`BTreeMap::entry` API?

[dpc]

If I need to update a record, I think I need to do a get + unwrap_or_default, modify, and then insert? Does it cost me a lookup twice? Would it make sense to have an API like BTreeMap::entry?

[dpc]

Hmmm... insert_reserve is kind of like it, but I don't see an Option anywhere, and the value_len doesn't seem like what I have.

[cberner]

yes, it does cost two lookups. However, they will be cached, so it probably doesn't matter very much. BtreeMap::entry is probably not a very good fit for redb, because values can't, generally, be mutated in-place because the serialized size would have to match

[adamreichold]

because values can't, generally, be mutated in-place because the serialized size would have to match

Not wanting to relitigate this whole thing, but I found that this case is somewhat common, at least by number of accesses. For example for persistent caches where I have bincode-encoded structs which contain last-accessed/updated timestamps, they are frequently updated without their serialized size changing as only the timestamp actually changes which has a fixed size even in the serialized representation.