* feat: enable project-scoped preferences extension tabs
Add a typed extension schema so subprojects can define extra settings,
render them in the shared preferences drawer only when configured, and
consume them in playground as a real feature demo. Extension labels now
follow locale keys instead of hardcoded app-specific strings.
Constraint: Reuse the shared preferences drawer and field blocks
Rejected: Add app-specific fields to core preferences | too tightly coupled
Rejected: Inline localized label objects | breaks existing locale-key flow
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: Keep extension labels as locale keys rendered via $t in UI
Tested: Vitest preferences tests
Tested: Turbo typecheck for preferences, layouts, web-antd, and playground
Tested: ESLint for touched preferences and playground files
Not-tested: Manual browser interaction in playground preferences drawer
* fix: satisfy lint formatting for preferences extension demo
Adjust the playground preferences extension demo template so formatter and
Vue template lint rules agree on the rendered markup. This keeps CI green
without changing runtime behavior.
Constraint: Must preserve the existing demo behavior while fixing CI only
Rejected: Disable the Vue newline rule | would weaken shared lint guarantees
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Prefer computed/template structures that avoid formatter-vs-lint conflicts
Tested: pnpm run lint
Not-tested: Manual browser interaction in playground preferences extension demo
* fix: harden custom preferences validation and i18n labels
Tighten custom preferences handling so numeric extension fields respect
min, max, and step constraints. Number inputs now ignore NaN values,
and web-antd extension metadata uses locale keys instead of raw strings.
Also align tip-based hover guards in shared preference inputs/selects.
Constraint: Keep fixes scoped to verified findings only
Rejected: Broader refactor of preferences field components | not needed for these fixes
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Reuse the same validation path for updates and cache hydration
Tested: Vitest preferences tests
Tested: ESLint for touched preferences and widget files
Tested: Typecheck for web-antd, layouts, and core preferences
Not-tested: Manual browser interaction for all preference field variants
* fix: remove localized default from playground extension config
Drop the hardcoded Chinese default value from the playground extension
report title field and fall back to an empty string instead. This keeps
extension config locale-neutral while preserving localized labels and
placeholders through translation keys.
Constraint: Keep the fix limited to the verified localized default issue
Rejected: Compute the default from runtime locale in config | unnecessary for this finding
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Avoid embedding localized literals in extension default values
Tested: ESLint for playground/src/preferences.ts
Tested: Oxfmt check for playground/src/preferences.ts
Not-tested: Manual playground preferences interaction
* docs: document project-scoped preferences extension workflow
Add Chinese and English guide sections explaining how to define,
initialize, read, and update project-scoped preferences extensions.
Also document numeric field validation and point readers to the
playground demo for a complete example.
Constraint: Keep this docs-only and aligned with the shipped API
Rejected: Update only Chinese docs | would leave English docs inconsistent
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep zh/en examples and playground demo paths synchronized
Tested: git diff --check; pnpm build:docs
Not-tested: Manual browser review of the rendered docs site
* fix: harden custom preferences defaults and baselines
Use a locale-neutral default for the web-antd report title.
Also stop preference getters from exposing mutable baseline
or extension schema objects, and add a regression test for
external mutation attempts.
Constraint: Keep behavior compatible with the shipped preferences API
Rejected: Return raw refs with readonly typing only | callers could still mutate internals
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep defensive copies for baseline and schema getters unless storage semantics change
Tested: eslint, oxlint, targeted vitest, filtered typecheck, git diff --check
Not-tested: Full monorepo typecheck and test suite
* test: relax custom preference cache key matching
Avoid coupling the custom-number cache test to one exact
localStorage key string. Match the intended cache lookup
more loosely so the test still verifies filtering behavior
without depending on the full namespaced cache key.
Constraint: Focus the test on cache filtering behavior
Rejected: Assert one exact key | brittle with namespace changes
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Prefer behavior tests over literal storage keys
Tested: targeted vitest, eslint, git diff --check
Not-tested: Full monorepo test suite
---------
Co-authored-by: caisin <caisin@caisins-Mac-mini.local>
* feat(@vben-core/form-ui): support schema valueFormat on getValues
Some form fields emit UI-friendly structures such as time-range arrays,
while consumers and backend APIs often need a different payload shape.
This adds schema-level `valueFormat` hooks so `getValues()` can
normalize field output at read time without forcing callers to
post-process every submission path.
Constraint: Must preserve existing range-time mapping and nested field behavior
Constraint: Must not mutate live vee-validate form state while formatting output
Rejected: Global formatter config | too coarse for per-field payload shaping
Rejected: Post-submit-only transform | misses reset/query/change handlers
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep `getValues()` output derivation side-effect free
Directive: Clone raw form values before formatting derived payloads
Tested: vitest form-api test for valueFormat and existing getValues paths
Tested: oxlint on changed form-ui source and test files
Not-tested: Full repo typecheck baseline has unrelated .vue module resolution errors
* fix(@vben-core/form-ui): restore mount compatibility and share field path parsing
Follow-up review found two real regressions and one missing assertion in the
new value formatting flow. `FormApi.mount()` had become breaking by requiring
`componentRefMap`, and delete path resolution duplicated field-name parsing
instead of sharing the reader grammar. This patch restores backward
compatibility, centralizes field-name path parsing, and extends the test to
prove formatting does not mutate live form values.
Constraint: Must preserve current valueFormat behavior and nested field support
Constraint: Must not reintroduce mutation of live vee-validate values
Rejected: Keep duplicated delete parsing | risks grammar drift from read path
Rejected: Only loosen mount tests | would leave consumer-facing API breakage
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Reuse shared field-name parsing for read/delete semantics in form-ui
Tested: vitest form-api test suite
Tested: oxlint on changed form-ui files
Not-tested: Full repo typecheck baseline has unrelated .vue module resolution errors
EOF && git push hekx feature-form-value-format
* fix(@vben-core/form-ui): clear stale component refs on unmount
A follow-up review found that `unmount()` left the private component ref map
populated. Because `mount()` now accepts an optional `componentRefMap`, a later
mount without a new map could silently reuse stale refs from a prior form
instance. This change clears the ref map on unmount and adds a regression test
covering remount behavior without a new ref map.
Constraint: Must preserve backward-compatible optional `mount()` ref map behavior
Constraint: Focus and field-ref lookups must not observe stale refs after unmount
Rejected: Clear refs only during next mount | stale state would still leak between lifecycle calls
Rejected: Remove mount fallback entirely | would undo the compatibility fix
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: When mount falls back to internal refs, unmount must always reset that cache
Tested: vitest form-api test suite
Tested: oxlint on changed form-api source and test files
Not-tested: Full repo typecheck baseline has unrelated .vue module resolution errors
* refactor(@vben-core/form-ui): trim redundant valueFormat plumbing
Review feedback identified a few small cleanups in the value formatting path.
This removes an unnecessary shallow clone in `getValues()`, reuses the
already-parsed `rawKey` from `resolveFieldNamePath()` instead of re-resolving
it in multiple helpers, and clarifies the `FormValueFormat` contract for
undefined-as-delete decomposition behavior.
Constraint: Must not change runtime valueFormat behavior or payload shape
Constraint: Documentation and helper cleanup should stay behavior-preserving
Rejected: Leave duplicate raw-key resolution in place | adds needless parsing churn
Rejected: Expand the formatter API further | outside the scope of this cleanup
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep read/format helper plumbing lean and avoid duplicate field-name parsing
Tested: vitest form-api test suite
Tested: oxlint on changed form-ui source and test files
Not-tested: Full repo typecheck baseline has unrelated .vue module resolution errors
* feat(@vben-core/form-ui): document valueFormat with live examples
The new `valueFormat` feature needed a concrete usage path in both the
playground and the docs so users can understand how raw component values differ
from the final payload returned by `getValues()`. This adds a dedicated form
example, wires it into the playground menu, and documents the API with an
interactive docs demo. The preview panels now stay in sync when values are set,
reset, or submitted.
Constraint: Must demonstrate both return-value and setValue decomposition flows
Constraint: Example previews must react to setValues, reset, and manual edits
Rejected: Only document via markdown snippet | insufficient for verifying live payload behavior
Rejected: Reuse an existing basic form page | would bury feature-specific behavior
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep playground and docs demos behaviorally aligned when extending valueFormat examples
Tested: eslint on playground/docs valueFormat demo files and route module
Tested: oxlint on playground route module
Not-tested: Full docs/playground app runtime was not launched in this session
* chore(@vben-core/form-ui): normalize valueFormat demo formatting
The previous feature/docs commit left a few formatter-only adjustments unstaged
after hooks rewrote line wrapping in the new demo and docs pages. This commit
captures those final non-behavioral formatting updates so the branch matches the
current working tree.
Constraint: Must not change runtime behavior or docs meaning
Rejected: Leave post-hook diffs unstaged | branch would not reflect local state
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: After hook-driven rewrites, verify the working tree is clean before final push
Tested: Git diff inspection of remaining changes
Not-tested: No additional runtime verification needed; formatting-only follow-up
EOF && git push hekx feature-form-value-format
* fix(@vben-core/form-ui): remove docs demo dayjs dependency
The docs valueFormat demo imported `dayjs` directly even though the docs
package does not declare it as a dependency. That caused `@vben/docs:build`
to fail in CI during VitePress bundling. This change removes the direct
import, keeps the preview formatter generic for day-like values, and drops
the docs-only preset button that required constructing dayjs instances.
Constraint: Docs build must succeed without adding new package dependencies
Constraint: Playground example should remain unchanged and fully interactive
Rejected: Add dayjs to docs dependencies | unnecessary for a small display demo
Rejected: Externalize dayjs in VitePress build | hides a package boundary issue
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Docs demos should avoid imports only available through transitive deps
Tested: pnpm exec eslint docs/src/demos/vben-form/value-format/index.vue
Tested: pnpm --dir docs run build
Not-tested: No browser-side manual verification of the docs demo in this session
---------
Co-authored-by: caisin <caisin@caisins-Mac-mini.local>
* fix(menu): update hover color variable to use the correct reference
Medium Severity
In the horizontal .is-light menu section, --menu-item-hover-color is set to hsl(var(--menu-item-color)), but --menu-item-color is already defined as hsl(var(--accent-foreground)). This results in hsl(hsl(...)) at computed-value time, which is invalid CSS. The non-horizontal .is-light block correctly uses var(--menu-item-color) without the extra hsl() wrapper.
* fix(menu): simplify hover styles by removing redundant nested hover rules
Low Severity
The SCSS &:not(.is-active):hover { &:hover { ... } } compiles to a :hover:hover pseudo-class chain, which is functionally identical to a single :hover. The inner &:hover nesting is redundant and adds unnecessary complexity compared to placing styles directly inside the &:not(.is-active):hover block.
Caisin