* Frontend: inject orgId into all SPA navigations via LocationService
Append ?orgId=<N> to every SPA navigation at the history layer in
LocationService (push/replace), gated on multi-org, so URLs stay
shareable across orgs. Wires the orgId getter at startup in app.ts and
covers the behaviour in LocationService tests.
Fixes#105040
Signed-off-by: QuentinBisson <quentin@giantswarm.io>
* refactor: fix test
Signed-off-by: QuentinBisson <quentin@giantswarm.io>
* prune suppressions
Signed-off-by: QuentinBisson <quentin@giantswarm.io>
* Frontend: keep current path for query/hash-only orgId navigations
appendOrgId resolved query-only ('?tab=x') and hash-only ('#h') strings
against a dummy base and always returned url.pathname, forcing relative
navigations to '/'. Omit pathname for those so the router keeps the
current path, and cover both cases in tests.
Signed-off-by: QuentinBisson <quentin@giantswarm.io>
* Frontend: register orgId getter before post-login redirect
---------
Signed-off-by: QuentinBisson <quentin@giantswarm.io>
Co-authored-by: Laura Benz <laura.benz@grafana.com>
Co-authored-by: Ashley Harrison <ashley.harrison@grafana.com>
Co-authored-by: joshhunt <josh.hunt@grafana.com>
* docs(skills): add panel-testing-strategy skill
Add a Claude skill capturing how the DataViz squad wants unit and E2E tests
written for Grafana visualization panels and DataViz-owned code, distilled from
the April–July 2026 test-coverage effort, its PR review discussions, and the
DataViz + DataPro frontend testing strategy manifesto.
The skill encodes:
- the (inverted) testing diamond and how to pick the right layer
- assert-real-behavior conventions (no bare toBeDefined / not.toThrow)
- honest test descriptions that match their assertions
- verifying tests actually reach the target branch
- the repo's data-frame and panel-props builders, with common gotchas
- the canvas draw-call snapshot harness for chart panels
- E2E: selectors-first, an a11y check for every panel, and interaction snapshots
- no-AI-slop authoring guidance (AI speeds up a test you already intend)
- SDLC-phase expectations for when to add which tests
- 10 anti-flake rules, each mapped to a real stabilization PR
Wire it (and the existing add-e2e-selectors skill) into AGENTS.md under a new
"Project skills" pointer, and un-ignore the skill directory in .gitignore to
match the add-e2e-selectors convention.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Update .claude/skills/panel-testing-strategy/SKILL.md
Co-authored-by: Galen Kistler <109082771+gtk-grafana@users.noreply.github.com>
* Update .claude/skills/panel-testing-strategy/SKILL.md
Co-authored-by: Galen Kistler <109082771+gtk-grafana@users.noreply.github.com>
* Update .claude/skills/panel-testing-strategy/SKILL.md
Co-authored-by: Galen Kistler <109082771+gtk-grafana@users.noreply.github.com>
* Update SKILL.md
* Update .claude/skills/panel-testing-strategy/SKILL.md
Co-authored-by: Galen Kistler <109082771+gtk-grafana@users.noreply.github.com>
* Update .claude/skills/panel-testing-strategy/SKILL.md
Co-authored-by: Galen Kistler <109082771+gtk-grafana@users.noreply.github.com>
* docs(skills): trim panel-testing-strategy rules checklist and exemplars
Make the rules checklist pure pointers into the sections above instead of
restating every rule, and drop the two exemplar bullets already cited inline
(panel-props builder in Step 1, canvas harness in Step 4). Removes duplicated
maintenance surface so a rule change no longer has to be edited in two places.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(skills): tighten panel-testing-strategy description
Trim the frontmatter description from ~184 to ~124 tokens by dropping the
redundant "Encodes ..." summary sentence (that content is in the body) and
thinning the panel/component enumerations, while keeping every distinctive
trigger phrase. The description is loaded into every session's skill listing,
so this is an always-on saving regardless of whether the skill is invoked.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(skills): correct canvas harness example and waitFor idiom
The Step 4 import example named createGrafanaUiMeasureTextJestMock but
attributed it to TimeSeriesPanel.canvasTestUtils.tsx, which actually imports
applyDefaultUPlotAxisMeasureTextMock. They are two different helpers: the
harness file imports applyDefaultUPlotAxisMeasureTextMock / installCanvasPath2DShim /
removeCanvasTransforms, while each *.canvas.test.tsx mocks
@grafana/ui/src/utils/measureText via createGrafanaUiMeasureTextJestMock in a
jest.mock factory. Show both halves so the example matches the exemplar.
Also fix the renderer-ready wait: waitFor(() => uPlotInstance?.status === 1)
never retries because a falsy return does not throw. Use
waitFor(() => expect(uPlotInstance?.status).toBe(1)), consistent with
anti-flake rule #4.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* build: make the .claude skill un-ignore actually take effect
!.claude/skills/panel-testing-strategy could not re-include files because its
parent .claude/ was excluded as a directory and git never descends into an
excluded dir, so new files in the skill dir still required a force-add. Ignore
directory contents with /* and re-include each parent step by step so the two
tracked skill dirs work without -f, while the rest of .claude/ stays ignored.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(skills): prefer jest.mocked over as jest.MockedFunction casts
Add Step 3 guidance to type mocked functions/modules with jest.mocked(fn)
rather than `fn as jest.MockedFunction<typeof fn>`. jest.mocked is the
type-safe repo convention (256 test files use it vs 52 using the cast) and
avoids a common review nit, so agent-generated tests should follow it. Also
note it in the rules-checklist Step 3 pointer.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(codeowners): assign panel-testing-strategy skill to dataviz-squad
The .claude/skills/panel-testing-strategy/ dir had no CODEOWNERS entry (no
global * owner exists), so it had no owner and review routing. Assign it to
@grafana/dataviz-squad, who maintain the skill's conventions.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(skills): scope skill pointers to directory-level AGENTS.md
Move the panel-testing-strategy and add-e2e-selectors pointers out of the
root AGENTS.md, where they cost tokens on every agent session repo-wide, and
into directory-scoped AGENTS.md files that only load when working in the
relevant frontend trees:
- public/app/plugins/panel/
- public/app/features/
- packages/grafana-ui/
The pointers only advertise the skills and nudge their use in-tree; each
skill's own trigger description (already in every session's skills manifest)
decides when it actually applies, so the AGENTS.md files avoid any ownership
lookup and do not duplicate the CODEOWNERS mapping.
This follows the directory-scoped context recommendation for large codebases
and avoids adding frontend-specific guidance to the repo-wide default, keeping
context lean for backend work and for contributors without a large LLM budget.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(skills): satisfy codeowners + prettier for un-ignored skill files
The .gitignore fix in this branch makes .claude/skills/add-e2e-selectors/
genuinely tracked for the first time, which exposed two CI checks that had
never seen the file:
- codeowners-validator flagged add-e2e-selectors/SKILL.md and the new
public/app/features/AGENTS.md as not-owned. Assign the e2e-selectors skill
to @grafana/grafana-frontend-platform (owns the e2e-selectors package) and
the features agent guide to @grafana/dataviz-squad.
- prettier flagged emphasis style in add-e2e-selectors/SKILL.md; normalize
*text* to _text_.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(skills): add rule against circular test expectations
Extend Principle 2 of panel-testing-strategy: expected values must be
frozen literals, never derived from the code under test, a collaborator
it calls internally, or a copy of the production formula (comparing a
value to itself asserts nothing). Promote mutate-then-confirm-red from a
Verify step to a first-class habit.
Prompted by review feedback on geomap test coverage (#129607).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(skills): require regression tests to fail on main
A test accompanying a bug fix must fail in the absence of the fix and
pass in its presence — otherwise it pins nothing and won't catch the
regression coming back. Spell out the check (revert the fix → red,
restore → green) and reference it from the SDLC checklist item.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(skills): drop squad reference from skill pointers
An agent reading the AGENTS.md pointer has no way to know who the
"DataViz squad" is, so phrase the panel-testing-strategy pointer by what
it does — how panel/visualization tests should be written — instead of
who wants them. Applies to the three directory-level AGENTS.md files.
Addresses review feedback from @joshhunt.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(skills): phrase skill by what it does, not who owns it
Drop the "DataViz squad" framing from the skill description and intro so
an agent decides relevance by the work (writing/reviewing viz tests),
not by knowing who the squad is. Keeps the coverage-gate fact, reworded
to reference the codeowner paths rather than the team.
The one remaining slug is the literal argument to
`test:coverage:by-codeowner`, left as-is so the command still runs.
Addresses review feedback from @joshhunt.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(skills): drop the run-tests-by-codeowner recommendation
Remove the `test:coverage:by-codeowner` Verify step and its checklist
entry — the codeowner coverage gate runs in CI and isn't something an
agent needs to run locally.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(skills): prettier-format add-e2e-selectors SKILL.md
Un-ignoring .claude/skills brings its files under the repo prettier:check.
main's copy of the add-e2e-selectors skill predates that gate and wasn't
formatted, so the branch-merged-with-main lint failed. Run Prettier on it
(emphasis markers + list-item indentation only).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(skills): prefer deletion over inflation
Add a Step 2 rule that deleting a redundant assertion (e.g. a toBeDefined()
prelude before a concrete check) or a duplicate test is a preferred fix over
inflating a weak assertion into a contrived value check.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Update .claude/skills/panel-testing-strategy/SKILL.md
Co-authored-by: Jesse David Peterson <jesdavpet@users.noreply.github.com>
* Update .claude/skills/panel-testing-strategy/SKILL.md
Co-authored-by: Jesse David Peterson <jesdavpet@users.noreply.github.com>
* Update .claude/skills/panel-testing-strategy/SKILL.md
Co-authored-by: Jesse David Peterson <jesdavpet@users.noreply.github.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: Galen Kistler <109082771+gtk-grafana@users.noreply.github.com>
Co-authored-by: Jesse David Peterson <jesdavpet@users.noreply.github.com>
* DataViz: Replace circular test expectations with hardcoded literals
Several DataViz-owned unit tests computed their expected value the same
way the code under test does — via a sibling production function, a copy
of the production formula, or (in one case) by comparing a value to
itself. A shared bug passes such a test, so it raises coverage without
verifying behavior.
Freeze the expected values as literals (or, for projected coordinates,
an independent WGS84 readback of the input), so a regression in the
resolver/formula/geometry now fails the test. Each was mutation-checked.
Surfaced in review of #129607.
Test-only; no product code changes.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Revert theme-color and gazetteer test changes per review
fastfrwrd pushed back on these: hardcoding a theme color hex (or a
gazetteer coordinate) couples the test to palette/geometry internals and
breaks on a legitimate theme change without catching a real bug —
resolving a named color via getColorByName is trusted lower-layer
behavior worth asserting against directly. Restore the original
assertions in color, gazetteer, barchart, and xychart.
Keeps the non-color literal fixes that weren't contested: getLayersExtent
extents, TableNG row height, the RadialGauge self-comparison no-op fix,
and the UPlotScaleBuilder range.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* use useTheme2 to get theme
* fix contrast checker on (by series name) matchers
* replace config.theme2 with hooks
* keep config.theme2 in sync
* fix review finding
* add theme to dep array
* remove unused export
* test(SeriesEditor): remove config casts and thin grafana/ui mocks
Two readability cleanups to the xychart SeriesEditor tests:
- Drop the `onChange.mock.calls[n][0] as XYSeriesConfig[]` casts. A typed
`renderEditor()` helper returns a `jest.MockedFunction` onChange spy plus
`configAt(n)` / `lastConfig()` accessors, so tests read emitted configs
without indexing or casting. The typing also surfaced that onChange's
argument is genuinely optional, which the cast had hidden — the accessors
now guard for it explicitly.
- Delete the `@grafana/ui` `Select` mock entirely; the frame field now
exercises the real Select via the `selectOptionInTest` helper and the real
clear control.
- Keep a thin `FieldNamePicker` stub but remove the `mockPickerItems` module
global. Picker settings (`filter`, `baseNameMode`) are read back through
the typed `jest.mocked(FieldNamePicker)` calls via a `pickerSettings` helper.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* XYChart: consolidate SeriesEditor test Options cast into one helper
Route every context mapping through a single mappingOptions() helper so the
narrowing 'as Options' cast lives in exactly one place.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Prettier: anchor `build` and `data` ignore patterns to repo root
The bare `build` and `data` entries in .prettierignore use gitignore-style
matching, so they match any directory of that name at any depth — not just
the repo-root output/runtime dirs they target. This silently hid real source
from prettier, notably public/app/plugins/panel/geomap/layers/data.
Anchor both with a leading slash so they match only the root dirs (mirroring
.gitignore's `/data/*`), and format the source files that were exposed as a
result so `prettier:check` stays green.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Prettier: format devenv data.js files exposed by anchoring
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Replace String.prototype.localeCompare with a module-scope Intl.Collator
in the scenario select sort. Spec-equivalent (default locale/options), so
sort order is unchanged. Removes the @grafana/no-locale-compare suppression
for this file.
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
* TestData: wire scenario select and labels input e2e selectors
Wires the existing-but-unused scenarioSelect aria-label onto the
Scenario <Select>, and adds+wires a new templated labelsInput(refId)
selector onto the per-query Labels input.
Split out of grafana/grafana#128750 per the MT service compatibility
check, since this touches an independently-deployed data source
plugin (public/app/plugins/datasource/*).
* Revert QueryEditor.tsx wiring, moved to a stacked follow-up PR
check-separation flags any PR touching both packages/grafana-e2e-selectors
and public/app/plugins/datasource/*, regardless of what else is in the
diff. Keeping this PR to just the new labelsInput selector definition;
the QueryEditor.tsx wiring (this + the pre-existing scenarioSelect) moves
to a PR stacked on top of this one.
* Fix version key on labelsInput to computed main version
Per the add-e2e-selectors skill: new selectors use the current main
version (13.2.0), not MIN_GRAFANA_VERSION.
* TestData: wire scenarioSelect and labelsInput onto QueryEditor
Stacked on jayclifford/first-dashboard-testdata-e2e-selectors, which
adds the labelsInput selector definition. This PR only touches the
datasource plugin file itself, so check-separation passes.
* chore: revert change
---------
Co-authored-by: Jack Westbrook <jack.westbrook@gmail.com>
Co-authored-by: Hugo Häggmark <hugo.haggmark@gmail.com>
* TimeComparison: Fix high cardinality mismatches
* Add functional test coverage
* Rework getCompareSeriesIdentityKey slightly
* TimeComparison: Add testing to verify color labels (#128840)
* Fix random walk test data use case
* Update tests
* Stat: Improve unit test coverage
Adds a common.test.ts (the option builders had no coverage), and expands
the migration and panel render tests to exercise previously-untested
branches: angular colorValue/lineColor migration paths, the
grafana-singlestat-panel plugin id, textMode None, the Area sparkline
path, percent change, and all-values rendering.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* improve tests in my imo
* missed snapshot in commit
* fix import
* updates from review
* clarify assertion
* Stat: strengthen migration/panel/common tests per panel-testing-strategy
Replace a .not.toThrow() assertion with concrete field-config checks, extract a
single time-series frame builder, and swap no-op toBeDefined() checks for exact
value/path assertions.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Stat: tighten migration test names and gates
Apply the panel-testing-strategy conventions to the Stat migration tests:
- Collapse the topic-named graphMode/colorMode/textMode cases into three
it.each tables whose names state the asserted value.
- Add a single makePanel() builder instead of per-test PanelModel casts.
- Cover the colorBackground-over-colorValue precedence and the case where a
sparkline lineColor must not become a fixed field color because the value
was already colored.
- Drop the duplicated colorBackground case.
- The non-singlestat test passed for the wrong reason: with empty prevOptions
the missing `angular` key short-circuited the migration, so the assertion
held for any plugin id. Pass migratable angular options so the plugin id is
the only gate.
- Remove format/decimals setup that no assertion used.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Geomap: Add unit tests for gazetteer, layer registry, esri, style dims
Focused unit-first coverage for geomap/geo pure logic, following the
panel-testing-strategy conventions (real ol as data constructors,
concrete-value assertions, it.each for variants).
- frameAsGazetter: lat/lng detection (incl. LATITUTE/LONGITUE aliases),
geohash and existing-Point derivation, non-point centroid, key detection
(*_CODE/UID/first-string fallback), case-insensitive lookup, count/examples
- getLayersOptions: basemap vs overlay sets, current selection, alpha gating
(label/bolt/ordering) and the beta fall-through, via deterministic layer mocks
- esri basemap create(): service->ArcGIS slug URL, default fallback, custom
passthrough, TileLayer/XYZ shape
- getStyleDimension: custom-config vs style.fields branches and the text gate
Test-only; no product code changed.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Geomap: Fix typecheck and prettier issues from CI
- Cast pointFieldFrom*().values[0] to Point before getCoordinates()
(Field values are typed as the Geometry base class)
- StyleConfigFields values are driving field-name strings, not booleans
- Apply prettier formatting
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Geomap: Add stretch unit tests for lastPointTracker, maplibre, marker fallbacks
Covers three previously-untested alpha/basemap paths:
- lastPointTracker create().init() + update() tracking the last row coordinate
- maplibre init() LayerGroup/opacity and style-URL fetch selection
- getMarkerMaker text-label and error-marker fallbacks
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
waitFor(() => uPlotInstance?.status === 1) never retries: a waitFor callback
must throw to trigger a retry, and a falsy return does not, so the uPlot
readiness gate resolved immediately without actually waiting. Use
waitFor(() => expect(uPlotInstance?.status).toBe(1)), matching the idiom
already used in TimelineChart.canvas.test.tsx.
Affects the HeatmapPanel, XYChartPanel, and SparklineCell canvas suites. No
snapshot changes: each suite's follow-up .u-over waitFor already gated actual
render, so this only hardens the readiness wait.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* Table: Add unit tests for cell editors, module, and suggestions
Backfills coverage for the previously untested custom cell-option editors,
the panel module wiring, and the suggestions supplier in the table panel.
Follows the panel-testing-strategy conventions: concrete-value and exact
call-argument assertions, behaviour-specific test names, it.each for
enumerable variants, and boundary cases for the suggestion scoring.
- suggestions: score thresholds (Best/Good/OK) incl. strict boundaries and
the preview card modifier
- module: plugin wiring plus the "Cell value inspect" and "Tooltip
placement" showIf branches
- TableCellOptionEditor: cell-type switching, sub-editor swapping, and the
settings cache restore-on-switch-back
- cell editors: each editor's onChange output and showIf filtering
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Table: Address self-review feedback on editor/module tests
- Cover multiple cell-type sub-editors (not just gauge) via it.each
- Rework the type-change test to start from a configured gauge and
affirmatively assert the previous settings are discarded
- Combine the "Cell value inspect" showIf cases into a single tagged
it.each table
- Reword the suggestions-supplier test name
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Table: Initialize standard editors registry in module test
Building the plugin's fieldConfigRegistry runs the table useCustomConfig
path, which resolves the 'stats-picker' standard editor. Without calling
standardEditorsRegistry.setInit(getAllOptionEditors) the registry access is
order-dependent and can throw when the file runs alone. Matches the setup in
bargauge/module.test.ts and SparklineCellOptionsEditor.test.tsx.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Table: Make sparkline showIf test assert real behavior
The previous assertion queried the bar-alignment option with getByLabelText,
but a wrapped radiogroup option editor has no label association, so that query
was null whether the option was shown or hidden — the test was vacuous.
Query by the field's label text instead and cover both states: shown for the
Bars draw style, hidden otherwise. The positive case proves the query is not
always-null.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Table: assert suggestions supplier delegation instead of duplicating output test
The module test now checks that the plugin delegates to tableSuggestionsSupplier
(scores flow through); the supplier's own count/score semantics stay in
suggestions.test.ts.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Table: guard suggestion supplier/plugin results for void return type
tableSuggestionsSupplier and PanelPlugin.getSuggestions are typed to return
void as well as an array; narrow both before mapping so the test typechecks.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* Gauges: Fix color for unfilled bars
* switch to action hover
* update
* Fix
* BarGauge: Refactor to function component and access theme via context
* remove update config.theme2
* remove suppression for function components for BarGaugePanel
---------
Co-authored-by: Paul Marbach <paul.marbach@grafana.com>
Newly added timeseries/graph panel options were declared with an explicit
`defaultValue` in their options-builder entries. PanelPlugin collects every
builder `defaultValue` into `plugin.defaults`, and `getPanelOptionsWithDefaults`
merges those into the saved panel options. As a result, opening and saving any
dashboard with graph panels (without making changes) wrote these fields into
every panel's JSON, producing noisy diffs:
- legend.enableFacetedFilter
- legend.overflow
- annotations.multiLane
- annotations.clustering
Remove the `defaultValue` from these builder entries so the options are only
serialized when a user actually sets them. Runtime behavior is unchanged: all
consumers already fall back to the previous defaults when the value is
undefined (faceted filter off, ellipsis overflow, multi-lane off, clustering
disabled).
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* Move incident card gate to parent
* Add news card when no incident card
* Tweak dashboard tab list padding
* Don't show error while loading
* Don't full-width news images for mobile
* remove loki backend
* remove loki frontend
* remove loki from various places
* remove loki from more places
* copy/paste loki stuff for the sake of the rest
* prettier
* go work sync
* update wire
* retrigger
* fix unit tests
* update
* fix adhoc e2e tests
* gen jsonnet
* remove loki tests
* update CODEOWNERS
* copy unit tests
* put loki devenv back
* make gen-jsonnet
* update yarn.lock
* put loki-promtail back
* Dashboards: Fix duplicate timeCompare series on streaming queries, drop processor fork
timeShiftAlignmentProcessor mutated refId/meta in place on frames owned
by Loki's split-query accumulator, so every re-emitted chunk appeared
as a new series to the merge layer and comparison series duplicated per
chunk. The processor now lives in @grafana/scenes (exported there) and
returns new frame objects; dashboard-scene's private fork is deleted.
alignTimeRangeCompareData in @grafana/data had the same in-place
mutation hazard at render time - it now returns a copied frame without
breaking its public signature.
Depends on the @grafana/scenes canary from grafana/scenes#1583; the pin
moves to stable once that releases.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Bump @grafana/scenes canary to 29911997371 (idempotent getCompareSeriesRefId)
Main pre-suffixes compare target refIds at request time (#128132), so
the scenes processor must not double-suffix them - the guard landed in
grafana/scenes#1583.
@grafana/scenes-react stays on the previous canary because its publish
kept failing on npm provenance tlog 409s; a resolutions entry forces its
nested @grafana/scenes to the new canary so a single copy is installed.
Both the pin and the resolution go away when scenes#1583 ships stable.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Bump @grafana/scenes to stable 8.13.4
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Ashley Harrison
Adela Almasan
Matt Cowley
Nathan Vērzemnieks
Galen Kistler
Cody Confer
Jack Westbrook
grafana-pr-automation[bot]andgrafana-pr-automation[bot] <140550294+grafana-pr-automation[bot]@users.noreply.github.com>
cursor[bot]andCursor Agent
Ryan Morris
Drew Slobodnjak
Victor Marin
ismail simsek
Torkel ÖdegaardandPaul Marbach
Kristina
Dominik ProkopandClaude Fable 5
Tom Ratcliffe
