Re: whats wrong with analysizing pointers this way ?

Perhaps this will work in restricted languages where pointer arithmetic
would not cross the variable boundary. Would it ?




shrey
dz wrote:
> Hi,
> Consider the following pointer analysis assuming the complete
> availability of the source code.
>
> I consider pointers pointing to a universe of variables whose
> address is taken, that includes variables that has been passed as
> parameters to functions, parameters whose address has been taken inside
> that function. For functions, that have calls to malloc the universe
> includes an additional generic element "heap"
>
> While I understand, u might be able to do better. But the bigger
> question is, that is this correct and feasible to do. It seems trivial
> in terms of feasibility unless the assumption that I made is not really
> right. The reason I am motivated to ask this is based on the current
> gcc's alias analysis, which reports a pointer as pointing to {ANYTHING}
> once it is passed as a parameter. The gcc guys cannot be wrong, so that
> brings me back to...whats wrong with the pointer analysis ?
>
> thanks
> dz

Return to the comp.compilers page.
Search the comp.compilers archives again.