Alyssa Ross

> @alyssais should these dependencies be added to any other rubies that compile? No other ruby that compiles seems to need these. Rubinius seems like it would, but doesn't compile...

Hi @tenderlove, thanks for investigating for me! I don't _think_ I've ever seen a plateau, although it's possible since the cache size is so big that I just wouldn't have...

Something else that's just occurred to me — there's a `Repository#close` method that looks interesting, now that it appears to be a caching problem: > Frees all the resources used...

With no changes, my program currently peaks at ~1.4GB memory, of which ~232MB is apparently cache. I tried calling `#close` on the repo after _every_ time I generated a diff...

It might be worth considering renaming (or at least aliasing) `#close`, too? It's a really non-obvious name, to me at least — especially the fact that the repository object is...

I was going to suggest `cleanup`, since that's what libgit2 seems to call it, or `clean`, maybe.

Bah, even though I'm calling `#clean` periodically, I'm still getting a memory buildup over time. Here's a graph (y-axis is %mem used by the script, x-axis is seconds): ``` 3...

I've also tried modifying it to create new repository objects every so often, and that appears to have no difference compared to calling `#close`.

> Are you seeing any GC stats rising? Specifically `GC.stat(:heap_sorted_length)` and `GC.stat(:heap_live_slots)`. `heap_sorted_length` is stable at 300, and `heap_live_slots` is stable at 78594±1. Here's the full data, with an entry...

A graph showing the same as the previous one, but with a much bigger repo (torvalds/linux — about 10x the number of commits). It's less noticeable because I loaded _so...