fix: stabilize Hermes external memory tests

[mrousavy]

Summary

• resample Hermes external-memory measurements when process-wide GC activity contaminates a snapshot
• retain native ArrayBuffers across measurement attempts so Nitro allocations cannot be collected
• cover both HybridObject and ArrayBuffer accounting with the same strict helper
• throw with external-byte and GC-count diagnostics when Hermes instrumentation is unavailable or no exact sample is observed

Root cause

The failed iOS ASan job expected a 1 MiB increase but observed 1 MiB minus 1 KiB. The immediately preceding test creates a 1 KiB native ArrayBuffer and releases it. ASan changed timing enough for Hermes to collect that unrelated buffer between the global js_externalBytes snapshots. This was a test measurement race, not an AddressSanitizer violation or missing Nitro accounting.

Validation

• bun typecheck
• bun --cwd example lint-ci
• local iOS ASan build
• local full iOS ASan harness: 544/544
• local focused external-memory iOS ASan harness: 4/4, repeated 6 times
• hosted iOS ASan harness: 544/544, passed 3 consecutive executions
• hosted iOS TSan and Default harnesses: 544/544
• complete PR matrix green, including Android ASan and Default
• original failing ASan job passed when rerun unchanged, confirming the timing-dependent flake

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
nitro-docs	Skipped		Aug 25, 2026 9:35pm

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 68fbf4699a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Reject samples that span garbage collection

When Hermes runs a GC between these snapshots, an exact net delta is still contaminated: the immediately preceding ArrayBuffer test leaves an eligible 1 KiB native buffer, so a new allocation over-reported by 1 KiB could coincide with that buffer's collection and produce the expected delta. Because garbageCollections is only logged, this condition would accept the sample and mask the accounting regression; require both the exact delta and zero intervening collections, otherwise retry.

Useful? React with 👍 / 👎.