[Soot-list] my StrongLocalMustAliasAnalysis gives unexpected results
Eric Bodden
eric.bodden at ec-spride.de
Tue Oct 23 10:41:03 EDT 2012
Hi Zell.
> Eric, could you pls. send me your new StrongLocalMustAliasAnalysis.java so I
> can look at it more closely?
It's on Github...
https://github.com/Sable/soot
> Besides, another imprecision issue may be due to the merge of
> LocalMustAliasAnalysis. But I am not that sure whether I misunderstand your
> merge operation.
>
> Here is a code fragment reproduces the imprecise results:
>
> if (...) { x = a; y = a ; }
> else { x= b; y = b ; }
>
> The current LocalMustAliasAnalysis associates different numbering for x and
> y and the end of the program, that is, the must-alias relation between x and
> y is not captured.
>
> Probably, you only wanted to capture must-alias over time, and must-alias
> between variables at the same control point is ignored ?
You are right, our value numbering currently does not capture this. I
actually wonder how this could be done with value numbers. Seems
tricky... I think one would have to use another implementation based
on alias-sets to model cases like that.
Eric
More information about the Soot-list
mailing list