Support buffer de-duplication
mdekstrand opened this issue · comments
Michael Ekstrand commented
It's possible that an object may have multiple numpy arrays with the same contents (this will arise in some LensKit use cases). We can support de-duplication by recording more robust checksums (MD5 or SHA) of buffers, and making the buffer store effectively content-addressed.
Michael Ekstrand commented
Format version 2 has file checksums, which is one of the prerequisites for this.