Revert "Simplify internals of {Rc,Arc}::default"

[alexcrichton]

This reverts #152591 following a perf run being done. I'm not really positioned at this time to dive in further and understand the performance regressions, so I'll back away from Rc/Arc and leave them as-is.

[rustbot]

r? @Mark-Simulacrum

rustbot has assigned @Mark-Simulacrum.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: libs
• libs expanded to 7 candidates
• Random selection from Mark-Simulacrum, joboet

[joboet]

It appears that the pointer write trick doesn't work, the write of T::default() still emits a memcpy... and using Box::new also means the layout calculation will be optimised to constants. In general Box::new is just much more likely to have good behaviour even in debug builds.

@bors r+

[rust-bors]

📌 Commit 57d20c1 has been approved by joboet

It is now in the queue for this repository.

[jieyouxu]

(This has perf differences so to not make rollup perf more noisy)
@bors rollup=never

[jieyouxu]

Scheduling...
@bors p=1

[rust-bors]

☀️ Test successful - CI
Approved by: joboet
Duration: 3h 21m 33s
Pushing 6a979b3 to main...

[github-actions]

What is this?

This is an experimental post-merge analysis report that shows differences in test outcomes between the merged PR and its parent PR.

Comparing 25396cf (parent) -> 6a979b3 (this PR)

Test differences

Show 812 test diffs

812 doctest diffs were found. These are ignored, as they are noisy.

Test dashboard

Run

cargo run --manifest-path src/ci/citool/Cargo.toml -- \
    test-dashboard 6a979b3e32522049d0acb4a47f7ae44b7c8abfd5 --output-dir test-dashboard

And then open test-dashboard/index.html in your browser to see an overview of all executed tests.

Job duration changes

1. dist-aarch64-apple: 2h 19m -> 1h 50m (-20.8%)
2. dist-x86_64-apple: 2h 28m -> 2h 4m (-15.8%)
3. pr-check-2: 35m 8s -> 39m 53s (+13.5%)
4. x86_64-gnu-llvm-20-3: 1h 54m -> 1h 39m (-12.7%)
5. x86_64-gnu-llvm-21-2: 1h 23m -> 1h 34m (+12.5%)
6. aarch64-apple: 2h 56m -> 3h 15m (+10.7%)
7. i686-gnu-2: 1h 35m -> 1h 26m (-9.5%)
8. x86_64-gnu-nopt: 2h 11m -> 2h 23m (+9.3%)
9. aarch64-msvc-1: 2h 11m -> 2h (-8.7%)
10. dist-powerpc-linux: 1h 36m -> 1h 28m (-8.3%)

How to interpret the job duration changes?

Job durations can vary a lot, based on the actual runner instance
that executed the job, system noise, invalidated caches, etc. The table above is provided
mostly for t-infra members, for simpler debugging of potential CI slow-downs.

[rust-timer]

Finished benchmarking commit (6a979b3): comparison URL.

Overall result: ✅ improvements - no action needed

@rustbot label: -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
Improvements ✅ (primary)	-0.4%	[-0.7%, -0.3%]	6
Improvements ✅ (secondary)	-0.6%	[-0.6%, -0.6%]	1
All ❌✅ (primary)	-0.4%	[-0.7%, -0.3%]	6

Max RSS (memory usage)

Results (primary 6.5%, secondary 7.5%)

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

	mean	range	count
Regressions ❌ (primary)	6.5%	[6.5%, 6.5%]	1
Regressions ❌ (secondary)	7.5%	[7.5%, 7.5%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	6.5%	[6.5%, 6.5%]	1

Cycles

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

Binary size

Results (primary -0.8%)

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)	-	-	0
Improvements ✅ (primary)	-0.8%	[-1.2%, -0.5%]	2
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-0.8%	[-1.2%, -0.5%]	2

Bootstrap: 481.72s -> 484.473s (0.57%)
Artifact size: 395.60 MiB -> 397.64 MiB (0.52%)

[alexcrichton]

@joboet FWIW I was just copying what Box does which references #136043. If that doesn't work for Rc it probably doesn't work for Box either.

I also suspect that's not the source of regressions since that was in (I think?) compiling Cargo and html5ever. My guess is they both heavily use Rc or Arc and LLVM had to work a bit harder to optimize this.

It sounds like the codegen of neither Rc nor Arc can change, even in debug mode, so I don't know how to refactor things with that constraint.