mirror of
https://github.com/Kpa-clawbot/meshcore-analyzer.git
synced 2026-10-11 13:17:20 +00:00
8933e2c703aaa48afcb7d3ffdccdd511bf091387
16
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 efitenandClaude Opus 5.5
|
8933e2c703 |
fix(channels): keep client-only state across a channel-list refresh (#2095) (#2096)
Fixes #2095. `loadChannels()` replaced the `channels` array with the server snapshot and carried nothing across, and `mergeUserChannels()` only ever ran from `init()`. So any refresh destroyed every field that exists only in the tab. **No race needed for the worst of it.** Change the region filter or toggle show-encrypted with a PSK channel open, and: - the **My Channels** section disappears — `renderChannelList()` derives it as `channels.filter(c => c.userAdded === true)`, and the server returns neither `userAdded` nor a `user:`-prefixed hash - every unread badge resets to 0 - the user's own labels vanish - `reconcileSelectionAfterChannelRefresh()` cannot find the `user:*` hash in the snapshot, so it nulls `selectedHash`, sets `messages = []`, rewrites the URL to `#/channels` and replaces the open conversation with "Choose a channel from the sidebar" ## What this does `mergeClientChannelState(fresh, prev)` mirrors `mergeWsAppendedIntoRest()`, which already does exactly this job for `messages` (#1498). Same shape: pure, takes both arrays as parameters, returns a fresh array, never aliases or mutates an input. It carries `unread`, `userAdded` and `userLabel` by hash, and keeps `lastActivityMs` / `lastSender` / `lastMessage` / `messageCount` when a WebSocket batch landed while the request was in flight and is therefore newer than the snapshot. Those four move together: a sender without its message reads as a different message. `mergeUserChannels()` now runs inside `loadChannels()`, before the render and before the reconcile, so all three call sites get it instead of `init()` alone. ## What it deliberately does not do **It does not resurrect a channel the snapshot left out.** A region-filter change legitimately narrows the list, so carrying survivors over would defeat the filter, which is a worse bug than the one being fixed. The helper only enriches rows already present in the fresh snapshot. That leaves half of finding 2 in the issue unfixed: a channel pushed by the WebSocket handler during the initial in-flight window is still dropped. That one self-heals on the channel's next packet, and the reverted preview line is overwritten by the next WS batch. Fixing it properly needs a way to tell "dropped because the snapshot is stale" from "dropped because the filter excludes it", which is a larger change than this. ## Tests `tests/unit/test-issue-2095-channels-client-state.js`, 11 cases, registered in `test-all.sh`. Part 1 exercises the helper directly. Part 2 drives the real `loadChannels()` through the existing `_channelsLoadChannelsForTest` hook with a stubbed `api()`, which is what proves the helper is wired in rather than merely defined. **Verified by mutation.** With the wiring removed from `loadChannels()` but the helper left in place, the three reproduction cases fail with exactly the reported symptoms: ``` FAIL a refresh keeps the My Channels rows My Channels lost on refresh (got ["public1"]) FAIL a refresh keeps unread badges the unread badge reset to 0 on refresh undefined !== 7 FAIL a refresh does not close an open PSK conversation the open PSK channel was deselected + null - 'user:MyPSK' ``` The fourth case, "a refresh still drops a channel the server filtered out", stays green throughout, so the fix cannot be defeating the region filter. Full frontend suite green: `sh test-all.sh` exits 0. ## Not done **No browser validation.** This is frontend JS covered by unit tests that call the production function through its own hook, but I did not run Playwright against a server, so I am not claiming a browser check I did not do. `init()` still calls `mergeUserChannels()` and `renderChannelList()` after `loadChannels()` resolves, which is now redundant and costs one extra full sidebar render per page load. Left alone deliberately: it is idempotent, and the comment there records the regression it was added to fix (`test-channel-issue-1111-e2e.js`, case 2). Worth removing separately with that e2e test watched. --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
n30nex
|
b9fb3d244c |
feat(analytics): split adverts using recorded route evidence (#2088)
Red commits: `dc7dcbe`, `76e3217` ([assertion failures](https://github.com/Kpa-clawbot/CoreScope/actions/runs/37059977385)). Fixes #2041; supplies #2085's backend contract. Records at most two route bits per retained advert. Node flood counts now use the same evidence, including mixed and transport-flood adverts. Analytics failures log/count errors while core observation, path and liveness writes continue. The unrelated recent-advert limit clamp is removed. Direct adverts (empty path) describes the observed remaining path across valid hash widths. Firmware can forward an advert down to an empty path; this cannot establish original send mode or RF distance. The `zero_hop` API spelling remains compatible. Successful post-upgrade evidence writes survive observation replacement and restart. Older overwritten frames and failed writes can leave gaps. The UI explains automatic background backfill. ## Measured validation Synthetic Windows workloads; timings are not production guarantees: - 300K retained adverts, one-hour window: median 64.35→39.84 ms; 13.98 MB→145 KB allocated. Seven-day allocations unchanged; five paired samples. - 1M transmissions/11M observations: backfill 5m43s. Simple 10 Hz writer: all 3,423 observations persisted; p99 28 ms, maximum 323 ms. - Simulated pre-backfill cursor: retained evidence ready 4m21s; full catch-up 6m41s, 122 cache invalidations. Normal completed-migration startup bulk-loads current evidence. Portable opt-in scale tests and window benchmarks are included. All 183 frontend suites, full server tests, targeted races, lint and real-API desktop/mobile checks passed. Full [Linux CI](https://github.com/Kpa-clawbot/CoreScope/actions/runs/37063210229) passed at `36def9c`; three independent reviews are clear. Local Windows testing hit the unchanged symlink-privilege test. No configuration changes. Preflight overrides: external script/OpenClaw profile unavailable; repository checks and Chromium used. |
||
|
Sylvain Rabot
|
a4a447acea |
fix(packets): pack short columns, one expand arrow, Full Names toggle (#2090)
The packets table wastes most of its width on short columns, draws two arrows on grouped rows, and cuts every observer and hop name short. This fixes all three and adds a way to read the full path when it does not fit. ## Why the columns were wide `makeColumnsResizable` measured each column once on load and stored the widths as **percentages** of the table. Percentages grow with the screen: on a 2000px display "17s ago" sat in a 180px column. The measurement also read the virtual-scroll spacer row — a single `colspan` cell as wide as the table — as column 0, so the expand column came out 160px wide locally and about 950px on a live instance with grouped rows. Measured on the fixture at 2000px: | Column | Before | After | |---|---|---| | Expand | 162–230px | 32px | | Time | 111px | 67px | | Size | 64px | 46px | | HB | 44px | 29px | | Scope | 80px | 48px | | Path / Details | ~180px each | ~650px each | ## The change **Column packing.** A new `fitColumnsToContent` (`app.js`) sizes every column except Path and Details to its widest cell, in px; those two split what is left. Every visible column gets an explicit width and the table gets their sum: a spacer row keeps a slot for each hidden column, and leaving any column `auto` hands those slots a share of the slack as a blank strip at the right edge. When space is short (detail panel open, narrow window) the widest fixed columns give way first, then Path and Details shrink to 60px, and only then does the wrapper scroll. The split itself is a pure function, `distributeColumnWidths`, with unit tests. It re-lays out when the container resizes (detail panel), when TableResponsive hides or reveals columns (a new `table-columns-changed` event), and after sorting or a column toggle. After each render `grow()` widens columns for the rows just inserted and never narrows them, so nothing jitters while scrolling. Drag handles store px; double-clicking one resets it. Path and Details have no handle. `makeColumnsResizable`, still used by the nodes, observers and analytics tables, now skips `colspan` rows when measuring; other rows are measured as before. **One expand arrow.** A CSS `::before` "▶" sat next to the SVG caret, which pointed up when collapsed. The glyph is gone and the caret points right, then down. **Full Names.** A toolbar toggle next to Hex Paths shows observer and hop names untruncated. It is saved per browser; a link can carry `fullNames=1`, which applies to that page without overwriting the visitor's saved preference. Paths stay on one line: the virtual scroller positions rows at one fixed height, so wrapping would make scrolling jump. **The "+N" path pill, and a full-path popover.** Hops past the column edge were meant to go behind a "+N" pill (#1124), but the pill was appended after the overflowing chips and so sat past the clipped edge: it never showed. It is now sticky at the right edge, opaque on hover, and counts hop chips only (it used to count warning icons too). Hovering or focusing it shows the full path, one hop per line and numbered; click, Enter or Space pins it until an outside click or Escape, and it closes when the page changes or its row scrolls away. Clicking the pill no longer selects the row underneath. **Two related fixes.** - Rows drawn before `/api/observers` resolves show raw 64-char pubkeys. `loadObservers` now re-renders and re-measures, instead of leaving Observer at the pubkey's width (~600px under Full Names). - Group child rows indented every cell by 20px, which misaligned them with the headers and widened every column on expand. Only the Time cell is indented now. Anyone who had dragged packet columns gets the new defaults once: the old percentage key is dropped. ## Performance `grow()` runs after every `renderVisibleRows`, so it is on the hot path. Measured in Chromium on the fixture, 300 iterations, three runs: | Case | Cost | |---|---| | Full render, 63 rows measured | +0.88–0.93ms over the forced reflow alone | | Scroll step (incremental render) | ~73 `Range` rect reads, only the inserted rows | | Large jump / re-render | ~569 rect reads | Before narrowing `grow()` to inserted rows, one scroll step with 92 rendered rows made 5,472 rect reads. The DOM row count is bounded by the virtual-scroll window, so the cost does not grow with the packet count. `distributeColumnWidths` is a binary search over at most ~10 columns. ## Tests - `tests/e2e/test-packets-compact-columns-e2e.js`, 16 cases: short columns stay under fixed px bounds at 1920px; Path and Details take the freed width; hiding both still fills the wrapper; no blank strip and no horizontal scroll, with and without the detail panel, at 1920, 1100, 1000 and 375px; responsive reveal and re-hide refit; stale percentage widths dropped; drag follows the pointer and double-click resets; expanding a group leaves widths alone; Full Names untruncates, keeps paths on one line, persists and round-trips through the URL; the Observer column shrinks back when observer names arrive late; the pill popover lists every hop vertically, closes when the pointer leaves, pins on click without selecting the row, closes on Escape and on navigation, and opens on keyboard focus. Registered in `scripts/non-unit-tests.json` and `deploy.yml`. - On `master` the suite fails where it should, and restoring each fix's old behaviour (child-row indent, the late-observer refit) turns its case red. - `tests/unit/test-packets.js`: `distributeColumnWidths` (room to spare, rounding, widest-first capping, floors, dragged columns, the 60px minimum, overflow), the single chevron and no `::before` glyph, Full Names on flat and grouped observer cells, and the Full Names CSS. - `tests/unit/test-xss-escape-sinks.js` now builds the real `obsCellName` from source: escaping moved into it. Removing its `escapeHtml` turns both observer-cell tests red. - `test-issue-1189-composed-cell.js` and `test-issue-2012-clear-filters-selection.js` evaluate packets.js fragments with injected helpers and needed `obsCellName` / `showFullNames` added. - Existing packets E2E suites pass on this branch and on master against a CI-prepared fixture: scope column, #1128 (layout and multi-viewport), #1122, #1188, #1648, table sort, #1692, #1486. - Unit: 182 of 183 suites pass locally; `test-issue-1956-release-routing.js` could not create a temp directory in my sandbox and is unrelated. ## Browser validation Checked in Chromium against the fixture at 2000, 1920, 1280 (panel open and closed), 1100 and 375px, and with the seeded grouped row expanded and collapsed. Screenshots taken at each step; the table fills its container exactly, grouped rows show one arrow, and the pill popover lists full names. ## Not done, deliberately - **No width cap on Observer under Full Names.** A cap brings back the truncation the toggle removes, and Observer is already the first column to give way when space is short. - **Customizer.** The floors (60px flex minimum, 64px shrink floor, 32px expand, 58px Rpt) are hardcoded, per AGENTS rule 8; exposing them in the customizer can follow. - **Wrapping paths.** Rejected because of the fixed row height above; the pill popover is the way to read a long path. |
||
|
n30nex
|
410c82c02c |
fix(ui): use consistent 48px navigation and channel buttons (#2078)
Fixes #2052. Navigation (`.nav-btn`) and channel icon (`.ch-icon-btn`) buttons now use the shared 48px house minimum. Remove competing 44px component/coarse-pointer rules and the legacy 32px mobile rule, and correct comments that confused the house preference with accessibility requirements. At 768px, user-added channel rows already clipped Remove; the larger Share target exposed the same constraint. Allow only those rows' controls to wrap so both actions remain within the sidebar. Navigation height and normal network-channel rows are unchanged. - Red `7787d9b`: rendered target assertions fail at 375/390/768/1280px. Green `e6ed9c8` changes CSS only. - Validation: all 183 standalone suites; 33 computed touch-target checks; 15 actual-page layout/click checks. Broader browser suite: 131 passed, 3 fixture-dependent skips. Three independent reviews found no required changes. - Browser verified: local Chromium and fixture-backed Go server; screenshots `coverage/2052-targets-375.png` and `coverage/2052-targets-768.png`. - E2E assertion added: `tests/e2e/test-channel-fluid-e2e.js:112`, covering dimensions, containment, overflow and normal Filter/Share clicks. Existing CI selection is retained; Chromium-required runs cannot silently skip the touch suite. - Test-only follow-up `5144503` waits for packet readiness before measuring navbar controls. Five fresh mobile contexts passed; size and click assertions remain intact. - No new settings, requests, dependencies or runtime JavaScript. ## Preflight overrides - External `run-all.sh` unavailable; repository checks run directly. - Local unit runner uses UTF-8 and `GITHUB_REF_NAME=local-validation` for the existing release harness. Browser navigation uses a 60s local budget for slow assets; repository assertions/timeouts are unchanged. |
||
|
n30nex
|
f1edbbef3f |
fix(packets): preserve observation selection in detail URLs (#2093)
Red commit: `c165087` ([CI assertion failure](https://github.com/Kpa-clawbot/CoreScope/actions/runs/36644030012)). Fixes #2091. Packet detail links now retain the selected observation when page initialization or filter changes rebuild the URL. Initialization also restores `obs` from the complete hash after the router strips its query. Explicit ID links render the requested observation and retain it in Copy Link state. The shared updater reads selection from the current route, so returning to the list or selecting another packet cannot resurrect an old observation. Existing filter serialization and Clear Filters behavior remain intact. No new requests, configuration, dependencies or layout changes. ## Validation - Real fixture browser checks use nondefault observation 502: hash/ID load, type/observer/time-window changes, refresh, Clear, and another refresh. Both URL and selected row are asserted. - All 183 standalone frontend suites passed; focused filter browser suite 11/11. - Broader local core run reached the unrelated Live input readiness bug #2094, reproduced on unchanged master. The complete CI browser suite passed. - ESLint 8: zero errors; 91 existing warnings. XSS diff, syntax and whitespace checks passed. - Three independent reviews found no production defects; assertions additionally prove filter controls changed before checking selection. E2E assertion added: `tests/e2e/test-filter-ux-e2e.js:179`. OpenClaw profile/external preflight were unavailable; Chromium and repository checks ran directly. Local navigation used a 60-second budget. Final CI passed at `280ced5`, including Go, browsers, coverage and both container architectures ([run](https://github.com/Kpa-clawbot/CoreScope/actions/runs/36646175352)). |
||
|
n30nex
|
dc4db17c48 |
fix(nodes): remove unsupported region fields from Heard By (#2077)
Fixes #2062. The node health API does not emit observer region data, but Heard By rendered a Region column containing only dashes and a Regions summary that could never appear. Remove that column, its sort control, and the unsupported region displays from the full node page and side panel. Keep Observer, Packets, Avg SNR and Avg RSSI, along with existing escaping, signal placeholders, relay counts and badges. No API, dependency or configuration changes. - Red commit `4396492` fails on the unwanted Region header and summary; `a360711` removes the unsupported fields. - Validation: 183 standalone suites and 7 focused browser checks passed. Broader browser run: 131 passed, 3 fixture-dependent skips. Independent reviews completed; the sorting-test finding was addressed in `1452981`. - Unit coverage evaluates the real templates. The existing CI-selected browser suite checks four-column alignment and sorting on desktop/mobile using the actual API field names. - Browser verified: local Chromium against a fixture-backed Go server; screenshots recorded in `coverage/issue-2062-heard-by-1400.png` and `coverage/issue-2062-heard-by-390.png`. - E2E assertion added: `tests/e2e/test-issue-1151-orphan-separators-e2e.js:125`, the two #2062 desktop/mobile cases. - No added requests, loops or data structures. ## Preflight overrides - External `run-all.sh` unavailable; repository syntax, whitespace, CSS-variable, XSS and PII checks run directly. - Local unit execution supplies UTF-8 settings and `GITHUB_REF_NAME=local-validation`, required by the existing release-routing test harness. - Local browser navigation budget raised to 60s for slow local asset responses; repository assertions and timeout settings unchanged. |
||
|
efitenandClaude Opus 5
|
3caa847323 |
fix(nodes): the advert section is called "Recent Adverts", and says why (#2071)
Closes #2042, reported by @damn-simple-scripts. ## Which of the two fixes The report offered widening the section to all packets the node originated, or renaming it, and proposed the rename. Widening sounds like the better fix, so I measured before agreeing. On a production database: | | | |---|---| | transmissions total | 1,251,967 | | with `from_pubkey` populated | 191,237 | | of those, `payload_type = 4` (ADVERT) | **191,237** | | with `from_pubkey` and **not** an advert | **0** | The ingestor fills that column for adverts only. Attributing a relayed CHAN or TXT packet back to its sender is the path-resolution problem, not a filter this section could apply — so "show all packets from this node" is not a small change, it is a different feature resting on attribution the data does not carry. So the rename is correct, and the numbers say so rather than my preference. ## The change Both copies renamed — the full node page and the side pane. Both already read `nodeData.recentAdverts`, so the field feeding them said "adverts" while the heading said "packets". The heading also gained a `title` naming `from_pubkey` as the reason it is adverts only. Renaming without explaining invites the same report from the next reader; the tooltip is where that explanation costs nothing. ## Tests `tests/unit/test-issue-2042-recent-adverts-label.js`, four cases: both copies present and headed "Recent Adverts", no copy back to "Recent Packets", the advert field still feeding them, and the explanation still in place. The third matters most — it ties the label to its data source, so pointing this section at a different field in future fails here rather than silently making the label wrong again. ## Not done, deliberately The reporter's follow-up idea, splitting flood adverts from zero-hop adverts into two lists. They wrote that it can be dropped if an issue should tackle one thing, so it is not here. It is feasible now: a zero-hop advert is `ROUTE_TYPE_DIRECT` with an empty path, established while fixing #2064. That deserves its own issue rather than a paragraph in this one — say the word and I will open it. --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
efitenandClaude Opus 5
|
2cfe9cbbc5 |
fix(a11y): raise the Scope Audit observed-chip contrast above 4.5:1 (#2070)
Closes #1996. ## Reproduced first The chips tint their own background — `color-mix(in srgb, var(--status-green) 16%, transparent)` — which darkens whatever surface sits behind them. With `--status-green-text` (green-700, `#15803d`) on top: | surface | composited chip | ratio | |---|---|---| | `--surface-0` `#f4f5f7` | `#d2eddf` | **4.04:1** | | `--card-bg` `#ffffff` | `#dcf6e5` | **4.38:1** | Both under the 4.5:1 the #1719 gate requires for normal text. The first row is the same ratio **and the same hex** @n30nex measured in the browser, which is how I know the model in the test agrees with what a visitor actually sees. ## Fixed on the text, not by thinning the tint Dropping the tint to 12% reaches 4.52:1. That is two hundredths above the line, and a margin that thin fails again the next time a surface value moves — the same trap #2039 hit with a threshold sitting inside the healthy band. One palette step darker gives **6.23:1** on white and **5.74:1** on `--surface-0`, and keeps the tint that makes a chip read as a chip rather than as plain text. - `--palette-green-800: #166534` added. The greens ran 300–700 while the blues already reach 900, so this fills the scale rather than inventing a colour. - `--sa-chip-observed-fg` defined per theme: green-800 in light, and the bright `#22c55e` **kept** in dark, where the chip already passed at 6.48:1 / 5.73:1. Dark is deliberately untouched. - The chip reads the variable, so the customizer still governs it and no literal enters a component. `.sa-chip-verified` needed no change: `scope-audit.js:106` only ever adds it alongside `sa-chip-observed` or `sa-chip-unobserved`, and it contributes an underline. So the fix covers both classes the issue names — worth stating, since the title mentions both. ## Tests `tests/unit/test-a11y-1996-scope-audit-chips.js`, 7 cases: both themes × both surfaces, that the chip still tints (so the suite cannot pass by testing nothing), that its colour comes from a variable rather than a literal, and a guard asserting green-700 **would** still fail — so a quiet revert to `--status-green-text` turns this red instead of passing. **Red-run confirmed:** with the old colour restored it fails at exactly 4.04:1 and 4.38:1, naming the composited `rgb(210,237,223)`. Two deliberate choices in the test: - **A separate suite, not a case in `test-a11y-1719`.** That suite's `parseColor` handles hex and `rgb()` only; teaching it `color-mix()` is a larger change than this fix. These chips are the only contrast-critical user of the function today. If a second appears, the two should merge, and the file says so. - **Derived from the stylesheets, not a rendered page.** The default Scope Audit fixture renders no chips at all, which is precisely why the existing browser coverage missed this. A stylesheet-derived check cannot be defeated by a fixture that shows nothing. `check-css-vars` passes (180 definitions, 0 undefined) and `test-test-inventory` passes with the new file classified. --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
|
435ac25dd6 |
feat(nav): show the running version in the nav-drawer footer (#2068)
Takes over #1985 by @SaarMesh-Bot, as offered there on 2026-09-16 and 2026-09-17. **Their commit is the first here, unchanged and under their authorship**; the rest clears the two review points. Closing #1985 in favour of this so the rebase and the fixes travel together, not to reassign the work. ## The feature, unchanged The frontend never surfaced which build was running, though `/api/health` has reported `{version, commit, buildTime}` all along. A footer on the nav drawer now renders `CoreScope <version>`, linking to the releases page, with commit and build time in the tooltip. Colours come from existing CSS variables, the label is set with `textContent`, and a failed health call leaves a neutral label rather than an empty footer — all as the author wrote it. ## Review point 1: fetch on open, not on page load The `/api/health` call sat in `buildDom()`, which runs on page load. The drawer may never be opened, and **cannot** be opened at ≤768px, where the module is disabled by design. So every visitor's browser was requesting the endpoint to fill a footer most of them would never see. Moved into `open()`, after the width gate. `fetchVersion` still caches its promise for the page lifetime, so re-opening costs nothing. ## Review point 2: the fallback is pinned `tests/unit/test-nav-drawer-version-footer.js`, six cases: - a rejected fetch, a non-ok response, and a 200 without `version` each leave the neutral `CoreScope` label — not a blank footer and not `CoreScope undefined`, which is what an instance shows exactly when someone is trying to read its version - a normal response renders the version, with commit and build time in the tooltip - a version carrying markup lands verbatim in `textContent`, and `innerHTML` is never touched - the endpoint is requested once however often the footer is filled, pinning the cache from the other side It **slices `fetchVersion` and `fillVersion` out of the shipped `public/nav-drawer.js`** and evaluates them rather than copying them into the test, so it exercises what ships — same approach as `tests/unit/test-direct-rf-heard-by.js`. Both slice markers are asserted, so a rename fails loudly instead of quietly testing nothing. Each case re-evaluates the slice, because `versionPromise` caches for the page lifetime and a shared sandbox would hand the second case the first case's answer. Listed in `test-all.sh`, which per #2036 is the only frontend runner. ## A note on the third commit The first version of the test used `setImmediate` to settle the promise queue. It ran fine under node and **failed eslint**, which treats these as browser code. I ran eslint and committed in the same command and pushed without reading its output. Fixed in the commit after, using `setTimeout(r, 0)`. Recording it because the PR would otherwise show a lint failure in its history with no explanation. ## Verification 6 of 6 in the new suite, eslint clean on both changed files, `test-test-inventory.js` passes with the new file classified. Cherry-picked cleanly onto current master. --------- Co-authored-by: SaarMesh-Bot <bot@saarmesh.de> Co-authored-by: Claude <noreply@anthropic.com> |
||
|
liquidraverandClaude Opus 5
|
b695b979a2 |
fix(nodes): keep paginating past a page that post-LIMIT filtering shortened (#2061)
## Problem `handleNodes` runs the geo-filter, `nodeBlacklist`, `hiddenNamePrefixes` and area passes **after** the SQL `LIMIT/OFFSET`, and rewrites `total` to the filtered length. A page that loses a row is therefore short **without being the last page**, and neither the page length nor `total` can tell a client whether to ask for another page. #1606 added the pagination loop and chose the page length as the canonical stop. That is correct only where nothing is ever filtered. Everywhere else the list truncates at the first filtered page boundary and strands every node behind it — the #1598 symptom reached by a different route: a node that is relaying right now simply stops being in the list. The comment at `app.js:240` rejects `total` for exactly the right reason, then picks the signal the same code path also breaks. ## Measured on a live 2346-node deployment Page sizes for the query the map issues: ``` offset=0 returned=500 ← full, loop continues offset=500 returned=499 ← one row filtered AFTER the LIMIT → loop STOPS offset=1000 returned=500 ← never requested offset=1500 returned=500 ← never requested offset=2000 returned=345 ← never requested ``` | stop rule | requests | nodes reached | |---|---:|---:| | short page (master) | 2 | **999** | | `has_more`, else empty page | 6 | **2344** | **1341 nodes, 57%, unreachable through the UI.** ### One hidden node truncates the whole list The deployment this came from has no `geoFilter` (`/api/config/geo-filter` returns `polygon: null`) and no `nodeBlacklist`. It has a single `hiddenNamePrefixes` entry — a deliberate operator choice — and exactly one node whose name starts with it: ``` public_key d4a46ea2…1054 (64 clean hex chars) name 🚫🔥☀️ role repeater last_seen 2026-09-22T10:09:38Z ``` `handleNodes` drops that row in the `IsNameHidden` pass, which runs after the SQL `LIMIT`. The row is counted by the `LIMIT` and by `COUNT(*)`, so the page it lands in comes back exactly one short — and stops every client that treats a short page as the end. Isolated against SQL on the same database, seconds apart: ``` SELECT lower(public_key) FROM nodes ORDER BY last_seen DESC LIMIT 500 OFFSET 500 -> 500 rows GET /api/nodes?limit=500&offset=500 -> 499 rows comm -23 sql.txt api.txt -> d4a46ea2e99cab132a3286ef3d9cce9099318790af7f25671fe83de453721054 ``` Deterministic — `offset=500` returned 499 on three consecutive requests. Not a CDN artifact either: `cf-cache-status: DYNAMIC`, origin `cache-control: no-store`, no `age` header, and four requests with deliberately unique cache keys all returned 499. So **one deliberately hidden node makes 1341 of 2344 nodes unreachable.** The hiding feature does exactly what it was asked to do for that one node, and takes 57% of the network with it, silently. A single `hiddenNamePrefixes` entry is enough; no geo-filter, blacklist or area filter is needed to reach this state. ### The cutoff moves, which is why this reads as intermittent The visible set is the sum of the pages up to and including the first short one, so the boundary sits wherever the unreturnable row currently sorts by `last_seen`, and jumps a whole page as ingest reorders the list. Same deployment, same code, same config, ~2h apart: | dropped row's rank | first short page | nodes visible | |---|---|---:| | inside 0–499 | page 1 | 499 | | inside 500–999 | page 2 | 999 | A node is visible or invisible purely by where it lands relative to that moving line, so affected nodes appear to vanish and return on their own. Two operators on this deployment reported exactly that, independently, while I was measuring. ### A named reproduction `HU-ZA-Lentihegy` (`5287a33f…`), reported missing from the map by an operator whose companion had logged its advert at 04:20 local the same morning. Ingest was fine. The row is in `nodes` with `last_seen` `2026-09-22T02:20:35Z` — the same advert, to the second — valid GPS, role `repeater`, 1033 adverts, and `/api/nodes/search?q=lentihegy` returns it. ``` rank by last_seen : 1081 cutoff at the time: 999 ``` It missed by 82 positions. Walking the same live endpoint, same moment: | stop rule | requests | nodes reached | Lentihegy | |---|---:|---:|---| | short page (master) | 2 | 999 | **not reached** | | `has_more`, else empty page | 6 | 2340 | reached | The practical shape of this on a busy mesh: 1081 nodes had been heard more recently than 9.4 hours, so on that deployment **anything last heard more than ~9 hours ago was invisible**, alive or not. `#/nodes` compounds it — its search box filters client-side over the truncated set, so the server-side `?search=` never runs and an operator cannot find the node by searching for it either, even though the endpoint would return it. ## Change **Server** — `NodeListResponse` gains `has_more`, computed from the raw SQL page against the real `COUNT(*)` before the filter passes run, so it survives them: ```go hasMore := offset+len(nodes) < total ``` Always emitted (no `omitempty`) so a client can tell `false` from an old server. No extra request in the fixed path: `has_more` ends the loop exactly, where the old rule needed a probe page. **Clients** — `app.js` `fetchAllNodes`, `nodes.js` `loadNodes` and `area-map.html`'s inline helper stop on `has_more`, falling back to a zero-length page against a server that predates it. An empty page always ends the loop, so a `has_more` against a concurrently-shrinking table cannot spin to `safetyCap`. Left alone: the three loops are still three copies. Collapsing them onto `fetchAllNodes` is a bigger change than this fix needs, and `nodes.js` has its own inter-page progress UI. Happy to do it separately if you want it. ## Testing - **Unit** (`tests/unit/test-fetch-all-nodes-pagination.js`): the fixture now models the real handler — a row counted by the LIMIT and by `COUNT(*)`, then removed from the page. Three new cases. Fails on the old rule at 499 of 1199. - **E2E** (`tests/e2e/test-map-nodes-pagination-e2e.js`, already wired into `deploy.yml`): the mock drops a page-1 row and emits `has_more`. Mutation-checked — restoring master's stop rule fails 3 of its steps. - **Go** (`cmd/server/nodes_pagination_has_more_test.go`): asserts `has_more` stays true on a page filtering shortened. Mutation-checked — recomputing it after the filter block fails the test. - Full server suite `go test -race`: ok, 41.4s. `gofmt` clean, `go vet` passes. - **Against a real binary**, not just mocks: fixture DB migrated with `corescope-migrate`, `hiddenNamePrefixes: ["SKCE"]`, `limit=3`. Page 1 returns 2 of 3 with `total` rewritten to 2 and `has_more=true`. Walking the real server with master's rule reaches 2 nodes; with `has_more`, all 199 visible of 200, the hidden one still hidden. The real frontend against that server loads 199 with no JS errors. Two existing expectations changed, both deliberate: 1. `surfaces ALL nodes past the 500 server cap` — 3 → 4 requests. That mock emits no `has_more`, so the 200-row final page can no longer end the loop (a short page is exactly what a filtered page looks like) and a zero-length probe follows. Against a current server `has_more` still ends it at 3. 2. `rows missing public_key are NOT collapsed into one` — its stub returned a constant body, which would now be paged to `safetyCap`. It serves one page then empties. Local `test-all.sh` exits 1 on two XSS-gate self-tests (`good-2-tested.js`, `good-4-tested.js`) via a `UnicodeEncodeError` printing an emoji under Windows cp1252. Identical on clean `origin/master` in a scratch worktree, so it is pre-existing and platform-local, not this branch. There is a second identical filter block further down `routes.go` on another list endpoint. Likely the same class; not touched here. If you would rather land your own version of this, say so and I will close mine. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Opus 5 <noreply@anthropic.com> |
||
|
efitenandClaude Opus 5
|
980c5c4515 |
fix(nodes): stop the Heard By empty state claiming the node is out of range (#2063)
Follow-up to #2057, which is correct in what it does and overclaims in one sentence. ## The sentence When no observer heard the node directly, the card said: > No observer is within radio range of this node. #2057's own rule cannot establish that: - every **direct route** is discarded, because the firmware removes the sender from the path before retransmitting (`Mesh.cpp`, `removeSelfFromPath`), so the packet cannot say who transmitted it. #2057 measured these at **38% of transmissions over 7 days**. - **28.8% of flood observations with a path** are dropped because the last hop resolves to more than one candidate, and the rule under-attributes rather than guesses. So an empty direct list is missing evidence, not evidence of missing coverage. ## Why it matters in practice Sampled 40 repeaters on a production instance after #2057 shipped: | | | |---|---| | at least one direct observer | 24 | | empty card, "no observer is within radio range" | **16** | | of those 16, with a non-zero `relayObserverCount` | **16** | Every node showing "nobody is in radio range" also showed "Seen via relay by N observers" two lines below. An operator reading that about a working repeater concludes they have a coverage problem they do not have. ## The change Wording only, in both copies of the card (full page and side pane): > No observation proves a direct reception here, which is not the same as being out of range. with the reason in a `title`, so the card stays one line: > Only flood-routed transmissions identify who was heard: a direct route removes the sender from the path before retransmitting (firmware `Mesh.cpp`, `removeSelfFromPath`), and an ambiguous relay hop is left unattributed rather than guessed. So an empty list is missing evidence, not proof of missing coverage. No API change. #2057's rule, shape and performance work are untouched — I verified its firmware derivation against the clone at `0679dbef` before writing this: `Packet.h:83`, the forwarder appending with `packet->getPathHashSize()` at `Mesh.cpp:349`, and `removeSelfFromPath` on the direct path at `Mesh.cpp:89-105` all read as described. ## Test `tests/unit/test-direct-rf-heard-by.js` slices this template out of `public/nodes.js`, so it pins the shipped markup. It now asserts the old sentence is gone and the qualifier is present. Worth recording how that assertion was reached: my first version banned the phrase "out of range" from the card, and it failed — on the new line, which contains that phrase precisely in order to deny it. A word ban was the wrong instrument. Matching the old sentence and requiring the new qualifier is the assertion that actually distinguishes the two states. 8 of 8 in that suite, eslint clean. ## Not in this PR The **Regions** line and **Region** column on the same card read `o.iata`, which `HealthObserverRow` does not emit, so both have always been dead. #2057 named this and left it; it is now **#2062** with the file and line references, rather than a remark inside a merged description. Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
efitenandClaude Opus 5
|
d1b615fc0d |
fix(node-health): list only observers that heard the node on air (#2057)
Closes #2056. ## What changes The node detail "Heard By" card now lists only observers that received the node's **own transmission off the air**, and reports the rest as a count. ``` HEARD BY — DIRECT (8 OBSERVERS) OBSERVER REGION PACKETS AVG SNR AVG RSSI BE-DUF-SiSCD-01 — 16276 7.9 dB -108 dBm BE-BRU-Moris repeater — 13775 -5.6 dB -122 dBm ... Seen via relay by 29 observers. Those observers heard a repeater that forwarded this node's traffic, not this node. ``` and for a node nothing hears: ``` HEARD BY — DIRECT (0 OBSERVERS) No observer is within radio range of this node. Seen via relay by 2 observers. … ``` ## The rule, and where it comes from Read out of the firmware rather than assumed: | | | |---|---| | `Packet.h:83` | `setPathHashSizeAndCount(sz,n) { path_len = ((sz-1)<<6) \| (n&63); }` — hash size rides in the packet's `path_len` byte | | `Mesh.cpp:649,678` | only `sendFlood()` sets it, so the **originator** decides; `CommonCLI.h:69` defaults `path_hash_mode = 0`, i.e. one byte | | `Mesh.cpp:349` | a forwarding repeater appends its hash with the packet's size — it cannot upgrade a packet, and the **last hop is who was heard** | | `Mesh.cpp:89,103` | on a direct route a forwarder matches the head of the path and calls `removeSelfFromPath` before retransmitting, so the path is the **remaining** route and the transmitter is not in it | So an observation credits exactly one node: 1. Route type must be `ROUTE_TYPE_FLOOD` or `ROUTE_TYPE_TRANSPORT_FLOOD`. Direct routes never qualify (38% of transmissions over 7 days). 2. Empty path → the originator, known only for ADVERTs. 3. Otherwise the last hop. 4. The hop must resolve to exactly one candidate. Same gate `resolvePathForObsColdLoad` already applies: under-attribute rather than guess. It drops 418,530 of 1,455,721 flood observations with a path over 7 days (28.8%), and it is what stops the wrong-band credits. ## Measured effect | node | before | after | |---|---|---| | BE-BRU-Moris | 36 observers | 3 | | BE-KRO-RP01 \| ON1KW | 40 | 3 | | BE-BRE-ON8AR | 38 | 2 | | NL-BXE-RP01 \| 433 | 35 | 0 | Network-wide over 7 days, 234 of 1,860 nodes have at least one direct observer (161 have exactly one, maximum 8). The direct list is therefore empty for most nodes, with the relay count below it. That is the correct reading: no observer is in radio range of them. Independent corroboration on staging: for BE-WIL-3EIK-01 the eight direct observers are exactly the top eight entries of its Neighbors table by score and observation count. ## Perf justification `GetNodeHealth` is fast today precisely because it never walks observations — it uses one representative observation per transmission. Direct-RF needs the per-observation path, and that cannot be a per-request walk: the reference store holds **232,928 transmissions / 2,887,861 observations**, one node's `byNode` slice alone holds **55,458 transmissions / 1,450,544 observations**, and `/api/nodes/bulk-health?limit=200` would multiply that. So the aggregate is rebuilt by a background recomputer on the existing `newAnalyticsRecomputer` pattern, published into an `atomic.Value`. Reads are `O(direct observers)`, which is **cheaper than before** — the old code built per-observer sums over every transmission in `byNode` on every request. Proof, `BenchmarkBuildDirectHeardIndex`: ``` BenchmarkBuildDirectHeardIndex-12 1 63067900 ns/op ``` 3,000,000 observations (60,000 transmissions × 50 observations, 8-hop paths, 64 candidate repeaters) in **63 ms**, once per recompute interval. Per observation the walk does one route-type check, one backward scan of `PathJSON` for the last quoted token (no allocation, no `json.Unmarshal`), one prefix-map lookup and one counter update. Rebuilding wholesale also means eviction needs no bookkeeping: a pass simply does not see evicted transmissions. The alternative — a field on `StoreObs` updated incrementally — would have needed the call at five construction sites (`store.go:942,1264,2854,3179`, `chunked_load.go:609`), which is the duplication that caused #1558, plus matching decrements at eviction. ## API Both `GetNodeHealth` and `GetBulkHealth` carried a near-identical copy of the observer loop; they now share one builder. - `observers` — direct-RF only. Same field names, so no client migration. Rows are a named `HealthObserverRow` instead of `map[string]interface{}` (one fewer occurrence in a touched file, per the AGENTS.md ratchet). - `relayObserverCount` — new integer, observers that saw traffic through the node without hearing it. `stats.totalPackets` and `stats.avgHops` still count relayed traffic, so without this number the card would contradict the figures printed beside it. `docs/api-spec.md` is updated for both endpoints. It also documented an `iata` field on these rows that the endpoint has never emitted; removed. ## Tests - `cmd/server/direct_heard_test.go` — table test over the rule: flood with empty path and known originator, flood whose last hop is the node, flood whose last hop is another node, direct and transport-direct routes (never credit), ambiguous last-hop prefix, listener-only candidate, 1-byte and 2-byte hop sizes; plus aggregation and row-building. - `cmd/server/node_health_direct_rf_test.go` — end-to-end through the handler: an observer that only saw relayed traffic must not appear in `observers` but must be counted in `relayObserverCount`. Plus the benchmark. - `tests/unit/test-direct-rf-heard-by.js` — slices the card template out of `public/nodes.js` and evaluates it, so it tests the shipped markup rather than a copy: heading, empty state, relay line, singular/plural, signal columns, listener/repeater badge tri-state. - `cmd/server/node_health_can_relay_case_1290_test.go` — updated to seed a genuinely direct reception, since a relay-only observer no longer carries a badge. - `cmd/server/analytics_recompute_after_load_test.go` — recomputer count 10 → 11. Verified locally: `cmd/server` suite green, `sh test-all.sh` green (180 suites), `tests/e2e/test-e2e-playwright.js` 131/134 passed with 3 skipped and 0 failures against the seeded fixture, plus `test-issue-1147-section-order-e2e.js`, `test-issue-1151-orphan-separators-e2e.js` and `test-issue-1281-location-row-e2e.js`, which all assert on this card. `gofmt` clean, `vet` clean across all modules. Browser-validated on staging: both the full detail page and the side pane, on a node with 8 direct observers and on the 433 MHz node with none. No console errors. ## What this does not do `prefixMap.resolveWithContext` still guesses on ambiguous hops, so paths, neighbor edges and analytics keep their current attribution. Making it abstain is a much larger change and needs its own issue. The "Regions" line and Region column on this card read `o.iata`, which this endpoint has never emitted, so both have always been dead. Left as found rather than widened into this change. Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
efitenandClaude Opus 5
|
5c016a210a |
fix(analytics): show a building state on the distance index's 202, and stop caching it (#2051)
Closes #1997. ## What was wrong `/api/analytics/distance` answers `202 {status:"building", retry_after_seconds:5}` with no `summary` until the lazy index (#1011) has been built. 1. `renderDistanceTab` read `data.summary.totalHops` straight away. The TypeError is caught by the tab's own try/catch, so it never reaches `window.onerror`: it is painted into the tab as `Failed to load distance analytics: Cannot read properties of undefined (reading 'totalHops')`. 2. `api()` caches any `res.ok` body, and `res.ok` is true for 202, so the placeholder was stored for the `analyticsRF` TTL. Even a correct retry read the cached "building" body back. That is the half that made the broken state outlast the index build. ## What changed - `public/app.js:173` skips the cache write when `res.status === 202`. A 200 still caches, unchanged. - `public/analytics.js` renders a building notice and retries itself, honouring `retry_after_seconds` clamped to [1s, 30s]. The timer is cleared on tab switch and in `destroy()`, and a new render supersedes a pending retry, so two renders cannot write into the same tab. The notice uses `.text-center`/`.text-muted` rather than the `.spinner` class used at `analytics.js:1683`, because `.spinner` has no CSS anywhere in the repo and renders nothing. ## Tests `tests/unit/test-issue-1997-distance-building.js` (7 assertions, wired into `test-all.sh`) pins the two pure decisions the renderer makes and `api()`'s refusal to cache a 202 while still caching a 200. Red-run on the unfixed sources: 6 of 7 fail, and the "a 200 is still cached" control stays green. `tests/e2e/test-issue-1997-distance-building-e2e.js` (classified in `scripts/non-unit-tests.json`, invoked from `deploy.yml` with `CHROMIUM_REQUIRE=1`) serves both responses by route interception, so it does not depend on whether the server under test has an index built. It asserts: the building state appears, the tab does not paint the error text, a retry arrives with no interaction, the retry replaces the placeholder once the server answers 200, and no retry fires after leaving the tab. Check (2) deliberately asserts on the rendered text and not on `pageerror`: the TypeError is caught, so a `pageerror` assertion would pass on the broken build too. ## Verification No local cgo toolchain here since #1992, so I could not build a server to run the E2E against. Instead I ran its five steps in Playwright against a live instance with this branch's `public/app.js` and `public/analytics.js` injected in place of the deployed ones (both files are byte-identical between that instance and upstream master, so the injection is faithful): | check | deployed build | this branch | |---|---|---| | (1) building state shown | fail | pass | | (2) not rendered as data | fail | pass | | (3) retried on its own | fail (1 request) | pass (2 requests) | | (4) real payload after retry | fail | pass | | (5) no retry after leaving the tab | pass | pass | (5) passes on the broken build too: it schedules no retry at all, so it is a control and only means anything together with (3). The committed E2E suite has not been run as committed. CI is its first real run. ## Not done - The server still recomputes on every 202 poll rather than signalling readiness. - No other analytics tab was audited for the same assume-a-summary pattern. - One pre-existing unit suite (`test-preflight-xss-gate.js`) fails on this Windows machine with a cp1252 `UnicodeEncodeError` from its Python helper, on a clean tree as well as with this change. Unrelated, and green on Linux CI. Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
liquidraver
|
84eba41592 |
fix(live): anchor the legend toggle to .live-page so the VCR bar stops eating its clicks (#2049)
The PACKET TYPES legend could not be dismissed on Live: clicking the palette toggle did nothing, while document.querySelector('#legendToggleBtn').click() from the console worked. That asymmetry was the diagnosis, since a synthesised click skips hit testing.
Residual half of #1833. That fix corrected the button's offset to calc(var(--vcr-bar-height) + 10px) but left the anchor: .legend-toggle-btn was position: fixed, so the offset resolved against the viewport, while the .vcr-bar it clears is absolute inside .live-page, whose height subtracts --bottom-nav-reserve (56px + safe-area at <=768). The two anchors disagreed by exactly the reserve, dropping the button into the bar's band, and .vcr-bar at z-index 1000 against the button's 500 took every real click. The window is 641-768px: below 641 both buttons are display:none, above 768 the reserve is 0, which is why neither the desktop nor the phone layout test saw it.
Both buttons are now position: absolute, resolving against the same containing block as the bar. .feed-show-btn carried the identical defect but missed the bar by 16px because --legend-toggle-stack parks it a row higher, so that change prevents a break rather than repairs one.
Three review rounds, each verified rather than argued:
- The new E2E was classified in non-unit-tests.json but had no deploy.yml line, so it would never have run (#2037). Now wired in, and it covers both buttons.
- The '.feed-show-btn is display:none at <=768' claim was wrong (it is <=640), which the author measured and corrected.
- Its first real execution failed on its own setup: getComputedStyle().getPropertyValue() on an unregistered custom property returns the unevaluated declaration ('calc(56px + 0px)'), so the non-zero check could not work. It now asserts the measured gap instead, and deliberately does not pin 56 because the gap measures 58 on CI.
Final run: 14 passed, 0 failed. The discriminating assertion is (f), each button's offset net of the bar being identical at 720 and 1440 (11px and 123px slack); under the old anchor those differ by the reserve. (g) pins the desktop layout unmoved.
Merged by the interim maintainer without a second human reviewer: CI and the review above are the independent checks.
Fixes #1833
|
||
|
liquidraver
|
bbf54cfe1b |
fix(rx-coverage): open at the configured map default, with its own saved viewport (#2033)
The coverage page opened at a hardcoded [51.0, 4.8] zoom 8 regardless of deployment, ignoring /api/config/map (#2032). It now follows the same precedence as the main map (URL hash, then saved position, then /api/config/map, then [37.6, -122.1] zoom 9) and persists its own position across visits, syncing lat/lon/zoom into the hash so a view is shareable. The saved position lives under its own key, rx-coverage-view, and the page never reads or writes the main map's map-view: sharing the configured default was the bug, sharing the session position was not. Both suites assert map-view stays untouched after a pan, so reintroducing a shared write fails instead of passing quietly. Also fixed here: selectedRx is now percent-encoded into the hash, and a generation counter stops a late /api/config/map response or a stale 150ms layout timer from building a map for a page that was already left. Reviewed twice. Verified by mutation rather than by reading: writing map-view too, ignoring the saved coverage position, and dropping the /api/config/map fetch each make the unit suite exit 1, so it covers the feature, the fix and the rejected alternative. The deploy.yml invocation was confirmed to land inside Run Playwright E2E tests (fail-fast) by parsing the workflow, and CI run 35261689694 is the E2E suite's first real execution: Go prints 'RX coverage viewport regressions OK' and Playwright prints 'RX coverage viewport browser regressions OK'. Worth recording for the next reviewer: that E2E asserted localStorage.getItem('map-view'), so wiring it into deploy.yml without updating the assertion would have turned the job red on its first ever run. It was registered in scripts/non-unit-tests.json but invoked by nothing, the gap tracked as #2037. Merged by the interim maintainer without a second human reviewer: CI and the mutation checks above are the independent checks. Fixes #2032 |
||
|
Alex B
|
893773338e |
chore(tests): move root test-*.js into tests/unit and tests/e2e (#2036)
Moves 290 root test-*.js into tests/unit (177, listed in test-all.sh) and tests/e2e (113, classified in scripts/non-unit-tests.json), per #1981 and PR-D of #1385. Root goes from 348 entries to 48. test-all.sh and test-fixtures/ stay put. The inventory guard now fails if a test reappears in the root or sits in the wrong folder.
Verified independently of the diff: the invoked sets are unchanged (test-all.sh 177 before and after, deploy.yml 96 before and after, both identical as sets), and a full local run of test-all.sh on master and on the branch produced 4702 output lines each whose only differences are absolute paths, stack-trace line numbers shifted by the REPO_ROOT line, the inventory wording and two perf ratios. The guard was mutation-checked: a test back in the root, a unit suite in tests/e2e, and a suite dropped from test-all.sh each make it exit 1. CI run 35246304316 ran 97 suites from tests/e2e and is green.
Follow-up
|