somethingelseentirely
commented
5 months ago Because we simplified the architecture to use the BlobSet for strong blobs, the BlobCache now only needs to wrap a BlobPull.
Some open questions are:
- Do we want a hierarchical cache that stores both on disk and in memory, potentially using memory mapping?
- Do we want to also cache the parsed representations?
We still need to implement the BlobCache. Because we can't use the WeakMap hack that we use in a GCed language, we have to use a proper cache.
Quick Cache looks like the ideal candidate for this, but it's missing cache pinning, which we need for strong blobs. So the next step is to implement cache pinning for it.