could hasFlow support multi core?

[18Fl]

Hey, I write a codeql-query for chromium. And it works work well.

But I have a problem at here.

1. for query chromium, we always suggest only compile single module(just remove one folder, and create database).
2. But some chromium bug is cross module. It means I could find it once I know which files maybe have bug. It is hard.

I delete all relate .o file the bug happened. and create the database. Then my query works well.

But I think the more common situation is :

I don't know which file will have same pattern, so I just create a big database, which include multi module. Then run my query.

I run my query on this big database, seems it can't handle this. Because of that it will spend too much time on hasFlow Function(I ensure my query is right, because with a small database, it success which I mentioned before).

I search this on github slack, and one people said:

"I'd guess there is probably one core active that computes data-flow related stuff and all other cores/queries wait for this..."

So I just set --thread=0 in the vscode plugin, but seems it still can't work. Any tips? could make my query run more fast? Thx.

[aibaars]

Currently the CodeQL evaluator is able to evaluate different predicates in different threads. This normally works well, unless there is a very expensive predicate such as hasFlow in your case hat acts as a bottleneck, and there are no other predicates for other threads to be working on until it completes. The team is investigating finer grained parallelism to allow multiple threads to work together on a single predicate.

It is also possible that your database is so large that CodeQL requires more memory to analyze it efficiently. When memory is low CodeQL likely spends most of its time swapping things to disk or garbage collecting. Another possibility is that your query was poorly optimized by the CodeQL compiler. Here are some rough guidelines: https://codeql.github.com/docs/writing-codeql-queries/troubleshooting-query-performance/ .

To investigate the performance problem further, you can use the --evalator-log and --tuple-counting flags of codeql database analyze .