[cfe-dev] Symbolic Extents

[Jordy Rose]

On Wed, 30 Jun 2010 21:11:26 -0700, Ted Kremenek <kremenek at apple.com>
wrote:
> On Jun 30, 2010, at 5:47 PM, Jordy Rose wrote:
>> I wasn't sure how to implement the liveness tracking for SymbolExtent.
We
>> can't just say "SR.markLive(R->getExtent().getAsSymbol())" since that
>> will
>> create the extent symbol if it doesn't already exist. Any
>> ideas?<symbolic-extents.patch>
> 
> Great work Jordy.  In response to your last point, instead of explicitly
> marking extent symbols live, how about we just define them as live if
the
> corresponding region is live?  That is, put the smarts in
> SymbolManager::isLive(), and don't even bother marking them as live.

<snip>

> See my comment above.  We can define a SymbolExtent as live if the
region
> is live.  This involves changing:
> 
>   bool SymbolReaper::isLive(SymbolRef sym) {
> 
> to:
> 
>   bool SymbolReaper::isLive(const Stmt* Loc, SymbolRef sym) {
> 
> and then querying the region of SymbolExtent:
> 
>     if (const SymbolExtent *extent = dyn_cast<SymbolExtent>(sym)) {
> 	const SubRegion *R = extent->getRegion();
>         while (R) {
>           if (const VarRegion *VR = dyn_cast<VarRegion>(R))
>             return isLive(Loc, VR);
>           if (const SymbolicRegion *SR = dyn_cast<SymbolicRegion>(R))
>             return isLive(SR->getSymbol());
>          R = dyn_cast<SubRegion>(R->getSuperRegion());
>        }
>        return false;
>     }

Can we make the current statement a SymbolReaper ivar? We're already
creating one per statement in GRExprEngine, and we pass the statement
around to most -- but not quite all -- of the dead-symbol-clearing code, so
it's not much functional difference. And it would let us take /out/ the Loc
parameter from the other isLive() methods.