compute temporary scopes when building MIR, not THIR

[dianne]

This accomplishes two things:

• Makes the THIR slightly smaller by not attaching a full TempLifetime to every expression.
• Reduces the number of traversals of the ScopeTree by only calling ScopeTree::temporary_scope when building MIR for something that needs to be dropped in a temporary scope.

[dianne]

@bors try @rust-timer queue

[rust-bors]

☀️ Try build successful (CI)
Build commit: 9de26a5 (9de26a5918b5655a37eefd1b02cdb1ff0cdcdc74, parent: 87f9dcd5e28a301fce81515bd4bea792444dcf29)

[rust-timer]

Finished benchmarking commit (9de26a5): comparison URL.

Overall result: ✅ improvements - no action needed

Benchmarking this pull request means it may be perf-sensitive – we'll automatically label it not fit for rolling up. You can override this, but we strongly advise not to, due to possible changes in compiler perf.

@bors rollup=never
@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)	1.2%	[1.2%, 1.2%]	1
Regressions ❌ (secondary)	0.7%	[0.2%, 1.2%]	2
Improvements ✅ (primary)	-0.4%	[-2.0%, -0.1%]	56
Improvements ✅ (secondary)	-0.9%	[-2.5%, -0.0%]	81
All ❌✅ (primary)	-0.4%	[-2.0%, 1.2%]	57

Max RSS (memory usage)

Results (primary -0.5%, secondary -2.8%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	1.4%	[1.4%, 1.4%]	1
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-1.5%	[-1.9%, -1.2%]	2
Improvements ✅ (secondary)	-2.8%	[-4.8%, -1.3%]	28
All ❌✅ (primary)	-0.5%	[-1.9%, 1.4%]	3

Cycles

Results (primary -2.3%, secondary -4.0%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	4.9%	[4.2%, 5.5%]	2
Improvements ✅ (primary)	-2.3%	[-2.4%, -2.2%]	3
Improvements ✅ (secondary)	-4.6%	[-8.7%, -2.3%]	29
All ❌✅ (primary)	-2.3%	[-2.4%, -2.2%]	3

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 475.809s -> 473.629s (-0.46%)
Artifact size: 390.93 MiB -> 390.83 MiB (-0.03%)

[dianne]

This could be removed entirely from thir::Expr, and we could use thir::ExprKind::Scopes to set up temporary scopes, but it's a little subtle/inconvenient for certain THIR lowerings (looking at &pin mut here) so for now I've opted to keep it simple.

[cjgillot]

The temp_scope_id annotations here are even more obscure than the previous version. Should we dump the region tree alongside THIR for debugging purposes?

[dianne]

That might help, yeah. I'm not sure a raw Debug dump of the scope tree would be that legible either, but maybe for the -Zunpretty=thir-tree output it'd make sense to inline the relevant bits? I worry a little it could be confusing to have information in there that isn't actually part of the THIR, but it'd also let us print scopes for binding patterns (which aren't straightforward for super let).

I think my ideal though would be to add a new way of pretty-printing the THIR that contains the actual drop scopes used to build the MIR (and maybe also get rid of temp_scope_id entirely). It'd be more work to implement, but it might be more useful? Speaking for myself at least, currently when I want to know drop scopes I read a promoted MIR dump since the THIR doesn't tell me at a glance which expressions correspond to places/operands/temporaries/constants/etc. (and since THIR dumps are verbose enough that they're not particularly more readable than MIR, but I'm not sure how fixable that is).

[dianne]

Opening for review now that #148658 is merged. r? cjgillot if you don't mind

[rustbot]

Some changes occurred in match lowering

cc @Nadrieril