You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Auto merge of #119238 - Mark-Simulacrum:def-hash-efficiency, r=<try>
Specialize DefPathHash table to skip encoding crate IDs
The current implementation is ad-hoc and likely should be replaced with a non-table based approach (i.e., fully pulling out DefPathHash from the rmeta table infrastructure, of which we use ~none now), but this was an easy way to get an initial PR out.
The main pending question is whether the assumption made here that there is exactly one shared prefix accurate? If not, is it right that the number should be typically small? (If so a deduplication scheme of which this is a special case almost certainly makes sense).
We encode a lot of these (1000s) so the savings of 8 bytes/hash add up quickly. Opening this PR to get opinions more on the general idea and to run perf on whether the underlying impl will perform OK.
0 commit comments