Peak memory is 2× QuickJS: arena reservation, startup floor and representation

Open
#158 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Assessment

Difficulty
5/5
Estimated time
Over a week
Newbie friendliness
35/100
Issue type
Feature
Clarity
Mostly clear
Activity status
Active
Tech stack
rust

Research direction

Start by reviewing the heap reservation and collection logic, then measure the empty-script RSS floor and object-heavy benchmark cases named in the issue. The issue gives no file paths or test commands, so first locate the arena and startup allocation code and reproduce the reported measurements. Done means peak RSS is within 1.3× QuickJS across the 37 programs while runtime stays within 2% of today's.

Written by the indexing model from the issue text.

Description

Peak RSS of 0.3.0 is about twice QuickJS 2026-06-04's on the 37 fixed-work programs (median of 3, Ryzen 7 7840HS):

Suite Peak RSS, oxide / QuickJS
V8 v7 (8) 2.09
Octane 2.0 (15) 1.83
AWFY (14) 2.19
all 37 2.02

Time on the same programs is 1.03× QuickJS's, so the two together are the trade to improve. (CodeLoad, 9.6×, is a separate leak: #156.)

The collection threshold is not the cause

Halving --min-heap-mib from 4 (default) to 2, same binary, interleaved, 2 runs per program:

Suite time, 2 MiB / 4 MiB peak RSS, 2 MiB / 4 MiB RSS vs QuickJS at 2 MiB
V8 v7 1.006 0.979 2.04
Octane 0.985 0.958 1.75
AWFY 1.002 0.906 2.00
all 37 0.996 0.942 1.90

Time does not change (largest: NBody +4%); memory drops only 6%.

Where it goes: AWFY Bounce

Live data is 215 KiB. QuickJS peaks at 4.4 MiB.

--min-heap-mib collections heap reserved peak RSS
4 2 7,948 KiB 12.4 MiB
2 4 4,090 KiB 10.3 MiB
1 9 2,161 KiB 9.1 MiB
  1. Reserved is twice the threshold. At the 4 MiB threshold the heap reserves 7.9 MiB: object slots have 3.3× the capacity of the live count at the peak and value slots 3.3× (--memory-report: objects 20,679 live of 68,725, values 107,607 of 358,320). The threshold counts used bytes; the arenas grow by more than that, and the free slots are split by size class.
  2. About 7 MiB outside the heap even at a 1 MiB threshold: the empty-script floor is 5.1 MiB against QuickJS's 3.3 MiB (binary pages touched, the realm's builtins, atoms), plus compiler allocations (the stack-code intermediate of #154) that malloc keeps.
  3. Object representation in object-heavy programs: Splay 264.6 vs 144.0 MiB, Earley-Boyer 24.9 vs 14.6 MiB.

Direction

  • Count reserved bytes, not used bytes, toward the threshold, and grow arenas in steps bounded by it; return empty arena chunks after a collection.
  • Measure and cut the startup floor (lazy builtins, atom table sizing).
  • Release compile-time allocations (follows from #154).

Acceptance: peak RSS within 1.3× of QuickJS over the 37 programs (geometric mean) with time within 2% of today's.

Dominant language
Rust
Stars
61
Forks
2
PR merge metrics
No merged PRs in 30d

Getting set up

We have not checked this project's setup files yet. Start from its README, and see our first-contribution guide for the general steps.

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from pocket-stack/quickjs-oxide

All issues in pocket-stack/quickjs-oxide

Similar issues

More Rust issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.