A series of Polonius Alpha refactors

[jackh726]

Based on #161776

Best reviewed by commit.

This is a series of refactors to enable lazy liveness calculation. Each commit is relatively targeted.

r? lqd

[jackh726]

Disclosure: An LLM was used for experiments here, but all code here is written fresh.

[lqd]

The rationale is not clear for a handful of these (in the code, comments, or the commit message). I expect them to make sense in light of #161938, so I'll have to review them later when I look at that PR.

• commits 1 and 2: I've reviewed these as part of Add a couple polonius tests and move universal region liveness generation #161776, and it's already in the queue
• commit 3: this feels like it could be simplified, and I imagine it makes more sense seen together with Defer extra liveness calculation for Polonius Alpha #161938, so let's remove it from this PR.
• commit 4: I left a comment
• commit 5: I'm not sure why, but ok, why not.
• commit 6: I left a few comments
• commit 7: I imagine it's same as commit 3; the split into a third context (liveness results, liveness context, liveness calculation) feels suboptimal, let's remove it from this PR as well.

We can land the straightforward commits soon, and leave the other two for later.

[jackh726]

commit 3: this feels like it could be simplified, and I imagine it makes more sense seen together with Defer extra liveness calculation for Polonius Alpha #161938, so let's remove it from this PR.

commit 7: I imagine it's same as commit 3; the split into a third context (liveness results, liveness context, liveness calculation) feels suboptimal, let's remove it from this PR as well.

Indeed, these make more sense in the context of #161938. In short, the LivenessCalculation struct factors essentially the logic to calculate the liveness for a single Local. This is helpful because we eagerly calculate not just liveness for the set NLL-relevant locals, but also things like dropck region constraints and legacy polonius facts (that's all in LivenessResults); but, we lazily only need the use/drop liveness points, which LivenessCalculation does. There is no longer a LivenessContext - it was LivenessContext and LivenessResults, now it's LivenessResults and LivenessCalculation.

The visitor -> visitor+traversal split is basically just type-safety around ensuring that liveness is calculated for a given region prior to liveness values being used; LocalizedConstraintGraphTraversal could almost just be replaced by an impl FnMut(RegionVid) -> impl LocalizedConstraintGraphVisitor.

Anyways, removed them from this PR - we can see if/how they fit later.

commit 5: I'm not sure why, but ok, why not.

Because when computing deferred liveness, I'm passing in just liveness: &mut LivenessValues and live_region_variances: &mut BTreeMap<RegionVid, ConstraintDirection>, not PoloniusContext.

[lqd]

I think we’ll need to do a handful of passes to simplify the entire process afterwards. The end state is starting to feel less like a series of high level steps than it is today, and while that is somewhat expected to allow liveness to be produced from two different places, we probably would benefit from clearly defined components/pieces of state.

@bors r+ rollup

[rust-bors]

📌 Commit b58ee5b has been approved by lqd

It is now in the queue for this repository.

[rust-timer]

Finished benchmarking commit (7dfde5a): comparison URL.

Overall result: ❌✅ regressions and improvements - no action needed

Benchmarking means the PR may be perf-sensitive. Consider adding rollup=never if this change is not fit for rolling up.

@rustbot label: -S-waiting-on-perf -perf-regression

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	0.2%	[0.1%, 0.3%]	3
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.4%	[-0.4%, -0.4%]	1
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

This perf run didn't have relevant results for this metric.

Cycles

This perf run didn't have relevant results for this metric.

Binary size

This perf run didn't have relevant results for this metric.

Bootstrap: missing data
Artifact size: 400.59 MiB -> 402.55 MiB (0.49%)