Vec::push in consts MVP

[fee1-dead]

Example:

const X: &'static [u32] = {
    let mut v = Vec::with_capacity(6);
    let mut x = 1;
    while x < 42 {
        v.push(x);
        x *= 2;
    }
    assert!(v.len() == 6);
    v.const_leak()
};

assert_eq!([1, 2, 4, 8, 16, 32], X);

Oh this is fun...

• We split out the implementation of Global such that it calls intrinsics::const_allocate and intrinsics::const_deallocate during compile time. This is achieved using const_eval_select
• This allows us to impl const Allocator for Global
• We then constify everything necessary for Vec::with_capacity and Vec::push.
• Added Vec::const_leak to leak the final value via intrinsics::const_make_global. If we see any pointer in the final value of a const that did not call const_make_global, we error as implemented in add const_make_global; err for const_allocate ptrs if didn't call #143595. (should we name the Vec method const_make_global as well?)

r? @rust-lang/wg-const-eval

To-do for me:

• Assess the rustdoc impact of additional bounds in the method
• Increase test coverage

[RalfJung]

We used to have the plan to have a dedicated compile-time allocator type... this is an interesting alternative. Given that it also involves the library surface, we should probably involve t-libs-api.

Why have you picked Vec as the first type for this? I think it'd make more sense for Box to go first since that is the most primitive type for heap allocations.

View changes since this review

[RalfJung]

It would be much more efficient to use write_bytes here.

[RalfJung]

This is conceptually quite different from leak which IMO should be reflected in the name, so I would indeed prefer const_make_global.

[RalfJung]

This needs a where T: Freeze for soundness.

[RalfJung]

It doesn't need to be done for all Vec created during compile-time -- only for those that you want to carry over to runtime. You can use intermediate Vec as scratch space or so during compile time that you never call this method on.

[fee1-dead]

I think it'd make more sense for Box to go first since that is the most primitive type for heap allocations.

That will either require us doing a const_eval_select to do box_new in runtime and something else in compile time, or to constify box_new. I chose Vec because I thought it would be much more useful than Box.

[oli-obk]

@bors try @rust-timer queue
This affects core pieces of vec and alloc, let's make sure everything gets optimized well

[rust-bors]

☀️ Try build successful (CI)
Build commit: b3f7ade (b3f7ade975cf0d5d02740da43e496064c977dfb4, parent: f977dfc388ea39c9886b7f8c49abce26e6918df6)

[rust-timer]

Finished benchmarking commit (b3f7ade): comparison URL.

Overall result: ❌✅ regressions and improvements - please read the text below

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.

Next Steps: If you can justify the regressions found in this try perf run, please do so in sufficient writing along with @rustbot label: +perf-regression-triaged. If not, please fix the regressions and do another perf run. If its results are neutral or positive, the label will be automatically removed.

@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)	0.3%	[0.3%, 0.3%]	1
Regressions ❌ (secondary)	0.2%	[0.1%, 0.2%]	3
Improvements ✅ (primary)	-0.6%	[-0.6%, -0.6%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-0.2%	[-0.6%, 0.3%]	2

Max RSS (memory usage)

Results (primary 0.4%, secondary -0.3%)

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

	mean	range	count
Regressions ❌ (primary)	2.7%	[1.3%, 4.5%]	5
Regressions ❌ (secondary)	3.7%	[3.7%, 3.7%]	1
Improvements ✅ (primary)	-2.0%	[-3.3%, -0.4%]	5
Improvements ✅ (secondary)	-2.3%	[-3.1%, -1.6%]	2
All ❌✅ (primary)	0.4%	[-3.3%, 4.5%]	10

Cycles

Results (primary 3.3%)

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

	mean	range	count
Regressions ❌ (primary)	3.3%	[3.3%, 3.3%]	1
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	3.3%	[3.3%, 3.3%]	1

Binary size

Results (primary 0.3%, secondary 0.6%)

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

	mean	range	count
Regressions ❌ (primary)	0.4%	[0.0%, 1.5%]	33
Regressions ❌ (secondary)	0.6%	[0.0%, 1.0%]	66
Improvements ✅ (primary)	-0.2%	[-0.5%, -0.0%]	5
Improvements ✅ (secondary)	-0.1%	[-0.1%, -0.1%]	1
All ❌✅ (primary)	0.3%	[-0.5%, 1.5%]	38

Bootstrap: 473.751s -> 473.82s (0.01%)
Artifact size: 390.43 MiB -> 390.55 MiB (0.03%)

[fee1-dead]

Yeah, rustdoc output is non-ideal. It shows all the where A: Allocator bounds. #148434 could probably make this better?