Skip to content

docs: close the react.dev reskin/orphaned-node gap (engine#84/#92) - #435

Merged
tannevaled merged 1 commit into
mainfrom
reskin-gap-close
Nov 15, 2026
Merged

tannevaled merged 1 commit into
mainfrom
reskin-gap-close

Conversation

@tannevaled

Copy link
Copy Markdown
Contributor

Summary

Surveyed FIDELITY.md's own "Known gaps" section fresh for the next confirmed-real-trigger item, bibliography-first rather than guessing. One bullet's own confirmed trigger — react.dev's settle-loop reskin guard failing to restyle nodes a rejected script pass had removed, first found (round 19, 2026-09-03) leaving the whole marketing page below the hero on a stale, washed-out light-mode style on a dark background — is the SAME page, and very plausibly the SAME underlying mechanism, round 38 (earlier today) already found no longer crashes at all.

Re-checked directly, not assumed from the adjacency alone: rendered the live production react.dev right now and counted real content. Round 19's own measurement was "after settle, react.dev's DOM has ZERO <h2> elements at all." Today: 15 <h2> elements, and both of round 19's own named missing sections ("Create user interfaces from components," "Write components with code and markup") are present in the real text.

The causal chain closes cleanly: the reskin guard only ever fires when a script pass is REJECTED (throws) and the settle loop falls back to the pre-failure layout. Round 38 already established react.dev's own hydration no longer throws at all. No rejected pass means the guard's own orphaned-node path is simply never reached on this page any more — not a separate fix, the same root cause's own downstream consequence resolving itself.

What this does NOT close: the general architectural gap itself — reskin() still cannot restyle a node a rejected pass removed, dom.RemoveChild still nulls .Parent at the moment of removal with no snapshot taken beforehand — remains exactly as true in the code as before. Only the one confirmed, cited, real-world trigger this bullet named is gone; the bullet is corrected to say so.

No other item surveyed in this pass had both a still-standing confirmed real trigger AND a tractable scope: rowspan/border-collapse explicitly have no trigger producing a visible difference today; subgrid, the animation-timeline gap, and min()/max() on width-like properties are each already flagged in their own entries as needing a materially larger architectural change; the oksvg/Go-Mono-font items are third-party dependency defects outside this engine's codebase.

Pure documentation, no code change.

Test plan

  • Verified live against production react.dev: 15 <h2> elements (not round 19's own zero), both named missing sections present in the real text
  • go build/go vet/gofmt -l/govulncheck ./... clean (nothing to run them against)

🤖 Generated with Claude Code

… no longer reproduces

Surveyed FIDELITY.md's Known Gaps section fresh for the next confirmed-
real-trigger item. One bullet's own trigger (react.dev's settle-loop
reskin guard failing to restyle nodes a rejected script pass removed,
first found leaving the whole page below the hero on a stale light-mode
style) is the same page, and the same underlying mechanism, round 38
already found no longer crashes at all.

Re-checked directly: round 19 measured "after settle, react.dev's DOM has
ZERO <h2> elements at all." Today: 15 <h2> elements, both of round 19's
own named missing sections present in the real text. The reskin guard's
orphaned-node path only ever fires on a REJECTED script pass; with
nothing rejecting any more (round 38), it is simply never reached.

The general architectural gap itself is untouched -- reskin() still
cannot restyle a node a rejected pass removed -- only this one confirmed,
cited trigger is gone.

Co-Authored-By: Claude Sonnet 5 <[email protected]>
@tannevaled
tannevaled merged commit 61bb952 into main Nov 15, 2026
13 checks passed
@tannevaled
tannevaled deleted the reskin-gap-close branch November 15, 2026 13:26
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant