[cfe-dev] Proposed: change tracking for RegionStore

[Ted Kremenek]

On Aug 1, 2010, at 4:53 PM, Jordy Rose wrote:

> That's a good idea. I'd be a little worried transferring from one
> GRExprEngine to the next during far inline calls, but I guess that can wait
> until there's more support for that. (Plenty of checkers, for example,
> assume the ASTContext doesn't change between invocations.) And I assume
> these callbacks would happen at the end of the StoreManager public
> interface methods.

Besides the ASTContext changing, we possibly will have different MemRegions as well.  We haven't worked those details out yet.

> I'd actually still like to push this up to GRStateManager, since that
> would allow checkers to mess with their GDM store as a side effect of a
> region change (in the case of CStringChecker, to invalidate any recorded
> strlen).

Yes, this makes sense.  Checkers don't get to manipulate stores directly, and GRState is the only place they can put new state.

The problem with just generating GRStates, as opposed to ExplodedNodes, is that it prevents Checkers from registering an error immediately when the memory "notification" takes place.