Preserve card menu-item subclass during cached swaps - #3093
Merged
steipete merged 3 commits intoAug 20, 2026
Merged
Conversation
|
🦞👀 Pull request received. I will update this pull request when review starts. |
clawsweeper
Bot
added
P2
|
Codex review: needs maintainer review before merge. Reviewed August 20, 2026, 4:02 AM ET / 08:02 UTC. ClawSweeper reviewWhat this changesThe PR preserves the dynamic NSMenuItem subclass during cached provider-content swaps, with switch-back regression coverage and native-menu proof. Regression provenancePossible regression — probable (reviewed change; known regression link). No predecessor PR is attributed. Merge readinessThis PR remains necessary: current main swaps cached contents across unequal NSMenuItem classes, while card rows rely on their subclass to suppress native highlighting. The patch is focused, regression-covered, and supported by native-menu evidence; keep it open for normal maintainer review and exact-head checks. Priority: P2 Review scores
Verification
Live VerificationCommand: Result: FAIL (failed) — execution before step 1 Assertions:
How this fits togetherCodexBar renders provider and overview content into a native macOS status menu. Cached provider switches reconcile existing rows, so the row shell must retain card-specific behavior while its displayed content changes. flowchart LR
A[Provider and overview content] --> B[Cached menu rows]
B --> C[Cached menu reconciliation]
C --> D{Row classes match?}
D -->|Yes| E[Swap row contents]
D -->|No| F[Replace row shell]
E --> G[Native status menu]
F --> G
Before merge
Agent review detailsSecurityNone. Review metrics
Root-cause clusterRelationship: Members:
Proposal only: this assessment does not dispatch repair, suppress jobs, mutate sibling items, close, or merge anything. Merge-risk optionsMaintainer options:
Technical reviewBest possible solution: Merge the narrow class-compatibility guard after normal maintainer review, retaining payload-only swaps for equal-class rows and shell replacement only for unequal-class rows. Do we have a high-confidence way to reproduce the issue? Yes. Current main's cached-swap path can exchange content between unequal NSMenuItem classes, while the card subclass is the mechanism that disables native highlighting. Is this the best way to solve the issue? Yes. Replacing only unequal-class shells directly preserves the card override while leaving the existing flash-free swap path for compatible rows intact. AGENTS.md: found and applied where relevant. Codex review notes: model internal, reasoning high; reviewed against 84a4ca725d5d. LabelsLabel justifications:
EvidenceWhat I checked:
Likely related people:
Rating scale
Overall follows the weaker of proof and patch quality. Workflow
History |
kiranmagic7
force-pushed
the
kiran/preserve-menu-item-subclasses-3091
branch
from
August 20, 2026 07:01
45ab5d8 to
3e7ddf4
Compare
clawsweeper
Bot
added
proof: 📸 screenshot
steipete
added a commit
that referenced
this pull request
Aug 20, 2026
steipete
added a commit
that referenced
this pull request
Aug 20, 2026
* feat(kiro): show overage credits against their cap kiro-cli /usage states plan credits alone and omits the overage section for organization accounts, so a spent plan looks like the account is out. Read GetUsageLimits with the CLI's own token (read-only) and surface overage as a second credit window plus charges against the overage budget. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(kiro): reject impossible usage counters from GetUsageLimits An overage larger than total usage would clamp planUsed to zero and overwrite valid CLI numbers. Honor API-disabled overage over a stale CLI Enabled line so enrichment cannot resurrect a cap the API says is off. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(kiro): reject GetUsageLimits payloads that exceed the plan Drop enrichment when plan usage is above the plan ceiling, and treat an unrecognized overage status as unknown so the CLI overage line can still stand. Co-authored-by: Cursor <cursoragent@cursor.com> * fix: import FoundationNetworking for Linux URLSession types * fix(kiro): resolve the CLI state database on Linux GetUsageLimits enrichment always looked under macOS Application Support, so Linux refreshes never found data.sqlite3. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(kiro): reject overage above cap and honor API currency Best-effort GetUsageLimits data should not present a spend window above its ceiling or format non-USD charges as dollars. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(kiro): propagate cancellation from usage-limits enrichment A cancelled GetUsageLimits call now fails the refresh instead of publishing a CLI-only snapshot as success. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(kiro): drop CLI USD overage estimate when the API currency is not USD A missing overageCharges field no longer lets a dollar CLI fallback render in the API's non-USD currency. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(kiro): add locale entries for overage copy Translate the new Overage window title and Overage credits left detail row in every app catalog so non-English UIs no longer fall back to English. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(kiro): honor API-disabled overage and localize cap phrases GetUsageLimits DISABLED now replaces a stale CLI Enabled status, and Kiro "of N" detail values go through the of %@ localization key. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(i18n): expect localized OpenRouter API key details in zh-Hant Provider detail titles and catalogued labels go through L after #3084, so the popup test must assert the Traditional Chinese strings. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(kiro): clamp overage that runs slightly past the cap Rejecting over-cap counters dropped the whole GetUsageLimits payload and hid overage on organization accounts, so the gauge now clamps to the cap instead. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(kiro): keep over-cap overage and localize credit units Clamping hid the amount Kiro actually billed, and overage usage still rendered the English "credits" suffix in other locales. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(kiro): keep CLI overage when API omits the cap ENABLED without overageCapWithPrecision is incomplete enrichment, not a disabled account. * fix(kiro): mark API-enabled overage when the CLI omits it Organization accounts skip the CLI overage section, so the menu-bar modes never saw an Enabled row. * fix(kiro): keep CLI plan usage when API bonuses are present GetUsageLimits folds bonus spend into currentUsage, so overwriting the plan gauge would double-count bonus credits. * fix(kiro): parse bonus-inclusive usage above the plan limit GetUsageLimits folds bonus spend into currentUsage, so rejecting planUsed > planLimit dropped overage enrichment for those accounts. * fix: preserve stacked menu bar layout line breaks (#3094) * Preserve card menu-item subclass during cached swaps (#3093) * Preserve menu item subclasses during cached swaps * Add native menu proof for cached shell swaps --------- Co-authored-by: Kiran Magic <262980978+kiranmagic7@users.noreply.github.com> Co-authored-by: Peter Steinberger <steipete@gmail.com> * docs: changelog for #3093 * Extend menu bar conditionals beyond usage percentages (#3088) * Extend menu bar conditionals beyond usage percentages Conditional predicates could only compare four percent-used windows. They now compare 18 metrics across four units: percent windows, the direct primary/secondary/tertiary lanes, four reset countdowns, three pace deltas, run-out, credit balance, and today/30-day cost. Metrics with two readings (percent windows, lanes, balance) gain a used/remaining select, so "session > 50% used and session resets in < 2h" is expressible. Pace, run-out, balance and cost were only carried as display strings, which cannot be compared, so MenuBarLayoutRenderMetrics carries their numeric twins pre-rounded to the same granularity as the text they mirror. Three refresh gates needed widening for the new data dependencies: - The title cache key had no component that moves with the clock, so a countdown predicate would have served its pre-flip title indefinitely. It now keys on the per-conditional outcome, evaluated once per render. - The four observation signatures gated on display tokens; a predicate on cost or balance has no token. They now also read the conditionals' metrics, which additionally fixes lane tokens inside conditional branches being invisible to the lane signature. - A reset-countdown predicate flips at an instant nothing else ticks on, so the countdown scheduler wakes at `resetsAt - threshold`. The conditional library is now decoded element-wise: this change makes forward-incompatible metric values possible for the first time, and one unknown value would otherwise have wiped the whole library on a downgrade. Ships an "Auto % / Resets in" default that renders the automatic percentage while the lane has headroom and the reset countdown once it is spent. * Sign the readings conditional predicates actually compare Three observation-signature gaps let a predicate flip without a redraw: - Cost signatures recorded only the currency-formatted string, so two token-cost updates could cross a threshold while both formatted to the same cent. A referenced cost metric now signs the unrounded amount losslessly. - The balance signature recorded only the rendered "Remaining" row, so a `balance used` predicate — which reads the "Used" row no token surfaces — was entirely unsigned. Both amounts are now signed. - The lane signature recorded the displayed reading, which follows `usageBarsShowUsed` and clamps remaining at zero, while `RateWindow.usedPercent` deliberately preserves over-quota values. A used-direction predicate such as `primaryLane > 105%` could move 104% -> 106% against a constant `0.000`. The lane signature is now scoped to what the layout renders, and a new conditional-window signature covers what conditionals read: the raw used percent (which remaining derives from, so it covers both directions) plus `resetsAt`, which countdown predicates depend on and no display token contributes. * Tick clock-derived predicates that no token schedules `menuBarWeeklyPaceRefreshDelays` is gated on a placed `.pace(.weekly)` token and only wakes once, at the pace-eligibility boundary. Excluding `runsOutIn` from the conditional reset schedule on the assumption that scheduler covered it therefore left a hole: a layout whose only pace or run-out reference is a predicate got no clock wake-up at all, so it kept rendering the branch that was true when the value last moved. Referenced weekly-pace predicates now also trigger the eligibility wake-up, and any referenced pace or run-out predicate schedules a minute tick. Both numbers are pre-rounded to the granularity the menu bar shows -- whole percentage points and whole minutes -- so a minute tick is exactly enough, and it is the cadence a `.resetCountdown` token already costs. Money predicates deliberately schedule nothing: they move only when new provider data arrives. * Keep older releases' conditional libraries readable on downgrade Decoding the library element-wise only helps builds that already have the lenient decoder. The build a user actually downgrades to decodes `menuBarLayoutConditionals` strictly and falls back to `[]`, so one saved rule using a new metric would empty the entire library there. The conditional library now dual-writes the way layouts already do: `menuBarLayoutConditionalsV2` keeps full fidelity, and the original key keeps an older-readable projection. `loadLibrary` mirrors `preferredLayout` — the current key wins unless the legacy key disagrees with its own projection, which only happens when an older release wrote it, and that edit must survive. The projection drops an entry when any clause uses a metric outside the original four, and also when any clause uses a non-`.used` direction. The second case is the subtler one: an older release's synthesized decoder ignores the unknown `direction` key, so `session remaining > 80` would come back as `session used > 80` and render the opposite branch. A missing rule is visibly missing; an inverted one is not. * Drop cost metrics that could not be converted to USD `UsageFormatter.convertedCost` returns the source amount unchanged when it has no rate for the provider's currency, and both cost producers passed that value straight through as `costTodayUSD`/`cost30dUSD`. A `Cost today > 5 USD` rule would then compare, say, €6 against a $5 threshold and pick the wrong branch. Both producers now keep the amount only when the conversion actually landed in USD. Otherwise the predicate sees no value and evaluates false, which is the existing contract for a metric the provider does not report. The rendered text is untouched and still shows the provider's own currency. --------- Co-authored-by: Peter Steinberger <steipete@gmail.com> * Prevent RPC pipe writes from aborting after child exit (#3095) * fix: prevent RPC pipe writes from aborting after child exit * test: repin UsageFetcher codex identity anchor after pipe-write refactor * Publish live Grok tokens and xAI spend into Usage & Spend (#3085) * Publish live Grok tokens and xAI spend into Usage & Spend Enabled Grok and xAI now join the shared spend catalog instead of only inflating the unavailable denominator. xAI contributes vendor-metered daily USD from the Management API chart; Grok contributes local session tokens. SuperGrok credits and xAI prepaid balance stay quotas, not spend. * docs: add #3085 to changelog * Fix lint on Grok and xAI spend messages Wrap no-data copy under 120 characters and drop a redundant throws on the prepaid-balance mapping test. * Fix Grok/xAI spend publication, Today, and coverage Preserve xAI analytics failures as unavailable instead of known-zero spend, publish local Grok tokens when remote billing fails, pin Today to the current UTC/local day, and keep xAI history as a 30-day source. * Fix Grok and xAI spend edge cases * Fix OpenRouter localization test after #3086 #3086 scoped localizedProviderDetails to DeepSeek and z.ai, so generic OpenRouter details keep canonical English. The merge test still expected zh-Hant "API 金鑰" and failed macOS shard 0. * Align OpenRouter localization test with main Peter restored generic title/row L() localization in 84a4ca7 after #3086 scoped it away. The merge kept the canonical-English assertion from the earlier CI fix; match the restored shared catalog instead. * test: reconcile gatekeeper anchors and fingerprints with Grok/xAI spend clusters * test: include Grok and xAI in the cost-capable dashboard source contract --------- Co-authored-by: Peter Steinberger <steipete@gmail.com> * docs: credit #3085 and #3088 changelog entries * chore: finalize 0.54.0 changelog and bump build to 127 * docs: update appcast for 0.54.0 * chore: open 0.54.1 unreleased changelog section * Clarify five-hour quota wording in Simplified Chinese (#3070) * Clarify Simplified Chinese five-hour quota label * Derive Simplified Chinese session quota labels from duration --------- Co-authored-by: UNGETSU <ungetsu@UNGETSUdeMacBook-Air.local> * Fix agent session menu width (#3096) * fix(alibaba): resolve mainland Personal/Solo sec_token from the console shell (#3098) Mainland Personal/Solo Token Plan (cn-personal) fails with a 200 `BailianGateway.Login.NotLogined` body ("Alibaba Token Plan login required") even with fresh, valid aliyun cookies, because the request lacks the `sec_token` the OneConsole gateway requires. #2533 already forwards the token when present, but it was never resolvable for this path for two reasons: - The console shell only server-renders `window.ALIYUN_CONSOLE_CONFIG.SEC_TOKEN` for a genuine same-origin document navigation; a bare GET receives a token-less shell. Send the browser-navigation headers (Referer, Sec-Fetch-*, Accept-Language) so the shell includes the token. - The shell embeds it as an upper-case, unquoted key (`SEC_TOKEN: "..."`), but `extractSECToken` only matched the lower-case `secToken`/`sec_token` shapes. Add the `SEC_TOKEN` pattern. With both, the scraper resolves the Personal `sec_token`, the gateway returns real usage, and the mainland Personal/Solo card renders. Verified end-to-end on a real cn-personal account: `secTokenSource=resolved`, body `message=Success`, 5-hour/weekly windows populated (was "login required"). Adds AlibabaTokenPlanSECTokenScrapeTests covering the upper-case shell format, the existing lower-case shapes, and the no-token case. Refs #2500, #2349, #2370, #2533. Co-authored-by: LeoLin <leolin990405@gmail.com> * Add BigModel CN account balance to z.ai provider (#3109) * Add BigModel CN account balance to z.ai provider * Bound balance timeout, region-aware override validation, null-safe amounts --------- Co-authored-by: RunhuaHuang <RunhuaHuang@users.noreply.github.com> * docs: credit #3070 #3096 #3098 #3109 changelog entries * test: repair gatekeeper anchors after #3096 menu refactor --------- Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Peter Steinberger <steipete@gmail.com> Co-authored-by: kiranmagic7 <kiranmagic@proton.me> Co-authored-by: Kiran Magic <262980978+kiranmagic7@users.noreply.github.com> Co-authored-by: William Mitchell <13799071+wdmitchelluk@users.noreply.github.com> Co-authored-by: Alec Gutman, Chip <44984861+Chipagosfinest@users.noreply.github.com> Co-authored-by: UNGETSU <44204222+YunyueLi@users.noreply.github.com> Co-authored-by: UNGETSU <ungetsu@UNGETSUdeMacBook-Air.local> Co-authored-by: KaranocaVe <39995482+KaranocaVe@users.noreply.github.com> Co-authored-by: Zhongyue Lin <101193087+LeoLin990405@users.noreply.github.com> Co-authored-by: LeoLin <leolin990405@gmail.com> Co-authored-by: Xigua <rh2897@columbia.edu> Co-authored-by: RunhuaHuang <RunhuaHuang@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #3091.
Cached provider switches can move a hosted card view onto a plain
NSMenuItem, which drops theMenuCardMenuItemhighlight override.This makes cached reconciliation replace the row shell when the dynamic item classes differ. Same-class rows still use the existing content swap, so the flash-free path remains unchanged. The regression test covers plain row → card → plain switch-back.
Native-menu proof
I ran the built head in a disposable ad-hoc app bundle with keychain access and web extras disabled. The repository's
MenuSwitchFlickerProbedrove the real mergedNSMenuin two sessions: 12 cached provider/overview switches, 412 WindowServer captures, and zero nil captures.Across the first cached Claude → Codex switch, stable before/after frames had no system-blue pixels below the switcher header. The only blue region is the selected provider segment:
Capture log
This host was macOS 26.3 (25D125), so this does not claim confirmation on the reporter's macOS 27 build.
Testing
CODEXBAR_SUPPRESS_TEST_KEYCHAIN_ACCESS=1 swift test --filter MenuCardViewRecyclingTests— 25 tests passed on the rebased headmake checkAdaptiveRefreshTimerTests; two timer cases returnedCancellationError()after their 30-second timeoutsCompatibility risk is low: only cross-class cached rows take the existing replace-shell path.
Review focus: whether exact dynamic class equality is the right compatibility boundary for cached row shells.