Gábor Horváth

Diagnose meaningless annotations

2

Diagnose when we annotate a non-pointer with `gsl::lifetime`, `gsl::lifetime_out`, `gsl::lifetime_in`. Diagnose when we annotate a top-level pointer with `gsl::lifetime_out`. Diagnose when we annotate a non-owner with `gsl::lifetime_const`. Is there more...

Prettier printing for variables

2

Would be nice to condense `(*a).f` into `a->f` for diagnostics. It should work well for multiple dereferences. So: ``` (*(*a)).f => (*a)->f ```

False negative with returning address of a temporary

1

Repro: ``` namespace std { template struct reference_wrapper { reference_wrapper(T &&); }; template reference_wrapper ref(T& t) noexcept; } std::reference_wrapper treatForwardingRefAsLifetimeConst() { const int &b = int(6); return std::ref(b); } ```

Make setPSet/getPSet more robust?

2

It is more of a question than a bug or feature. Lots of the visit functions have extra checks if an expression can have a PSet. We might also miss...

Divergent loop

2

In the following code the analysis diverge, the pset of `sp` oscillates between unkown and invalid: ``` struct stuff { stuff(); stuff(const stuff&); stuff *next; }; void test() { stuff...

There are much more lifetime_const methods/functions in the STL than supported by the current implementation. Rather than convoluting the code it might be better to have an array consisting of...

Cache type categories for types

4

Right now each time we see a type we will walk the corresponding declaration and base classes to check its category. We could have a DenseMap to avoid the redundant...

- [ ] Assigning aggregates - [ ] Expanding aggregates for function calls - [x] Ability to track fields in PMap/PSets

In case the tag is not provided for a git checkout, the script should embed the git hash in the output HTML to make the experiment reproducible.