fix: Remove the assert for interned structs that `last_interned_at >= last_changed_revision`
#804Comparing
ChayimFriedman2:outside-intern
(fafa4ca
) withmaster
(ab7ecb4
)-5%
- Improvements
- 0
- Regressions
- 1
- Untouched
- 11
- New
- 0
- Dropped
- 0
- Ignored
- 0
Benchmarks
Failed
amortized[SupertypeInput]Regression
benches/compare.rs::benches::inputs::Mutating Inputs::amortized[SupertypeInput]-5%
4 µs
4.2 µs
Passed
amortized[Input]
benches/compare.rs::benches::inputs::Mutating Inputs::amortized[Input]+1%
3.5 µs
3.5 µs
many_tracked_structs
benches/incremental.rs::benches::many_tracked_structs::many_tracked_structs0%
38.3 µs
38.2 µs
accumulator
benches/accumulator.rs::benches::accumulator::accumulator0%
4.3 ms
4.3 ms
converge_diverge
benches/dataflow.rs::benches::dataflow::converge_diverge0%
137.7 µs
137.8 µs
new[Input]
benches/compare.rs::benches::inputs::Mutating Inputs::new[Input]-1%
10.4 µs
10.5 µs
amortized[InternedInput]
benches/compare.rs::benches::inputs::Mutating Inputs::amortized[InternedInput]-1%
3.4 µs
3.5 µs
new[InternedInput]
benches/compare.rs::benches::inputs::Mutating Inputs::new[InternedInput]-1%
5.8 µs
5.9 µs
mutating[30]
benches/compare.rs::benches::mutating_inputs::Mutating Inputs::mutating[30]-1%
14.1 µs
14.3 µs
mutating[20]
benches/compare.rs::benches::mutating_inputs::Mutating Inputs::mutating[20]-1%
14 µs
14.2 µs
mutating[10]
benches/compare.rs::benches::mutating_inputs::Mutating Inputs::mutating[10]-1%
13.8 µs
14 µs
new[SupertypeInput]
benches/compare.rs::benches::inputs::Mutating Inputs::new[SupertypeInput]-2%
17 µs
17.3 µs
Commits
Click on a commit to change the comparison rangeBase
master
ab7ecb4
-5%
Remove the assert for interned structs that `last_interned_at >= last_changed_revision`
This assert was incorrect, in a subtle but important case (which caused half of rust-analyzer tests to panic): if a struct was interned inside a query, there is indeed no way to make `last_changed_revision` pass past it. But if it was created *outside* any query, it actually can. In the simplest case (which is also what happened in rust-analyzer), the same non-query function that interned it can create a new revision, then obtain its fields.
I couldn't find a way to make this assert reliable, so I just deleted it. I think (hope) this assert was just defensive and not really needed, in which case removing it is fine.
fafa4ca
19 hours ago
by ChayimFriedman2