[cfe-dev] coverage analysis with template-heavy code

[Manuel Holtgrewe]

ping

On 02/08/2012 05:40 PM, Manuel Holtgrewe wrote:
> Dear all,
>
> maybe this was asked already or is documented elsewhere. In this case,
> please point me in the right direction.
>
> I want to compute code coverage of a C++ program. Since some time,
> llvm/clang can do this. Apparently, this functionality has moved from
> clang into clang some time back (I had an old checkout where it appeared
> to be in clang, now it appears to be in llvm).
>
> Now, I tested this and it worked just like GCC. While this is nice for
> most people, I would like to do more. Here is where I see limitations
> with being on par with GCC:
>
> Our library consists almost only of templates. If a template is
> instantiated, the instrumented clang output gives me branches etc. as I
> would expect. However, no code is generated for template functions and
> thus no code is generated.
>
> I would like to programatically find such uninstantiated functions, however.
>
> What is the best way to do so? Would it be possible/make sense to extend
> the llvm/clang instrumentation such that a dummy symbol is generated for
> each uninstantiated template function?
>
> Cheers,
> Manuel