Previously Operand's getLocation would take it from the Operand use.
This lead to slightly confusing query results, where for example an
issue related to a call argument would highlight the function part of
the call instead of the parameter.
When inserting side effect instructions for argument indirections, we now insert side effects for smart pointers as we would for raw pointers. The address operand of the side effect instruction is the smart pointer object, which is a bit odd. However, I'd like to think through the design of a more principled solution before doing additional work.
A few new tests are added to the existing IR tests. In addition, the IR tests now `#include` some of the shared STL headers. I've disabled IR dumps for functions from those headers, since they only get in the way of the test cases we intended.
This change was necessary for my upcoming changes to introduce side effect instructions for indirections of smart pointers. The code to decide which parameters have which side effects appeared in both the IPA constructor for `TTranslatedSideEffect` and in `TranslatedCall`. These two versions didn't quite agree, especially once the `SideEffectFunction` model provides its own side effects instead of the defaults.
The relevant code has now been factored out into `SideEffects.qll`. This queries the model if one exists, and provides default side effects if no model exists. This fixes at least one existing issue, where we were emitting a buffer read side effect for `*this` instead of an indirect read side effect. This accounts for all of the IR diffs in the tests.
Removes the IR consistency checks for conflated memory and marks
instructions that have a conflated result with a percent sign (%)
instead. This avoids reimplementing part of the alias analysis logic
in the consistency check.
This fixes the IR generation for member accesses where the qualifier is a prvalue that is _not_ the load of a `TemporaryObjectExpr`. We synthesize a temporary variable during IR generation instead. It fits into the IR construction code at the same spot as `TranslatedLoad`, since it's basically the opposite of `TranslatedLoad` (prvalue->glvalue instead of vice versa). Note that array prvalues require special treatment.
This fixes some consistency errors in the `syntax-zoo`. It introduces three new ones in `dataflow-ir-consistency.expected`, but those are along the same lines as tons of existing failures.
For historical reasons, the extractor marks the temporary object expression used as the qualifier of a member access as a prvalue(load), even though the current C++ standard says that the temporary object materialization results in a glvalue. Added some special handling to ignore the load for both field accesses and member function calls.
This fixes all of the consistency failures in our regular tests, and all of the related failures in `syntax-zoo` other than the ones that deal with pointers-to-member, which aren't really supported yet anyway.
