Performance issues caused by synchronization

[xixixi0hehehe]

Hi @StevenArzt,

I found that many threads are blocked in data flow analysis phase：

Check and debug the code， and find that it is blocked in this place：

Is this a bug, can remove the synchronized modifier of the checkStaticFieldUsed method?

[StevenArzt]

There is a trade-off between threads getting blocked because of synchronization, and computing the same data multiple times. This algorithm is inter-procedural, therefore, we have overlaps even if two threads ask for different methods. The two methods might end up in the same callee, even if they started from different methods. That's why we have a hard synchronoization here - make sure we know the field uses for a method X and all of its transitive callees, before computing it again for some other method Z, because we might then find that we already know Z, since Z was in the transitive callee list of X.

I know that this solution is not optimal. I wonder whether it would even make sense to pre-compute all uses of all static fields in thw hole program before running the data flow analysis - that lead to some unneessary computation, because not all static fields actually end up having tainted data, but on the other hand, we wouldn't need any synchronization during the taint analysis. I guess someone should try that out and measure what is more efficient.

[xixixi0hehehe]

Thank you for your response.
From my current practice, i think pre-compute is necessary. But i currently have no ability and time to solve this problem, hope can do it future.