mirror of
https://github.com/urbit/shrub.git
synced 2024-12-23 19:05:48 +03:00
7ded3f127c
We were unconditionally adding equal items to the hashtable of already-compared pointers. This table exists so that if we've already determined two non-pointer-equal nouns are value-equal, we don't have to check them again. However, atoms (especially direct) ended up in this hashtable even though we don't look those up. This makes sure we don't add items to the hashtable if they're "pointer"-equal (which for direct atoms is value-equal). The impact of this inefficiency was greatly magnified by the issue with +mug where (mug a a) = (mug b b) for all a, b. For this reason, these identical pairs added to the hashtable always had the same mug, so they were added to the same bucket, which meant adding to that list required linearly traversing the entire bucket. This was the first barrier that was causing |pack to take a long time on ships which distribute OTAs, but it isn't a complete solution for |pack. |
||
|---|---|---|
| .. | ||
| arvo | ||
| ent | ||
| ge-additions | ||
| herb | ||
| hs | ||
| interface | ||
| libaes_siv | ||
| urbit | ||