test(xterm): upstream master baseline (post-#5817, pre-#5821)

[srid]

Purpose

Third prod arm in the #5821 A/B. This branch pins xterm.js + addon-webgl to the exact npm beta that was auto-published from upstream master seconds after our dispose-leaks PR (xtermjs/xterm.js#5817) merged — i.e. xterm master with #5817 landed, but without the open #5821 WeakRef fix.

This corresponds exactly to the "Unpatched" column in the local A/B. If the memory regression is real in production Chrome, this branch is where it reappears.

Pinned versions

\"@xterm/xterm\":        \"6.1.0-beta.199\",
\"@xterm/addon-webgl\":  \"0.20.0-beta.198\"

Fact-check — commit traceability

$ gh pr view 5817 --repo xtermjs/xterm.js --json mergeCommit
  → mergeCommit.oid = 0911688295fce79ba3b880e14af74f4f2718bef4

$ npm pack @xterm/xterm@6.1.0-beta.199 → package.json.commit
  → 0911688295fce79ba3b880e14af74f4f2718bef4   ✓

$ npm pack @xterm/addon-webgl@0.20.0-beta.198 → package.json.commit
  → 0911688295fce79ba3b880e14af74f4f2718bef4   ✓

All three SHAs match — the two npm betas are both built from the 0911688 merge commit of #5817. This is the same snapshot as xterm.js@master at the moment of writing (no commits after #5817 have landed).

Runtime verification of node_modules/@xterm/xterm/lib/xterm.mjs:

• _pausedResizeTask = this._register(new Qi(...)) — present → #5817 applied ✓
• new WeakRef(this) in _registerIntersectionObserver — absent → #5821 not applied ✓
• Plain strong-this IntersectionObserver callback — matches the pre-#5821 upstream code

The three-arm A/B on pureintent

Run the same 30× canvas-maximize toggle test across all three branches and compare Chrome Task Manager Memory Footprint Δ:

Branch	xterm state	Expected (per local + #617 priors)
this (test/xterm-upstream-master)	upstream master, pre-#5821	+367 MB / 30 toggles if #617 repros in prod
#652 (test/observer-nullify-variant)	master + observer = undefined	? — the experiment
master (shipped)	master + WeakRef	~0 MB / 30 toggles (per #617)

The full recipe — 7 terminals, DevTools console 30-cycle one-liner, fallback via Ctrl+Enter/Ctrl+D — is in the #652 description.

Interpreting the three outcomes

• This arm ≈ +367 MB, other two ≈ 0 MB → fix(xterm): patch IntersectionObserver retention (-367 MB/30 mode-toggles) #617 regression is real in prod Chrome, both WeakRef and observer-nullify defuse it. Tell #5821: ship the simpler variant, evidence attached.
• This arm ≈ +367 MB, test(xterm): A/B observer-nullify variant vs WeakRef (upstream #5821) #652 also ≈ +367 MB, master ≈ 0 MB → only WeakRef defuses the retention. Tell #5821: defend the WeakRef, evidence attached.
• All three ≈ 0 MB → fix(xterm): patch IntersectionObserver retention (-367 MB/30 mode-toggles) #617 is no longer reproducible in current pureintent state (Chrome update, extension change, something). Drop #5821 quietly or re-investigate under a fresh profile.
• This arm ≈ 0 MB, test(xterm): A/B observer-nullify variant vs WeakRef (upstream #5821) #652 ≈ 0 MB, master ≈ 0 MB → same as above.

Rollback

Not for merge. Close after data collected. The real decision lands as a change on #5821 (keep WeakRef, collapse to observer-nullify, or close).

Refs xtermjs/xterm.js#5817, xtermjs/xterm.js#5821, #617, #652