The data we need for the GC Test Pilot is not currently exposed.

I think the implementation will look like this: - At every GC, we store the results (whatever they are) in memory. - At some interval, Test Pilot will call some function we've exposed, which returns the data. - If we do too many GCs between polls, we throw the data away, or overwrite it. Test Pilot handles everything else, AIUI. It asks the user if they're OK with starting the study, polls our data and stores it in a sqlite DB until it's ready to be uploaded, then asks the user if they'd like to upload it.

Taking assignment. Intended instrumentation should provide data on: - frequency - duration - cause (i.e. full, local) - compartment count - compartment heap size Ideally we would also measure the time it takes to get to the major select statement in the event loop and determine what percentage of that time is JS GC, but we assume that instrumenting the main event loop is overly intrusive.

More information is also in 619561 comment 1 and onward.