[cfe-dev] MemRegions as ProgramState info (was: checkBind: distinguish between MemRegionVal/ElementRegion)

[Aitor San Juan]

Yes, you're right. Maybe I was too focused on using the ProgramState.
Many thanks, Jordan.


2014-05-28 5:23 GMT+02:00 Jordan Rose <jordan_rose at apple.com>:

>
> On May 27, 2014, at 10:57 , Aitor San Juan <aitor.sj at opendeusto.es> wrote:
>
> Jordan, thanks for the comments.
>
> Doubts (1) and (2) are now clear to me.
>
> With regard to point (3), I didn't know *MemRegion*s can also be saved as
> custom ProgramState information, the same way as *SymbolRef*s. So to be
> able to keep track of which variables currently point to a String literal,
> if I haven't misunderstood you, I've tried saving as the checker's custom
> ProgramState information a MemRegion. In this scenario, I'm only interested
> in whether a *char ** variable points to a String literal or not. So
> instead of registering a Map with ProgramState, I think is enough to use a
> Set that contains variable which point to a String literal at any given
> time. Thus, if a given variable does not exist in the Set, the checker can
> assume it doesn't point to a String literal. I've done the registering with
> ProgramState as:
>
> typedef const MemRegion * MemRegionRef;
> *// This typedef does not officially exist, so I've privately defined it
> to the checker as a helper *REGISTER_SET_WITH_PROGRAMSTATE(StrLit,
> MemRegionRef)
>
> So far, so good. No run-time error for the time being. However, when I
> retrieve the current State and try to add/remove an *interested* variable
> (what I actually add/remove is its MemRegion), nothing happens. The method
> "contains" always returns false.
>
> I wonder if this approach (using a Set) is correct. I see that the
> contents of the ProgramState, are, for example, ElementRegions. Is it a
> matter of casting, as you pointed out in your previous comments?
>
>
> What you've done ought to work (if you're being careful about the region
> that represents the *pointer variable* and the region that represents
> what it points *to)*, but again it'd be far easier to just ask if the
> MemRegion is a StringRegion.
>
> Jordan
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20140530/828ba637/attachment.html>