Ah, this probably also needs some changes to load the metadata back into memory.

[EDIT]: I guess it depends on the metadata? For coverage, it actually is probablynot important to load the metadata back into memory.

Ah, this probably also needs some changes to load the metadata back into memory.

[EDIT]: I guess it depends on the metadata? For coverage, it actually is probablynot important to load the metadata back into memory.

Is there a downside to loading it? You don't know what people do with the metadata, they may look at the coverage data for some random things later, too..

No, agreed that there's no downside to loading it. However, there's also not an existing API for it. The metadata path, for instance, isonly calculated just before storing it to disk.

I think loading metadata from disk would be good, but the issue of keeping it in memory is a critical blocker for me right now - I just can't run my fuzzer for very long before it OOMs. I don't personally use metadata except for coverage in this fuzzer, and I don't use the coverage metadata outside of the coverage feedback. So for me, evicting the metadata is very high priority and loading it back in is less important.

We can't drop metadata without reloading it I think?
Add more swap? :P

Maybe we can make it an option for the time being until someone is bored enough to implement it right?

Alright, will reopen when I have a minute to be more thorough.