Re: [abc] PLDI paper

On Sat, Apr 09, 2005 at 08:35:32PM +0200, Damien Sereni wrote:
> > 8) Damien mentioned a variation of our inter-proc algorithm ... Damien ...
> > what was it - something about not needed something on "all paths"?
>
> I just noticed in chicago that the condition for setting a cflow
> residue to alwaysMatch is stronger than it needs to be -
>
> what i would expect would be
> a query shadow qsh is statically true if for every interprocedural
> path p to qsh there is an update shadow in p
> [where an interprocedural path includes calls but not returns]
>
> whereas the condition that there is an update shadow sh such that qsh
> is in mustCflow(sh) is equivalent to :
> every interprocedural path p to qsh goes through sh [ie all paths must
> go through the same shadow]
>
> (please check that i haven't made some mistake along the way!)

I just checked the code, and it turns out the code implements the first
(improved) variation, while the paper describes the second (worse)
variation.

Ondrej

> I guess the mustCflow version is probably easier / more efficient to
> compute, but maybe we should mention the other one (of course, the
> benchmarks show the mustCflow is precise enough)
>
> Cheers,
> Damien
>
Received on Sat Apr 9 20:01:33 2005