Dockview UI redesign + default @internal filesystem provider#920
Merged
Dockview UI redesign + default @internal filesystem provider#920
Conversation
Reframes the redesign around VS Code's two-tier model: tools (single-panel-per-bar on Left/Right/Bottom activity bars) vs. views (splittable editor area). Adds a Plots Browser tool panel and treats plots as views in the editor area. Drops VS Code multi-panel-native integration as out-of-scope. Switches to feature-flagged incremental rollout. Confirms zero backend changes are needed. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Revised spec based on brainstorming: drop feature-flag migration in favor of big-bang replacement with final verification sweep; add Rooms, Filesystem panels alongside Plots Browser; move chat to right activity bar default (closed); make viewer closable with F1 leave-room cascade (plots close, URL updates); replace ad-hoc panel defs with a single PANELS registry as the frontend source of truth; defer Pydantic panel control entirely (UI chrome, not a visualization feature); add default-layout table covering every panel and state. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
16-task plan for replacing fixed sidebar + react-rnd floating windows with dockview-react editor area and three activity bars with single-panel-per-bar zones. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Installs dockview-react@^5.2.0 in preparation for replacing the fixed sidebar + react-rnd floating window layout with a dockview-managed editor area and activity-bar-driven tool panels. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The ?panel=filesystem effect called toggleActive unconditionally, which flipped the panel on then off under React StrictMode's double-effect in dev. Check the current active first and only toggle when the target panel is not already active. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
… react-rnd Delete files fully replaced by dockview panels: - SideBar, PrimaryDrawer, WindowManager, FigureWindow, AddPlotButton, ChatWindow, windowManagerStore, formStore, pages/filesystemBrowser. - Drop react-rnd dependency (dockview provides native floating panels). - Port StaticInfoBox off react-rnd to a plain absolutely-positioned Box. - Prune now-unused openWindow plumbing from HandlerContext and useSocketManager. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…e cascade Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Scopes the resolution of the 7 ISSUES.md items plus a bottom-drawer shrink bug found during brainstorming into one coherent design. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Second-pass spec review resolved a blocker where the §8 tab-hide rule would have hidden the rightHeaderActionsComponent container, making the popout/maximize buttons and the viewer close affordance unreachable on single-panel groups. Dropped §8 entirely; updated goals, space reclaim math, and E2E coverage (added theme sync, plotly resize, plots browser simplification, and plot positioning assertions). Fixed registry.tsx extension, tightened sliver drag scope phrasing, concretized the dockview CSS variable list. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Third-pass spec review surfaced two ambiguities: §3/§4 did not make it explicit that rightHeaderActionsComponent is rendered per dockview group (so popout/maximize appear on every group, not just the root), and §7 step 1 conflated "oldest panel" (temporal) with "sorts first lexicographically" (alphabetical). Tightened both. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Maps the 8 issues in 2026-04-16-dockview-followup-design.md to 10 bite-sized TDD tasks. Each task contains a failing E2E assertion, the implementation, and a commit step. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Part of the dockview follow-up: empty right bar is a prerequisite for the sliver drop-zone change.
…draw uv build --reinstall is not a valid flag combination (verified via uv build --help). The hatch build hook handles frontend packaging on any wheel build, so uv sync --reinstall-package zndraw rebuilds frontend and reinstalls in one step.
Wrap DockviewReact in an absolutely-positioned box so its ResizeObserver sees the flex-parent's computed size. Before this, bottom-drawer opens did not shrink the viewer because DockviewReact rendered inline and its height was tied to content, not the parent's current size.
Tab chrome now reads MUI palette tokens so the dockview surface visually belongs to the app. Works for both light and dark themes because the MUI tokens themselves flip on color scheme change.
Three interacting bugs made the earlier Task 3 commit (f65b6dd) a no-op: 1. createTheme did not emit --mui-palette-* variables. ThemeProvider alone only wires the React context. Added cssVariables with colorSchemeSelector='data' so dark mode is togglable via the [data-dark] attribute MUI generates. 2. DockviewReact fell back to its internal themeAbyss default because no `theme` prop was passed — our className="dockview-theme-light" was additive, not a replacement. Switched to theme={themeLight} so the abyss class is no longer applied. 3. dockview-mui.css was loaded before dockview-react's runtime CSS injection, so dockview's base theme rule won on the cascade. Doubled the class selector to out-specificity the base rule, and replaced two phantom variables (--dv-tab-background-color and --dv-tab-header-foreground-color do not exist in dockview's CSS) with the real per-state tab-color variants. New E2E assertions lock in the behavior in both light and dark mode: each dockview CSS variable must resolve to the same value as the corresponding MUI palette token, and the MUI background-default must be light (#fff) or dark (#121212) depending on which mode is active. Root cause verified by reading dockview-core/theme.js (internal themeAbyss default on line 2236 of dockviewComponent.js) and inspecting the generated MUI stylesheet (only :root+[data-light] rule was emitted before the cssVariables config).
Read MUI's resolved color mode via useColorScheme() and pass themeLight or themeDark to DockviewReact. Combined with the MUI palette overrides from Task 3, dockview chrome now fully tracks the app's light/dark mode.
Eliminate env_parse_none_str='none' sentinel. Two explicit fields: filebrowser_enabled: bool = True filebrowser_path: str = '.' Removes the model-wide footgun where any string field accepting 'none' (e.g., guest_password) was silently coerced to None. Migrate all callsites: bootstrap.py and database.py None-checks now test filebrowser_enabled directly. conftest.py default changes from FILEBROWSER_PATH=none to FILEBROWSER_ENABLED=false; every test that opts in to the filebrowser now passes FILEBROWSER_ENABLED=true explicitly. Delete the weak tautological test; rename remaining two filesystem tests to match the new env-var names. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Commit d65d919 added TestFilebrowserConfig.test_default_filebrowser_path_is_cwd and a module-level test_filebrowser_path_default_is_cwd asserting the same invariant. Keep the class method. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
dockview-layout.spec.ts and pr920-fixes.spec.ts referenced selectors (data-sliver-state, text 'Drop to dock right') that do not exist in the implementation. Frontend e2e is not in CI; the files were misleading-docs masquerading as coverage. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
vite.config.ts uses @vitejs/plugin-react-swc; the non-SWC variant was installed alongside but never imported. Dropping removes ~7 transitive Babel packages. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replace the 'let sharedApi' module singleton + setInterval polling with
a Zustand store. Consumers:
- Components subscribe via useDockviewApi((s) => s.api) and re-render
when the api binds — no more polling.
- Non-component callers (socket handlers, effect callbacks) use
.getState().api.
Also closes the ViewerView.onLeaveRoom stale-reference race: the
onDidRemovePanel subscription now lives in a useEffect([dockApi, api.id])
that re-runs whenever the store api binds, so the handler is always
current.
Additional consumers migrated beyond the spec: RoomsPanel, FilesystemPanel,
and landingPage (viewer re-add + reset-layout handler) which also held
getDockviewApi references.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three paths previously re-declared the viewer panel config (id, component, title): DockviewLayout.onReady, roomId useEffect, and the Reset-layout MenuItem. Extract two helpers colocated with addViewerPanel; consumers import and call. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two consumers (useHasFilesystemProviders, FilesystemPanel) used the same queryKey with different staleTime. Per-observer staleTime means the Panel's default-0 defeated the Hook's 5_000 dedup whenever the Panel was open. Extract one hook owning both query + socket invalidation. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Extract ~90 LOC of duplicated drag-depth/setTimeout(0)/onDragEnter logic into useDragHover(position). Shared keyframes shimmer in dragStyles.ts (three copies removed). Empty bars now render as a 4px sliver instead of returning null, so dragging an icon off a bar still leaves a drop target to drag one back. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Covers moveIconToBar (with/without index), dropIconOnPanel, resetLayout, and toggleActive (open/close/switch). Also updates test-setup.ts mock to provide real default panel icons for slice tests. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replace useAppStore.getState() inside the effect with explicit selectors in the dep array. Effect now re-runs correctly when the active bar state changes out-of-band. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Safe autofix pass + hand fixes for noNonNullAssertion (FilesystemPanel early-returns on null roomId) and useExhaustiveDependencies (ChatPanel). - FilesystemPanel.tsx: replaced roomId!/selectedProvider! non-null assertions with early guard inside queryFn; fixed useTemplate string concatenation - ChatPanel.tsx: wrapped scrollToBottom in useCallback; added biome-ignore for intentional tick-trigger deps (data) in both effects - PlotsBrowserPanel.tsx: moved biome-ignore to useMemo declaration line (was orphaned on the closing brace); updated reason text - RoomsPanel.tsx: useOptionalChain applied to r.description check - PlotView.tsx: suppressed pre-existing noDoubleEquals (intentional loose equality for plotly string/number coercion) with targeted biome-ignore PlotView.tsx 'any' types and noNonNullAssertion are pre-existing legacy debt and explicitly out of scope per spec §3.9. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
for more information, see https://pre-commit.ci
PythonFZ
commented
Apr 20, 2026
Member
Author
PythonFZ
left a comment
PythonFZ
left a comment
There was a problem hiding this comment.
- style sometimes hard coded, sometimes using theme provider
- inconsistent testing / tests are missing in CI
- LoadFile no appears as modifier in the UI
- lots of defensive imports on the python side!
- lots of small util functions not necessarily needed! Evaluate each one of them if they should remain or be removed
getattr(app.state, "settings", None)patterns throughout the code, whilst thesettingsin this case always exist! Unecessary defensive programming!
Comment on lines
+30
to
+31
| const api = useDockviewApi.getState().api; | ||
| if (api) closePlotTab(api, data.key); |
Member
Author
There was a problem hiding this comment.
is this the intended REACT way of using the useHook?
Comment on lines
+1
to
+2
| import { act, renderHook } from "@testing-library/react"; | ||
| import type { DragEvent } from "react"; |
Member
Author
There was a problem hiding this comment.
there seems to be some inconsistency with the test, some are in __test__ and some are directly along the other files
Comment on lines
+13
to
+17
| const BAR_SX = { | ||
| left: { flexDirection: "column", width: 48, borderRight: 1 }, | ||
| right: { flexDirection: "column", width: 48, borderLeft: 1 }, | ||
| bottom: { flexDirection: "row", height: 40, borderTop: 1, width: "100%" }, | ||
| } as const; |
Member
Author
There was a problem hiding this comment.
Do we have a style provider somewhere?
| import { type BarPosition, PANELS, type PanelId } from "./registry"; | ||
| import { useDragHover } from "./useDragHover"; | ||
|
|
||
| const DRAG_MIME = "application/x-zndraw-panel-id"; |
Member
Author
There was a problem hiding this comment.
const DRAG_MIME = "application/x-zndraw-panel-id"; duplicated throughout the code
Comment on lines
+3
to
+6
| export const shimmer = keyframes` | ||
| 0%, 100% { background-color: rgba(25, 118, 210, 0.12); } | ||
| 50% { background-color: rgba(25, 118, 210, 0.28); } | ||
| `; |
Member
Author
There was a problem hiding this comment.
don't we have a style provider somehwer?
Comment on lines
+139
to
+145
| async def _execute( | ||
| request_id: str, | ||
| provider_id: str, | ||
| params_json: str, | ||
| token: str, | ||
| ) -> None: | ||
| await ex(cls, params_json, provider_id, request_id, token) |
Member
Author
There was a problem hiding this comment.
Pushback: taskiq rejects functools.partial. Verified with a minimal repro — broker.register_task(partial(executor, prov_cls), ...) raises:
TypeError: functools.partial(...) is not a module, class, method, or function.
from typing.get_type_hints(func) inside register_task. taskiq needs a real function object for type-hint introspection. Keeping the factory-closure pattern with default-arg binding.
| # Hide @internal:filesystem:* from non-superusers when the gate is on. | ||
| if settings.filebrowser_require_superuser and not _current_user.is_superuser: | ||
| gate_filter = ~( | ||
| (ProviderRecord.room_id == "@internal") |
Member
Author
There was a problem hiding this comment.
check that provider registration prohibits @internal prefix! TDD if not tested!
Member
Author
There was a problem hiding this comment.
Added 403 gate in register_provider (router.py:1062) mirroring the existing DELETE-side policy (@internal providers cannot be deleted). Two new TDD tests in test_providers.py: test_register_internal_provider_forbidden_normal_user and test_register_internal_provider_forbidden_superuser — both RED before the fix, GREEN after.
Comment on lines
+35
to
+37
| atoms = ase.Atoms("H2O", positions=[[0, 0, 0], [0, 0, 1], [1, 0, 0]]) | ||
| xyz_path = tmp_path / "water.xyz" | ||
| ase.io.write(xyz_path, atoms) |
Member
Author
There was a problem hiding this comment.
why not use a fixture. check conftest, otherwise add simple water file!
Member
Author
There was a problem hiding this comment.
Added a water_xyz fixture to tests/zndraw/worker/conftest.py that writes a 3-atom H2O xyz to tmp_path and returns the path. The e2e test now consumes the fixture directly — no more inline ase.Atoms(...) + ase.io.write(...).
Comment on lines
+259
to
+266
| def test_filebrowser_enabled_from_env(self) -> None: | ||
| """Configurable via ZNDRAW_SERVER_FILEBROWSER_ENABLED env var.""" | ||
| os.environ["ZNDRAW_SERVER_FILEBROWSER_ENABLED"] = "false" | ||
| try: | ||
| settings = Settings() | ||
| assert settings.filebrowser_enabled is False | ||
| finally: | ||
| os.environ.pop("ZNDRAW_SERVER_FILEBROWSER_ENABLED", None) |
Member
Author
There was a problem hiding this comment.
this is just testting that pydantic-settings work. We can trust the package and need no double testing here! Remove this and similiar non-tests! Keep tests for testing defaults!
Member
Author
There was a problem hiding this comment.
Dropped 12 *_from_env tests that just exercised pydantic-settings' own env-var plumbing (storage, media_path, host/port, simgen, worker, internal_url, filebrowser_enabled/path, task_queue_name, plus a generic "picks up env changes" smoke). Kept: all test_default_*, the env_parse_none_str-disabled tests (ZnDraw-specific behavior), customise_sources/pyproject.toml loader tests (ZnDraw-specific), and the /v1/config/global-settings endpoint tests. 32 → 20 tests, all green.
Comment on lines
+147
to
+148
| from zndraw_auth import User | ||
| from zndraw_joblib.models import ProviderRecord |
Member
Author
There was a problem hiding this comment.
Hoisted from zndraw_auth import User and from zndraw_joblib.models import ProviderRecord to the top of test_providers.py. Removed 10 duplicate function-local imports.
- docs: re-declare/re-declared -> redeclare/redeclared (codespell) - tests: split compound isinstance+len assert (PT018) - tests: extract error_body literal to keep line <=88 (E501) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@internal is a server-bootstrap namespace seeded at startup by register_internal_providers. Previously, any authenticated user could PUT /v1/joblib/rooms/@internal/providers and inject a row into the reserved namespace (201 Created). Gate in register_provider mirrors the existing DELETE-side policy (@internal providers cannot be deleted). 403 Forbidden for all callers, superuser or not. TDD: new tests RED before the fix, GREEN after. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…agic Three dependency sites used getattr(app.state, ..., None) as a defensive fallback: get_internal_provider_registry (deps:32), mint_internal_worker_token (deps:170, :175 for settings). Per reviewer: app.state.* attrs consumed by request-time deps must be assigned unconditionally in lifespan (to a value or to None). No getattr fallbacks, no half-populated-state shims. - database.py lifespan pre-inits internal_worker_user, internal_provider_registry, internal_registry to None up front so request-time access never AttributeErrors. - dependencies.py switches to direct app.state.X access. - conftest fixtures (_build_app, unguarded_client_factory) set internal_provider_registry + settings to satisfy the new contract. - test_mint_internal_worker_token_raises_runtime_error_on_missing_cache (specifically guarded the getattr fallback) is obsolete — the substantive user-is-None case is covered by test_read_remote_provider_works_with_no_internal_worker_cache. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Four lazy-import sites were flagged as defensive without a cycle or performance justification: - bootstrap.py:51,79 — from zndraw.database import _collect_providers Traced the import graph: zndraw.database only imports bootstrap lazily (lifespan, not module-top), so bootstrap can safely top-import zndraw.database. Same for zndraw_joblib.registry. - executor.py:80 — from zndraw_joblib.exceptions import ProviderExecutionFailed Called inside an except block; zndraw_joblib.exceptions has no back- import on the providers package. - executor.py:134-135 — import fsspec / DirFileSystem fsspec is the core dep of the filesystem provider; ~87 ms at startup is the correct place to pay it vs first-call-time. - test_providers.py:148 (and 9 duplicate sites) — User, ProviderRecord Consolidated to module-top imports. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
InternalProviderExecutor._resolve_handler was a one-line method called exactly once (from _run). Inlined the call to resolve_internal_provider_handler and dropped the wrapper method. Reviewer asked whether functools.partial fits here. It does not — the simpler fix is to remove the needless indirection entirely. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reviewer flagged executor.py:91 — the error-upload headers duplicated "application/problem+json" which is a property of the error class, not custom code at each call site. - Add ProblemDetail.MEDIA_TYPE ClassVar to both zndraw and zndraw_joblib copies of the ProblemDetail model. - Replace four literal "application/problem+json" sites (problem_exception_handler x2, router.py Response x2, executor.py upload headers) with ProblemDetail.MEDIA_TYPE. Auth/X-Request-Hash/X-Result-Status headers remain at the call site — those are upload-context, not error-context. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Four separate review comments consolidated: - database.py: drop _collect_providers() wrapper; call sites inline list(BUNDLED_PROVIDERS) directly. Hoist BUNDLED_PROVIDERS import to module top (no cycle — zndraw.providers does not import database). - result_backends.py: rename the cryptic _k() helper to _namespaced_key() across both RedisResultBackend and StorageResultBackend (and the one cross-backend access in CompositeResultBackend). - config.py: drop the two Pydantic "comment-docstring under the field" annotations on filebrowser_enabled / filebrowser_path; move the explanatory text to the Settings class docstring to match the rest of the file (which uses inline # comments). - executor.py: remove _extension_accepts_providers_kwarg and its runtime inspect.signature check. All subclass .run() overrides in selections.py (9 sites) and analysis.py (4 sites) now accept **_kwargs: t.Any — matching the base Extension.run signature. Caller can always pass providers= without a TypeError risk. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…s packages Reviewer flagged: the zndraw.exceptions and zndraw_joblib.exceptions packages both defined bit-identical ProblemDetail and ProblemError classes (same fields, same MEDIA_TYPE, same init signature). Fold the zndraw copies into re-exports from zndraw_joblib.exceptions. zndraw-specific extensions stay (ProblemType subclass with openapi_response and raise_for_client, plus problem_exception_handler auto-filling problem.instance from request.url.path — a real behaviour difference). Verified runtime identity: zndraw.exceptions.ProblemDetail is zndraw_joblib.exceptions.ProblemDetail -> True. Full test suite (1433 passed, 1 skipped) unchanged. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two separate review comments:
tests/zndraw/test_config.py:266
Reviewer: "just testing that pydantic-settings work. We can trust the
package and need no double testing here! Remove this and similar
non-tests!"
Dropped 12 "*_from_env" tests that only exercised pydantic-settings'
own env-var/type-coercion plumbing (storage, media_path, host, port,
simgen, worker, internal_url, filebrowser_*, task_queue_name, plus a
generic "picks up env changes" smoke). Kept:
* all test_default_*
* tests that exercise ZnDraw-specific behaviour (env_parse_none_str
disabled — "none" as literal string, customise_sources ordering,
pyproject.toml loader, /v1/config/global-settings endpoint).
32 → 20 tests, all still green.
tests/zndraw/worker/test_internal_loadfile_e2e.py:37
Reviewer: "why not use a fixture. check conftest, otherwise add
simple water file!"
Added `water_xyz` pytest fixture to tests/zndraw/worker/conftest.py
— writes a 3-atom H2O xyz to tmp_path and returns the path. Test
drops the inline ase.Atoms + ase.io.write call and uses the fixture.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Previously ``ensure_internal_providers`` only ran inside ``init_database`` (gated by ``settings.init_db_on_startup``). Existing production deployments — where the DB was bootstrapped before the @internal providers feature landed and ``init_db_on_startup=False`` so ``init_database`` never re-runs — started against DBs without any ``@internal:*`` ProviderRecord rows. Broker tasks registered fine in-process but the rows did not exist, so the frontend could not discover providers and lookup routes returned 404. Lift the seed step out into the lifespan after caching ``internal_worker_user``. Guarded by ``filebrowser_enabled`` (feature flag) and the cached user being present (init_database seeds both). Idempotent — the function already catches ``IntegrityError`` on the ``unique_provider`` constraint, safe under concurrent replica startup. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Member
Author
|
Triage of the latest review batch: Critical — ProblemDetail dedup: Already done in Important 1 — Important 2 — Important 3 — Migration for existing DBs: Real issue, fixed in Minors — acknowledged, not blocking this PR:
|
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.
Summary
Two bundled changes that landed on the same branch:
dockview-react: activity bars, side/bottom zones, panel registry, drag-to-bar with hover state, rooms panel rewrite, plot browser, filesystem panel, chat panel, etc. (55 commits)@internalProvider path (mirroring@internalextensions). Ships aFilesystemReadprovider registered on startup at@internal:filesystem:FilesystemRead, configurable viaZNDRAW_SERVER_FILEBROWSER_PATH(default".","none"disables). Addsregister_internal_providers+ensure_internal_providershelpers,InternalProviderExecutor, taskiq dispatch fork inread_provider, visibility widening, and auseHasFilesystemProvidershook that hides the filesystem activity-bar icon when no providers are registered.Plan:
docs/superpowers/plans/2026-04-17-internal-providers.md.Test plan
tests/zndraw_joblibsuite (313 + 6 new tests) — greentests/zndraw/test_providers_executor.py— greentests/zndraw/test_config.py::TestFilebrowserPath— greenbun --bun tsc --noEmit— zero errorsuvx prek run --files <pr-files>— green on all PR-touched filestest_default_internal_filesystem_read+ 7 pre-existing extension-e2e tests timeout locally due to a taskiq-receiver issue that does not reproduce on CI (confirmed onmainlocally; CI run24327808136green). Leaving the regression-investigation out of scope for this PR.🤖 Generated with Claude Code
Summary by CodeRabbit
New Features
Bug Fixes
Tests