Fix in-place collect not reallocating when necessary

[the8472]

Regression introduced in #110353.
This was caught by miri

r? @saethlin

[the8472]

This is not part of the fix. the condition was already enforced before, I'm just making it more explicit for readability.

[RalfJung]

I can confirm that with this patch applied, the test suite passes in Miri.

[saethlin]

The current implementation in this PR looks like it is trying to enumerate the cases where reallocation is not permitted, then if none of those cases are hit we fall through to permitting allocation reuse. I would much prefer to see this implemented with the inverse strategy; try to enumerate the cases where allocation reuse is permitted, then fall through to requiring reallocation. Alignment is trivial though, so I don't mind handling that differently.

So I'd prefer to see something where either we early-return false and default to returning true.

That being said, I slapped together this alternative implementation and it seems to pass the test suite:

const fn needs_realloc<SRC, DEST>(src_cap: usize, dst_cap: usize) -> bool {
    // Alignment must match exactly. Beyond this check we do not need to reason about it.
    if const { mem::align_of::<SRC>() != mem::align_of::<DEST>() } {
        return true;
    }

    // Exact allocation size matches can be reused
    let src_bytes = src_cap * mem::size_of::<SRC>();
    let dest_bytes = dst_cap * mem::size_of::<DEST>();
    if src_bytes == dest_bytes {
        return false;
    }

    // Everything else needs a realloc
    true
}

This implementation seems much simpler to me and I'd prefer to see it (or something similar) used if it is equivalent. Is it? If it isn't, I think we're missing a test.

[the8472]

The difference is that I'm doing more const checks to find cases where we don't need to do runtime checks to figure out whether realloc is needed. I'd have to write codegen tests to catch that.
But sure, I can invert the condition.

[the8472]

I rewrote the needs_realloc conditions and also added a check for zero-sized allocations to catch up with #113113 which happened after I originally wrote the realloc code.

[saethlin]

Suggested change

	// If src size is an integer multiple of the destination size
	// the resulting capacity will also always be an integer multiple without remainder
	// If src size is an integer multiple of the destination size, we know the capacity we calculate
	// with src_cap * size_of::<SRC> / size_of::<DEST> does not have a remainder.

It took me a long time to understand this comment. It's referring to the division done outside this function. I've tried to make that clear.

r=me with this suggestion or equivalent clarification

[the8472]

@bors r=saethlin

[bors]

📌 Commit 13a843e has been approved by saethlin

It is now in the queue for this repository.

[RalfJung]

@bors p=1
this unblocks miri-test-libstd

[bors]

⌛ Testing commit 13a843e with merge 15bb3e2...

[bors]

☀️ Test successful - checks-actions
Approved by: saethlin
Pushing 15bb3e2 to master...

[rust-timer]

Finished benchmarking commit (15bb3e2): comparison URL.

Overall result: ❌ regressions - no action needed

@rustbot label: -perf-regression

Instruction count

This is a highly reliable metric that was used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	0.4%	[0.4%, 0.4%]	1
Regressions ❌ (secondary)	1.0%	[1.0%, 1.0%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.4%	[0.4%, 0.4%]	1

Max RSS (memory usage)

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	3.2%	[3.2%, 3.2%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Cycles

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-2.1%	[-2.1%, -2.1%]	1
All ❌✅ (primary)	-	-	0

Binary size

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	0.0%	[0.0%, 0.2%]	26
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.2%	[-0.2%, -0.2%]	2
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.0%	[-0.2%, 0.2%]	28

Bootstrap: 674.58s -> 675.083s (0.07%)
Artifact size: 314.17 MiB -> 314.18 MiB (0.00%)