token annotation

[Argyrios Kyrtzidis]

On May 25, 2012, at 6:03 AM, Michał Staromiejski wrote:

> Hi everyone!
> 
> First of all, thank you for libclang ;)

You're welcome :)

> 
> I'm working on semantic highlight plugin for Code::Blocks IDE based on libclang.
> So far, everything works quite well, but I've noticed some issues with
> the functions clang_annotateTokens and (this is probably more specific) clang_getCursor.
> 
> I'm using svn version (rev. 157460) but the same is with 3.1.
> 
> From the documentation:
> "clang_getCursor() maps an arbitrary source location within a translation unit down to the most specific cursor that describes the entity at that location."

As a sidenote, in case it isn't obvious, if the location cannot be resolved it resolves to the enclosing entity, e.g. if you point at a comment inside a namespace it will give you the namespace.

> 
> And the weird things begin. Consider the following (valid) C++ code:
> 
> ---------------------------------8<----------------------------------------
> #ifdef _MY_MACRO_
> #define NUM 2
> #else
> #define NUM 5
> #endif
> 
> template <class T>
> class A
> {
> #ifdef _MY_MACRO_ // (1)
>     int a; // (2)
> #else
>     bool a;
> #endif
> 
> public:
>     static bool array[NUM]; // (3)
>     operator T();
> };
> 
> template <class T> // (4)
> bool A<T>::array[NUM]; // (5)
> 
> bool g(A<bool> a)
> {
>     return a; // (6)
> }
> 
> template <class T> // (4)
> T f(A<T> a)
> {
>     return a;
> }
> ---------------------------------8<----------------------------------------
> 
> Using clang_annotateTokens (clang_getCursor) or c-index-test -test-annotate-tokens (-cursor-at, resp.) everything but the following is OK:
> 
> (1) Inside any block (class, struct, namespace, ... declaration, function body, etc.) preprocessor directives
> are not annotated as such.

Please file a bug for this.

> (2) If the macro _MY_MACRO_ is not defined, all tokens in "int a;" are annotated in same way as the block they belong to;
> shouldn't they be annotated as, for example, CXXCursor_InactiveCode (new CXCursorKind value)?

This is a known limitation, we don't identify inactive source due to preprocessor directives so it resolves to the enclosing scope; it'd be good to do as you said, please file a bug to track it.

> (3) It's weird but here clang_annotateTokens behaves different from c-index-test: the former annotates the 'NUM' token
> as 'VarDecl=array' (wrong) and the latter as 'macro expansion=NUM', as expected...

File a bug with a test case that uses clang_annotateTokens and reproduces this.

> (4) The whole line is annotated here with CXCursor_FirstInvalid (70) and this concerns all such 'template' lines.

We should probably identify this as part of the 'array' declaration, and have 'T' be a TemplateTypeParameter, file a bug.

> (5) The whole line is marked as 'VarDecl=array' (IHMO 'A' should be 'TemplateRef=A' and 'T' should be 'TypeRef=T', 'NUM' behaves similarly as in (3)).

I believe you are right, file bug.

> (6) 'a' is annotated as 'CallExpr=operator _Bool', shouldn't be 'DeclRefExpr=a'? It is weird, but in the function 'f' there is no such problem.

You are right, and, you guessed it, please file a bug.

> 
> Is it a bug or do I miss something?

It's better that each separate issue is tracked in different bugs, the work involved varies greatly.

Thanks for the feedback!

> 
> Cheers,
> Michal Staromiejski
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev