• feat(track-memory-allocations): add total heap allocated bytes #24543 : 2 dependent PRs (#24539 , #24550 ) 👈 (View in Graphite)
• feat(allocator): record byte size of arena chunk allocations #24554
• feat(track-memory-allocations): add arena size #24553
• main

How to use the Graphite Merge Queue

Add either label to this PR to merge it via the merge queue:

• 0-merge - adds this PR to the back of the merge queue
• hotfix - for urgent changes, fast-track this PR to the front of the merge queue

You must have a Graphite account in order to use the merge queue. Sign up using this link.

This stack of pull requests is managed by Graphite. Learn more about stacking.

Merging this PR will not alter performance

✅ 62 untouched benchmarks
⏩ 9 skipped benchmarks¹

_{Comparing allocs-snap-bytes (f83d4e0) with main (246b426)²}

@camchenry tried this before, but problem was that the numbers aren't consistent across runs.

I narrowed down at least one source of the problem: #23728 (comment)

As it happens, I was talking to Claude about this last week.

The indeterminism is likely caused by Address Space Layout Randomization (ASLR).

On Linux, you can disable ASLR with setarch $(uname -m) -R ./binary, so we could probably get consistent numbers in CI. But Mac has no option to disable ASLR, so it wouldn't work on local dev machines.

One way to tackle it cross-platform is to shim the system allocator. Reserve a large chunk of address space from system at a fixed address with mmap, then pass that to Mimalloc to use as its heap. That would likely solve the problem (and also likely remove a lot of variance from our benchmarks), but it's not a trivial thing to implement.

Arena bytes should be consistent across platforms and runs for parser and semantic - I believe they're purely deterministic. But transformer introduces indeterminism, which also then affects minifier (since it has to run after transform). Not sure about formatter.

Sys alloc bytes may not be stable on any component (even parser).