Does gc_start use unusual values?

[richardhills]

gc_start is:

void gc_start(GarbageCollector* gc, void* bos) { gc_start_ext(gc, bos, 1024, 1024, 0.2, 0.8, 0.5); }

So that downsize_load_factor is 0.2, upsize_load_factor is 0.8 and sweep_factor is 0.5

If I have interpreted this correctly, if you genuinely have, say, 60% of allocations used and reachable, this will cause a sweep to run with each new allocation, until eventually you get to 80% of allocations used, when it will finally increase the capacity.

I suspect the actual configuration should be gc_start_ext(gc, bos, 1024, 1024, 0.2, 0.5, 0.8); so that we sweep at 80%, and upsize if we can't get allocations back below 50% after a sweep.

[mkirchner]

Hi @richardhills, thanks for reaching out! The way I read the code (and it's been a while that I wrote it) is that sweep decisions are based on sweep_limit and sweep_factor just influences how the sweep limit is updated here:

gc/src/gc.c

Line 224 in 7f6f17c

am->sweep_limit = am->size + am->sweep_factor * (am->capacity - am->size);

I do agree that this may not be an ideal set of parameters but I don't think it's as simple as a parameter flip. That said, there are two learnings here:

• gc is an educational piece of code; I have seen people use it in production but always recommend against it, just use Boehm's implementation. It's much faster and easy to integrate.
• When you play around with gc, be aware that you need to set parameters to match your environment - there is no such thing as a good default parameter.

If you have some time to track this down and provide a more detailed paragraph how the sweep_limit evolves over time and across a few different use cases, I'd be happy to add that to the README.