Skip to content

fix(apps): show why a refused builtin page is empty, not a chat redirect - #13004

Merged
bolichen97 merged 1 commit into
mainfrom
fix/seam-refusal-visible
Sep 23, 2026
Merged

bolichen97 merged 1 commit into
mainfrom
fix/seam-refusal-visible

Conversation

@bolichen97

@bolichen97 bolichen97 commented Sep 23, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

You click a page in the sidebar and land on chat. No error, no message. The page
you asked for is simply gone.

That is what a refused page registration looks like today. registerBuiltinComponents
refuses an entry it cannot use — a route that can never resolve, or a missing or
malformed appId. In dev and test it throws, so the refusal is loud. In a
production bundle it is one console.warn, and then BuiltinAppRoute treats the
route like any unknown path and redirects to /chat.

A downstream edition hit this: it registered ten builtin pages the old way, with
a bare component and no appId. All ten were refused. All ten redirected to chat.
Nobody saw a warning, because nobody has a production console open.

Why it matters

The warn-in-production policy is right, and this PR keeps it. Throwing at
registration would take the whole dashboard down over one bad page, which is worse
than the page being missing.

But degrading safe is not the same as degrading silently. A downstream edition
compiles its own bundle, so the dev/test throw never runs over its registrations.
The production warn is the only signal it ever gets, and a warn in a shipped app
reaches nobody. The result is a page that vanishes with no trace and no way for the
person looking at the empty space to find out why.

What changed (motivation → approach → change)

Symptom: a refused page is indistinguishable from a page that does not exist.
Root cause: the refusal is reported and then discarded, so by the time the user
navigates there is nothing left to read. Fix: keep the refusal, and show it where
the user meets it.

Before After
Registration warn, ignore entry warn, ignore entry, record reason
Navigating there redirect to /chat the page shows the reason
Unknown route redirect to /chat redirect to /chat
Dev / test throws throws
One page crashes, you open another stale card follows you each page gets its own boundary

website/src/apps/seamCollision.ts keeps its two-argument shape; the record lives in
website/src/apps/builtinRegistry.ts, next to the miss path that reads it, because
only a seam with somewhere to show a refusal benefits from remembering one. A
reportRefusal helper records then reports, making the ordering structural.

The key is the one the router will ask for, not the string the entry registered: nothing
can ask for /reports/daily, so a reason filed there is unreadable. Deriving it is
decoding, not slicing, because % is outside the route pattern so every escaped route is
refused. It mirrors react-router's two passes — cut query and hash on a literal ?/#,
decode the pathname's segments under one try/catch, then apply matchPath's %2F pass —
so /reports%2Ddaily is asked for as /reports-daily and %252F resolves to /. A
route with no leading / gets no key: no URL produces that spelling. A duplicate gets
none either — it still resolves to the winning entry.

website/src/apps/BuiltinAppRoute.tsx splits the miss path in two. An unknown path
is a typo or a stale link and still goes to chat. A path whose registration was
refused renders a panel supplied as the boundary's fallback: it names the page,
says in plain words that the build has to be corrected and retrying cannot help, and
carries the registration diagnostic under a label for whoever assembled the build. It
offers the agent hand-off and no Try Again, which on a refusal could never succeed.
The throw is kept because fallback overrides only what renders, so componentDidCatch
still journals to recordError and RUM. Both of
that component's boundaries are keyed by the resolved path, because App.tsx serves
every builtin page from one unkeyed <Route path="/:builtinApp/*">: a
builtin-to-builtin move is a param change on a mounted component, and ErrorBoundary
returns its fallback before it reads children with no reset on a prop change. Without
the key, one caught throw followed the user onto every builtin page they opened next
until a reload. Keying by the resolved segment rather than the full URL is what keeps
an app's own sub-path navigation from remounting the page on every step.

Its three strings are new apps.builtinAppRoute keys, translated in every catalogue
under website/src/i18n/locales/ with notes in website/src/i18n/en.context.json;
en-XA.json is regenerated.

website/docs/extension-seams.md, where website/AGENTS.md routes seam work, said a
non-conforming route redirects to chat. It now states that remembering a refusal
belongs to the seam with somewhere to show it, how the key is derived, and why the
refused page drops Try Again while keeping the throw.

Not a product-shape change: behaviour differs only on a refused route or a crashed
page, both defect states.

Backwards compatibility

Compatible: reportSeamCollision keeps its two-argument signature, so all ~23
existing calls are untouched and the throw/warn policy is unchanged; the refusal
record lives in builtinRegistry.ts, the one seam with a surface to show it on. In this
repository's own builds no registration is ever refused — the core table is
compile-time code held to isValidAppId by builtinRegistry.identity.test.tsx — so
no route reachable in a stock build changes. The key added to the entry branch's
boundary remounts a page only on a route change, which already unmounted it.

Tests

website/src/test/builtinRegistry.refusalVisible.test.tsx, 25 cases, stubbing
import.meta.env.DEV to false because production is the only build where a refusal
survives registration:

  • a missing appId and a malformed route each record their reason, and still do not register
  • a malformed route records under the segment the router resolves, and the message still names the full route
  • query-, hash- and trailing-slash routes likewise
  • a slashless route gets no invented key
  • a percent-escaped route shows its reason at the segment the router decodes to
  • an encoded slash stays inside the one segment, and a double-encoded one resolves through both passes
  • the query and hash are cut on a literal ?/#, not a decoded one
  • a malformed escape is left alone and does not take the registrar down, including when it sits in a later segment
  • the refused panel drops Try Again and keeps the agent hand-off
  • an ordinary page crash still offers Try Again — the control that scopes the override to the refusal branch
  • the refusal is still journaled, so the hand-off carries its context
  • the first reason survives a second refusal of the same route
  • a report with no key records nothing (the duplicate branch depends on this)
  • a duplicate route records nothing and keeps the first component
  • a refused route renders the reason and does not land on chat
  • leaving a refused page for a working one renders the working page
  • leaving a crashed page for a working one renders the working page
  • a page stays mounted across its own sub-path navigation
  • a route nobody registered still redirects to chat — the negative control that stops "render the reason" from becoming "never redirect"
  • the dev/test build still throws at registration, and records the same reason

Navigation runs through the real router via a useNavigate harness inside one
MemoryRouter; a mocked navigate or changed initialEntries would not reproduce the
param-change condition the boundary defect needs, and five of the decoding cases
assert end-to-end through the router rather than through seamRefusalReason. Every
new case was verified failing against the pre-fix code. Five negative controls each
turn exactly one test red: removing the entry branch's key, dropping the %2F pass,
decoding per segment instead of per pathname, cutting the query after decoding, and
reverting the fallback prop. website/src/test/BuiltinAppRoute.test.tsx gains the new
registry export to its total mock, which localizing the store made incomplete.

Manual verification

Both shots come from this repo's own screenshot harness
(website/scripts/lib/serve-dist.mjs and website/scripts/lib/boot-api.mjs): it
serves the real built SPA and answers every /api/** call from the shared boot
fixtures, so the shell boots without a gateway. Two production builds of this same
tree, one at this head and one at origin/main, each composed with one throwaway
edition that registers /aim the old way (bare component, no appId). Same
viewport both times, so the pair differs only in the branch.

Screenshots / video

Before, on origin/main — /aim is refused and the user silently lands on chat:

Dashboard on the chat page, showing "What can I do for you?" and a "Start a new chat" button, after the user asked for /aim

After, on this branch — /aim stays put and names the reason:

Dashboard at /aim, sidebar and top bar intact, showing the heading "/aim could not load", a plain explanation that Kiro Crew rejected the page when it started because this copy sets it up wrongly and retrying will not bring it back, the registration diagnostic under "Details for whoever assembled this copy", a line reading "Asking the agent opens a chat with these details already attached", and one "Ask the agent" button

Related Issues

no linked issue: found while tracing why an edition's ten pages bounced to chat.

Pattern harvest

Rule candidate: review-prompt
Pattern: a refusal reported into a log and then discarded, leaving the user-facing
miss path unable to tell "never registered" from "registered and rejected".

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

@bolichen97
bolichen97 requested review from a team and pepmach September 23, 2026 05:06
@bolichen97

Copy link
Copy Markdown
Contributor Author

Intent: A page whose registration the core refused must say why it is empty, instead of silently redirecting the user to chat.
Not a goal: Making a refusal hard-fail at registration (that takes the whole dashboard down over one bad page), changing the throw-in-dev / warn-in-prod policy, or adding this visibility to the other extension seams.

@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Sep 23, 2026
@github-actions

github-actions Bot commented Sep 23, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Fable 5) — ✅ PASS

Design-level review of 5ecb131642a1c77ef127925f1ab6593b236caa80 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Verify builtinRouteKey's mirrored decode against the react-router version actually pinned — drift on upgrade silently reverts escaped-route refusals to the chat redirect.

[DESIGN-REVIEWED] 5ecb131

@github-actions

github-actions Bot commented Sep 23, 2026 •

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of 5ecb131642a1c77ef127925f1ab6593b236caa80 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 5ecb131

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 5ecb131642a1c77ef127925f1ab6593b236caa80: <one-sentence reason>

@github-actions

github-actions Bot commented Sep 23, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Fable 5) — ✅ PASS

Premise-level review of 5ecb131642a1c77ef127925f1ab6593b236caa80 — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

First-Principles-Verdict: PASS

Verify the description's runtime claims the reviewer cannot: that every new test fails on base and the five named negative controls each turn one test red.

What this change ships

Inventory (9 items) — 9 justified

Intent: make a refused builtin-page registration readable where the user meets it, instead of a silent chat redirect — a FIX.

  1. A refused route renders why it is empty instead of redirecting to chat — justified
  2. Refusal reasons are recorded at registration, keyed by the path the router will ask for — justified
  3. The refused page drops Try Again, keeps the agent hand-off, states in-page what it does — justified
  4. A crashed builtin page no longer follows the user onto the next one (key={path} on both boundaries) — justified
  5. New export builtinRefusalReason — justified
  6. Four new strings translated across all 12 catalogues plus context sidecar — justified
  7. extension-seams.md documents the record, key derivation, and no-retry rationale — justified
  8. seamCollision.ts gains explanatory comment only, no behavior — justified
  9. New 25-case test file; total mock gains the new export — justified

Facts checked: ErrorBoundary renders fallback via 'fallback' in this.props with no reset on prop change and componentDidCatch unaffected (ErrorBoundary.tsx:76); one unkeyed <Route path="/:builtinApp/*"> in App.tsx:4841; findReport matches by exact message; ^report[A-Z]\w*$ i18n callee exemption exists (eslint.i18n.config.js:1031); cache retention is per-client, so the added key's remount costs nothing (appQuery.ts:56); defect provenance is the base code's unconditional Navigate to="/chat" plus tests that cannot pass on base; the sibling-seam count (theme/autolink/renderer refusals leave core surface rendered) is argued and recorded in the doc in the same commit.

[FIRST-PRINCIPLES-REVIEWED] 5ecb131

@github-actions

github-actions Bot commented Sep 23, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Fable 5) — ✅ PASS

UX-level review of 5ecb131642a1c77ef127925f1ab6593b236caa80 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

UX-Verdict: PASS

Nothing to check.

[UX-REVIEWED] 5ecb131

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Sep 23, 2026
@github-actions

github-actions Bot commented Sep 23, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5 Review — ✅ no blocking findings

Reviewed 5ecb131642a1c77ef127925f1ab6593b236caa80 — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 5ecb131

Verdict parsed from the review's SHA-scoped output markers for commit 5ecb131642a1c77ef127925f1ab6593b236caa80.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 5ecb131642a1c77ef127925f1ab6593b236caa80: <one-sentence reason>

@bolichen97
bolichen97 force-pushed the fix/seam-refusal-visible branch from dafe946 to 21afa18 Compare September 23, 2026 05:42
@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: action required A blocking check or review needs attention labels Sep 23, 2026
@bolichen97

Copy link
Copy Markdown
Contributor Author

span=14476f68e3fe
self-added: yes
mechanism: both ErrorBoundary returns in BuiltinAppRoute are now keyed by the resolved path

  • fixed: refused-route boundary remained crashed after navigation

Confirmed first-hand, not taken on trust. App.tsx serves all builtin pages from ONE unkeyed
<Route path="/:builtinApp/*">, so builtin-to-builtin is a param change on a mounted component;
both branches return ErrorBoundary as their root at the same position, and its render()
returns the fallback before reading children with no reset on a prop change.
Took your one key prop over the alternative fix of rendering the reason directly: that one is
bigger and loses recordError, the RUM react_error event and the agent hand-off.
Keyed BOTH boundaries, so the same mechanism is also closed for an ordinary page crash.
Keyed by the resolved segment, not the full URL, so intra-app nav does not remount.
Fixed in 21afa18.

@bolichen97

Copy link
Copy Markdown
Contributor Author

span=326aeca1cd28
self-added: yes
mechanism: builtinRouteKey() normalises a refusal key to the leading path segment

  • fixed: a nested route recorded a key the reader can never ask for

Correct, and it made this PR's own claim false for one of its two refusal arms.
The route-shape arm filed the refusal under the raw registration string while the reader asks
for `/${builtinApp}`, so /reports/daily, /reports?daily and /reports#x all still
fell through to the chat redirect as silently as before the record existed.
Narrower normaliser than a synthesising one: /^\/[^/?#]*/, no leading / invented for a
slashless registration, because no URL produces that spelling and inventing /reports for
reports would let one key's refusal answer for a different key.
Applied to both arms so reordering them cannot silently reintroduce it.
Fixed in 21afa18.

@bolichen97

Copy link
Copy Markdown
Contributor Author

span=b4a2654e5344
self-added: yes
mechanism: both ErrorBoundary returns in BuiltinAppRoute are now keyed by the resolved path

  • fixed: the refusal branch poisoned the boundary every builtin page renders through

Mechanism confirmed exactly as described, including ErrorBoundary.render() returning the
fallback before it reads children.
Chose your second option (a key on both boundaries) over rendering the reason directly: the
direct card drops recordError, the RUM event and the agent hand-off, which is the value this
PR routes through the boundary for, and it would leave the entry branch unfixed.
Your framing found a sibling instance: an ordinary page crash armed the same shared instance on
origin/main too, so a crash on one page showed its card on the next. Both are keyed now.
Negative control: with only the refusal branch keyed, the crash case fails and the refusal case
still passes, so the second key is load-bearing.
Fixed in 21afa18.

@bolichen97

Copy link
Copy Markdown
Contributor Author

span=621b4f938c54
self-added: yes
mechanism: builtinRouteKey() normalises a refusal key to the leading path segment

  • fixed: the route-shape refusal was keyed by the raw registration string

Correct, and it left exactly the class this registrar refuses most as silent as before.
Took a narrower normaliser than the one proposed: /^\/[^/?#]*/ via exec, which needs no new
string literal (CI's i18n gate is zero-tolerance on literals added on lines this branch wrote)
and deliberately does NOT synthesise a leading /. A slashless registration stays unnormalised:
no URL produces that spelling, so there is no empty page asking to be explained, and minting
/reports for reports would let one key's refusal answer for a different key.
Asserted at render as you asked, not only through seamRefusalReason: the multi-segment, query,
hash and trailing-slash cases each render the reason in a real-router test.
Fixed in 21afa18.

@bolichen97

Copy link
Copy Markdown
Contributor Author

span=b2342b2349e0
self-added: yes
mechanism: extension-seams.md now documents the third outcome and which refusals carry a key

  • fixed: the seam doc still described a two-outcome policy

Verified the routing claim before acting: website/AGENTS.md does send registry/seam work to
website/docs/extension-seams.md, and two of its sections were factually wrong about shipped
behaviour, not merely thin.
"Collision policy" now carries the third outcome and its reader seamRefusalReason, plus why a
refusal is still not thrown at registration. "Per-seam validation / Builtin routes" no longer
says these routes redirect to chat, and states that the key is the leading segment the router
would have asked for rather than the registered string.
Also named why duplicates and the other seams deliberately pass no key, so the next registrar
has a documented rule instead of a precedent to copy.
The same two stale claims had drifted into the registrar docstring; corrected there too.
Fixed in 21afa18.

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Sep 23, 2026
@bolichen97
bolichen97 force-pushed the fix/seam-refusal-visible branch from 21afa18 to d5195e0 Compare September 23, 2026 06:17
@bolichen97

Copy link
Copy Markdown
Contributor Author

span=326aeca1cd28
self-added: yes
mechanism: builtinRouteKey() now mirrors react-router's decode pipeline instead of slicing

  • fixed: the refusal key stored percent escapes verbatim

Legitimate and inside this PR's class, not beside it: % is outside _BUILTIN_ROUTE_RE, so
every escaped route is refused and the recorded reason is the ONLY thing a user could be shown.
Did not take the proposed fix. Probed react-router 7.18.2 with matchRoutes rather than reading:
the param comes from TWO passes, decodePath over the whole pathname under ONE try/catch, then
matchPath's %2F pass on the captured group. A plain decodeURIComponent on the segment gets
%252F wrong (one pass short) and a later-segment %ZZ wrong (decodes where the router does not).
Mirrored the pipeline instead: cut query/hash on a LITERAL ?/#, decode segments under one
try/catch, apply the %2F pass. Order matters — cutting after decoding collapses /x%3Fy.
8 tests, 5 asserting end-to-end through the real router; 3 negative controls each redden exactly
one. Checked the invariant has no sibling: only this file passes a key, the other 14 call sites
are 2-arg. Fixed in d5195e0.

@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Sep 23, 2026
@bolichen97
bolichen97 force-pushed the fix/seam-refusal-visible branch from d5195e0 to 69c036f Compare September 23, 2026 07:30
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Sep 23, 2026
A refused page registration was only a console.warn in a production bundle,
and the route then fell through BuiltinAppRoute's miss path to /chat. A
downstream edition shipped ten pages refused for a missing appId; clicking
any of them landed on chat with no signal a user could see.

The registry now remembers why, under the key the router will ask for, and the
route renders a panel naming the page, saying in plain words that retrying
cannot help, and carrying the registration diagnostic under a label for
whoever assembled the build. The throw is kept so the boundary still journals
to recordError and RUM; both of that component's boundaries are keyed by the
resolved path, because App.tsx serves all builtin pages from one unkeyed route
and without a key a single caught throw followed the user onto every builtin
page they opened next. Registration still degrades safe -- throwing there would
take the whole dashboard down over one bad entry.
@bolichen97
bolichen97 force-pushed the fix/seam-refusal-visible branch from a6f8aa7 to 5ecb131 Compare September 23, 2026 10:21
@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: action required A blocking check or review needs attention labels Sep 23, 2026
@bolichen97

Copy link
Copy Markdown
Contributor Author

span=9ab233ccef75
self-added: no
mechanism: "switched off" names a control the product really has, and it is not this one

  • fixed, with wording that differs from the suggestion

The concern is right and sharper than "vague": the Apps page has a real per-app on/off switch, so a
reader told this page was "switched off" would go find that control, and it has nothing to do with
this state. That is a concrete wrong turn.

The suggested "failed to load" was not taken, because a load that was attempted and went wrong is
exactly what invites a retry — one line above "Retrying will not bring it back". What happens is
narrower: registerBuiltinComponents validates each entry at startup, finds this one malformed, and
refuses to register it. The page was never loadable in this build, and the decision is immutable for
the life of the page load.

New English: "{{productName}} rejected this page when it started, because it is set up wrongly in this
copy. Retrying will not bring it back — whoever assembled this copy has to fix it." "Rejected ... when
it started" is the registration refusal, so it agrees with the diagnostic block below it on the same
page ("refused at registration"): the words the reader sees match the words they will paste. Both
mandated sentences survive.

Fixed in 5ecb131. All 13 locales retranslated, en-XA regenerated by its generator rather than
hand-edited, the translator-context entry rewritten to the new sense, and the test assertion moved off
the old words. The re-captured screenshot in the description is this wording.

One line beyond the finding rides with it, and it is this PR's own defect: the German value of the
sibling key this PR added last round addressed the reader as Sie, which deStyle.test.ts rejects at
zero tolerance. Reverting the German file to the previous head and re-running that test reproduces the
failure, so the branch was red on it independently of this reword.

@bolichen97

Copy link
Copy Markdown
Contributor Author

span=9b96a5f4ab4b
self-added: no
mechanism: the attachment upload had not finished propagating when the lane fetched it

  • fixed: the evidence is re-captured and confirmed downloadable

The lane was right that it had no usable image, and the cause was on this side, not in the lane. The
attachment it reported as HTTP 404 was a real upload of the real panel; gh pr edit --attach returns a
user-attachments URL immediately, but the object behind it was not yet fetchable when the lane read
it about 30 minutes later. The same URL served 200 and the exact local byte count later on, so this was
propagation latency, not a missing or mistyped asset.

Rather than re-run the lane against an image that no longer matched, the panel was re-captured, because
the copy changed in the same round (span=9ab233ccef75) and the earlier shot had stopped being what this
branch ships. The capture is a real production build: vite build against a throwaway edition that
registers /aim the old way (bare component, no appId), served on loopback, /aim opened in a
browser. The accessibility tree from that page was checked against the source strings before the shot
was taken, so the image is not the only evidence that the panel renders.

Attached and verified at 5ecb131: the URL in the description resolves and downloads as an image of
the exact byte size of the local capture. Its alt text names the heading, the reworded lead, the
diagnostic label, the hand-off note and the single button, so a reader who cannot load the image still
learns what the surface says.

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running readiness: passed Eligible automated validation passed for the current revision and removed readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention labels Sep 23, 2026
@bolichen97
bolichen97 enabled auto-merge (squash) September 23, 2026 18:32
@bolichen97
bolichen97 merged commit 5c23469 into main Sep 23, 2026
107 of 115 checks passed
@bolichen97
bolichen97 deleted the fix/seam-refusal-visible branch September 23, 2026 18:41
@github-actions github-actions Bot removed the readiness: passed Eligible automated validation passed for the current revision label Sep 23, 2026
wknight1 pushed a commit to wknight1/KiroCrew that referenced this pull request Oct 5, 2026
…ect (kirodotdev#13004)

A refused page registration was only a console.warn in a production bundle,
and the route then fell through BuiltinAppRoute's miss path to /chat. A
downstream edition shipped ten pages refused for a missing appId; clicking
any of them landed on chat with no signal a user could see.

The registry now remembers why, under the key the router will ask for, and the
route renders a panel naming the page, saying in plain words that retrying
cannot help, and carrying the registration diagnostic under a label for
whoever assembled the build. The throw is kept so the boundary still journals
to recordError and RUM; both of that component's boundaries are keyed by the
resolved path, because App.tsx serves all builtin pages from one unkeyed route
and without a key a single caught throw followed the user onto every builtin
page they opened next. Registration still degrades safe -- throwing there would
take the whole dashboard down over one bad entry.
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.

2 participants