@ids-group-ltd/ids-design-system 0.9.1 → 0.10.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +970 -3
- package/components.md +13 -13
- package/fesm2022/ids-group-ltd-ids-design-system.mjs +1473 -910
- package/fesm2022/ids-group-ltd-ids-design-system.mjs.map +1 -1
- package/migrations/0.10.0.json +407 -0
- package/migrations/0.10.1.json +159 -0
- package/migrations/0.6.1.json +9 -0
- package/migrations/0.6.2.json +24 -0
- package/migrations/0.8.0.json +4 -2
- package/migrations/0.9.0.json +27 -1
- package/migrations/0.9.1.json +10 -0
- package/migrations/AUTHORING.md +8 -0
- package/migrations/README.md +6 -3
- package/package.json +1 -1
- package/styles/_chip-row.scss +32 -0
- package/styles/_dropdown-overlay.scss +9 -2
- package/styles/_field-group-segment.scss +29 -0
- package/styles/_field-state.scss +38 -0
- package/styles/_focus.scss +8 -0
- package/styles/_link.scss +21 -0
- package/styles/_tokens.scss +6 -9
- package/styles/_visually-hidden.scss +14 -0
- package/types/ids-group-ltd-ids-design-system.d.ts +593 -479
|
@@ -0,0 +1,407 @@
|
|
|
1
|
+
{
|
|
2
|
+
"version": "0.10.0",
|
|
3
|
+
"items": [
|
|
4
|
+
{
|
|
5
|
+
"id": "chip-row-wrapper-in-multi-value-triggers",
|
|
6
|
+
"kind": "behavior",
|
|
7
|
+
"old": "ds-autocomplete .autocomplete-trigger > ds-tag",
|
|
8
|
+
"new": null,
|
|
9
|
+
"note": "The chips of ds-autocomplete and ds-multi-select now render inside a <span class=\"chip-row\"> instead of sitting directly in the trigger. LAYOUT IS UNCHANGED ONLY ON [maxVisibleChips]=\"null\": there the wrapper is display:contents, so it adds no box and the chips remain flex items of the trigger exactly as before — spacing, wrapping and height are identical. In the MEASURING modes, which now includes the DEFAULT \"fit\", the row becomes display:flex with flex-wrap:nowrap and overflow:hidden and the trigger itself flips to nowrap, so layout there is NOT unchanged (see chips-cap-by-width-by-default). WHAT BREAKS is a CSS override that walks the structure: any selector using the direct-child combinator from the trigger to a chip (`.autocomplete-trigger > ds-tag`, `ds-multi-select .field-trigger > ds-tag`, and `:first-child`/`:last-child`/`:nth-child` written against the trigger's children) now matches nothing or matches the wrapper. ACTION: drop the combinator (`.autocomplete-trigger ds-tag`) or insert `.chip-row`. Positional selectors should target `.chip-row > ds-tag:first-child` — note the chips are no longer siblings of the filter input, the addon slots or the chevron, so `ds-tag + .autocomplete-input` style adjacency is gone with no replacement. NAME THE RIGHT TRIGGER when you rewrite a multi-select selector: its trigger element is `.field-trigger`, NOT `.trigger` — a selector written against `ds-multi-select .trigger` matches nothing and always did. Chips also carry a `data-chip` attribute and the '+N more' marker a `data-chip-overflow` attribute; both are how the width measurement finds them, so do not strip them, but they are fine to select on.",
|
|
10
|
+
"searchHints": [
|
|
11
|
+
"autocomplete-trigger > ds-tag",
|
|
12
|
+
"autocomplete-trigger>ds-tag",
|
|
13
|
+
"> ds-tag",
|
|
14
|
+
"ds-tag:first-child",
|
|
15
|
+
"ds-tag:last-child",
|
|
16
|
+
"ds-tag + ",
|
|
17
|
+
"chip-row",
|
|
18
|
+
"data-chip",
|
|
19
|
+
".field-trigger > ds-tag"
|
|
20
|
+
]
|
|
21
|
+
},
|
|
22
|
+
{
|
|
23
|
+
"id": "exported-class-members-grew",
|
|
24
|
+
"kind": "behavior",
|
|
25
|
+
"old": "extends AutocompleteComponent",
|
|
26
|
+
"new": null,
|
|
27
|
+
"note": "This release's class-member-collision entry, the counterpart of 0.8.0's and 0.9.0's: it covers every exported class whose member surface changed, so a consumer who EXTENDS a DS component resolves them in one pass. Subclassing a DS component is unsupported (design-system.md), but where it happens the names collide and the build stops: TS4114 demands `override`, TS2415 fires on a redeclared private. GAINED MEMBERS — 30 classes, per `npm run class-surface` against v0.9.1: AuthLayoutComponent (brandName, letterGlyph, taglineText); AutocompleteComponent (chipBudget, chipRow, chipsThatFit, firstEnabledIndex, fitsChipsToWidth, nextEnabledIndex, sizeResolved); MultiSelectComponent (chipBudget, chipRow, chipsThatFit, fitsChipsToWidth); AvatarComponent (config, initialsRule, initialsText, name); BadgeComponent (withDotResolved); BottomSheetComponent (backdropPress, onBackdropRelease); CardComponent (elevationResolved, onActivationKey, paddingResolved); FilterBarComponent (triggerSize); ButtonComponent, CheckboxComponent, DatePickerComponent, FieldGroupComponent, IconButtonComponent, IconComponent, InputComponent, NumberStepperComponent, RadioComponent, SelectComponent, SwitchComponent, TextareaComponent, TimePickerComponent (sizeResolved); ModalComponent (backdropPress, dividersResolved, onBackdropRelease, status); ModalCardComponent (dividersResolved, resolvedIcon, resolvedRole, status); SearchComponent (cancel, debounce, flush, onText, push, sizeResolved, text); SidenavComponent (brandName, letterGlyph, markGlyph); TableComponent (activeSortDir, activeSortKey, borderedResolved, clipWidth, exportStatusState, exportingState, fitsContent, loadingMode, showsEmptyRow, sortDirChange, sortKeyChange, sortState, stickyScrollbarResolved); TabsComponent (memoryKey, remember, rememberLastTab, rememberedName). ALSO GAINED, from the shared field-state work: `controlState` on the nine bordered field components (AutocompleteComponent, ColorPickerComponent, DatePickerComponent, InputComponent, MultiSelectComponent, OtpInputComponent, SelectComponent, TextareaComponent, TimePickerComponent), plus TimeColumnsComponent.partPicked and TimePickerComponent.onPartPicked / .pickedParts from the single-mode time panel. These were invisible to class-surface until its field pattern was widened in this same release: it matched only `readonly x = call(...)`, and `controlState` is assigned from a property (`= this.fieldMessage.controlState`). ADD BY HAND, because the tool still cannot see them: SearchComponent also declares the plain private fields `pending` and `pushed` — class-surface matches `readonly` fields and methods, and a redeclared private still fails with TS2415. LOST MEMBERS — four classes, the case an `override` notices differently because it now overrides nothing: ModalComponent.error and .onBackdrop; ModalCardComponent.error; BottomSheetComponent.stop; InputComponent.floating (removed with the floating label). SEPARATELY, and invisible to the report because the NAME survives: TableComponent.exporting and .exportStatus are still there but are now PUBLIC readonly signals instead of the protected writable signals a subclass could set, so a subclass that assigned to them no longer compiles. ACTION: rename the subclass's own member, or add `override` where the base's semantics genuinely match; for the lost ones, drop the override and use the public API (runExport(), exporting(), exportStatus(), status). Consumers who only use the <ds-...> elements in templates need no change.",
|
|
28
|
+
"searchHints": [
|
|
29
|
+
"extends ButtonComponent",
|
|
30
|
+
"extends TableComponent",
|
|
31
|
+
"extends ModalComponent",
|
|
32
|
+
"extends AutocompleteComponent",
|
|
33
|
+
"extends MultiSelectComponent",
|
|
34
|
+
"extends SearchComponent",
|
|
35
|
+
"override ",
|
|
36
|
+
"sizeResolved",
|
|
37
|
+
"TS4114",
|
|
38
|
+
"exporting",
|
|
39
|
+
"exportStatus",
|
|
40
|
+
"onBackdrop",
|
|
41
|
+
"controlState",
|
|
42
|
+
"partPicked",
|
|
43
|
+
"onPartPicked",
|
|
44
|
+
"pickedParts"
|
|
45
|
+
]
|
|
46
|
+
},
|
|
47
|
+
{
|
|
48
|
+
"id": "auth-layout-brand-slot-drops-width-auto",
|
|
49
|
+
"kind": "behavior",
|
|
50
|
+
"old": "ds-auth-layout [brand]",
|
|
51
|
+
"new": null,
|
|
52
|
+
"note": "The rule constraining ds-auth-layout's projected [brand] slot no longer declares `width: auto` — only the `max-height` it shipped with in 0.9.0. That declaration sat at specificity (0,2,0), so it out-ranked a width the consumer set on their own logo, and with the slot's default `max-height: none` it achieved nothing: an <img> with no CSS width already keeps its aspect ratio under a max-height. ACTION: none for almost everyone. You are affected ONLY if you both opted into `--ds-auth-layout-brand-max-height` AND set an explicit width on the element you project into [brand] — that logo scaled to keep its ratio before and will now honour your width, i.e. it can squash. Add `width: auto` to your own logo rule to restore the old result. ds-sidenav's identical slot gained the same max-height constraint in this release and deliberately does NOT carry width: auto either, so the two match.",
|
|
53
|
+
"searchHints": [
|
|
54
|
+
"--ds-auth-layout-brand-max-height",
|
|
55
|
+
"[brand]",
|
|
56
|
+
"ds-auth-layout",
|
|
57
|
+
"brand slot",
|
|
58
|
+
"logo width"
|
|
59
|
+
]
|
|
60
|
+
},
|
|
61
|
+
{
|
|
62
|
+
"id": "tabs-url-param-slugs-the-label",
|
|
63
|
+
"kind": "behavior",
|
|
64
|
+
"old": "ds-tabs[urlParam]",
|
|
65
|
+
"new": null,
|
|
66
|
+
"note": "ds-tabs' urlParam now SLUGS a tab's label for the query parameter instead of using it raw. A tab that declares no `value` used to put its display string into the address - ?tab=License%20Verification%20Status - and now yields ?tab=license-verification-status (kebab, lowercased, diacritics decomposed and dropped; a label with nothing sluggable in it, CJK or bare punctuation, keeps its raw text). WHO IS AFFECTED: only consumers who adopted urlParam in 0.9.0 AND left `value` off their ds-tab elements. Their existing links stop matching and land on the first tab, which is what a stale bookmark has always done - nothing throws. ACTION: either accept the new parameter values, or declare `value` on each ds-tab to pin the names permanently, which is the better fix regardless: a slug is still derived from display copy, so renaming a tab still moves the link. NOT CHANGED: tabClosed still emits the raw label, because a host matches it against its own list - the two identifiers are deliberately different now, and the docs said they were the same until this release.",
|
|
67
|
+
"searchHints": [
|
|
68
|
+
"urlParam",
|
|
69
|
+
"ds-tabs",
|
|
70
|
+
"?tab=",
|
|
71
|
+
"tabClosed",
|
|
72
|
+
"ds-tab"
|
|
73
|
+
]
|
|
74
|
+
},
|
|
75
|
+
{
|
|
76
|
+
"id": "modal-error-boolean-to-status",
|
|
77
|
+
"kind": "input-rename",
|
|
78
|
+
"old": "ds-modal[error]",
|
|
79
|
+
"new": "ds-modal[status]",
|
|
80
|
+
"note": "ds-modal and ds-modal-card replace `error: boolean` with `status: DsStatus` ('info' | 'success' | 'warning' | 'error'), because a boolean cannot express a warning or an info dialog and the rest of the library settled on that vocabulary in 0.9.0. MECHANICAL MAPPING, no judgement needed: [error]=\"true\" becomes status=\"error\"; a bare `error` attribute becomes status=\"error\"; [error]=\"false\" and any unset case become nothing at all (unset is neutral and renders exactly as today); a COMPUTED [error]=\"expr\" becomes [status]=\"expr && 'error'\" - the input accepts false and undefined as \"no status\", so the shorter && form binds under strictTemplates; [status]=\"expr ? 'error' : null\" is equally correct. ALSO CHANGED, and worth knowing before you look at a screenshot: a status now drives four things rather than one - the icon-badge tint, the title colour, a DEFAULT icon (error/warn/success/info) and the ARIA role, where error and warning make the dialog an `alertdialog`. Both the icon and the role are overridable: `icon` beats the status's pick, an explicit `dialogRole` beats the status's role. The default icon is deliberate, not decoration - a tint on its own signals by colour alone.",
|
|
81
|
+
"searchHints": [
|
|
82
|
+
"[error]",
|
|
83
|
+
"ds-modal",
|
|
84
|
+
"ds-modal-card",
|
|
85
|
+
"error=\"true\"",
|
|
86
|
+
"ModalComponent",
|
|
87
|
+
"dialogRole"
|
|
88
|
+
]
|
|
89
|
+
},
|
|
90
|
+
{
|
|
91
|
+
"id": "chips-cap-by-width-by-default",
|
|
92
|
+
"kind": "behavior",
|
|
93
|
+
"old": "ds-autocomplete[maxVisibleChips]",
|
|
94
|
+
"new": null,
|
|
95
|
+
"note": "CHANGED DEFAULT, visual: maxVisibleChips on ds-autocomplete and ds-multi-select now defaults to \"fit\" instead of null. A multi-value trigger used to grow a line per selection and push the rest of the form down the page; it now keeps as many chips as it is wide and collapses the rest into '+N more', re-splitting on resize. ACTION: nothing, unless a screen relied on the growth - then bind [maxVisibleChips]=\"null\" to show every chip again. TWO THINGS RIDE ALONG with the default and are easy to miss. (1) A trigger in any measuring mode - \"fit\" or a number - is now flex-wrap: nowrap, and its chips are held to one line each and CLIPPED rather than wrapped, so a chip carrying more than a short label gets squeezed; only null still wraps the row. (2) A NUMBER is no longer an exact count: it is a ceiling that is then measured, so a narrow trigger shows FEWER than N (see the separate maxVisibleChips-number-is-a-ceiling item). CHECK: a form where a trigger sits above other fields, any layout whose height was measured or asserted with several chips selected, and any chip whose content is longer than a word. ALSO SCOPED HERE: Applies identically to ds-multi-select[maxVisibleChips]; the scope names one element only because the manifest format takes one.",
|
|
96
|
+
"searchHints": [
|
|
97
|
+
"maxVisibleChips",
|
|
98
|
+
"ds-multi-select",
|
|
99
|
+
"ds-autocomplete",
|
|
100
|
+
"chips"
|
|
101
|
+
]
|
|
102
|
+
},
|
|
103
|
+
{
|
|
104
|
+
"id": "ultra-compact-compresses-container-pad",
|
|
105
|
+
"kind": "behavior",
|
|
106
|
+
"old": "[data-density=\"ultra-compact\"]",
|
|
107
|
+
"new": null,
|
|
108
|
+
"note": "The ultra-compact density tier now also compresses --ds-container-pad, the inner padding ds-card and ds-stat share, from 24px (--space-6) to 16px (--space-4). WHO IS AFFECTED: only screens that set [data-density=\"ultra-compact\"] on a container that holds a ds-card or ds-stat. The default and the other three tiers (compact, cozy, comfy) are untouched and still resolve to 24px, so an app that never opts into ultra-compact sees no change. WHY: the tier already compressed field padding and table-cell padding, so a card kept its full 24px frame while its own cells halved - the surface padding was the one it missed. ACTION, if you want the old value inside an ultra-compact container: re-declare --ds-container-pad: var(--space-6) on it. If instead you have per-component pins (--ds-card-pad or --ds-stat-pad) set to work around the shared default, they still win and are unaffected - but consider collapsing them into one --ds-container-pad at :root, which is the lever both components default to.",
|
|
109
|
+
"searchHints": [
|
|
110
|
+
"data-density",
|
|
111
|
+
"ultra-compact",
|
|
112
|
+
"--ds-container-pad",
|
|
113
|
+
"--ds-card-pad",
|
|
114
|
+
"--ds-stat-pad"
|
|
115
|
+
]
|
|
116
|
+
},
|
|
117
|
+
{
|
|
118
|
+
"id": "control-size-input-becomes-nullable",
|
|
119
|
+
"kind": "behavior",
|
|
120
|
+
"old": "size()",
|
|
121
|
+
"new": null,
|
|
122
|
+
"note": "The size input on fourteen controls (ds-button, ds-icon-button, ds-input, ds-textarea, ds-select, ds-autocomplete, ds-search, ds-date-picker, ds-time-picker, ds-number-stepper, ds-checkbox, ds-radio, ds-switch, ds-field-group) is now typed DsSize | null and defaults to null, so an unbound size can fall through to the new DS_CONFIG.control.size. NOTHING CHANGES IN A TEMPLATE: size=\"sm\" binds exactly as before, and a control with no size still renders md. WHAT CHANGES is reading the signal back - size() returns null when nothing was bound, where it returned 'md' before. THERE IS NO PUBLIC REPLACEMENT ACCESSOR, and this is the trap: each component holds a sizeResolved() but declares it `protected`, so `cmp.sizeResolved()` from a host, a viewChild or a test does NOT compile (TS2445). ds-field-group is the one exception, where sizeResolved is public. ACTION: keep the value you bound and read that, or resolve DS_CONFIG.control.size yourself; the built-in fallbacks are unchanged ('md' for size, 'flat' for card elevation, 16 for icon size, false for badge withDot and for table bordered / stickyScrollbar). The same nullable-plus-protected-resolver shape now applies to ds-card elevation and padding, ds-icon size, ds-badge withDot, ds-table bordered / stickyScrollbar, dividers on ds-modal and ds-modal-card, and brand / logoLetter / tagline on ds-sidenav and ds-auth-layout - all of those inputs now read back null when unbound, where they used to return the old default. SEPARATELY: if you implement DsFieldGroupContext yourself, its member is now sizeResolved rather than size.",
|
|
123
|
+
"searchHints": [
|
|
124
|
+
".size()",
|
|
125
|
+
"sizeResolved",
|
|
126
|
+
"DsFieldGroupContext",
|
|
127
|
+
"ds-field-group",
|
|
128
|
+
"viewChild"
|
|
129
|
+
]
|
|
130
|
+
},
|
|
131
|
+
{
|
|
132
|
+
"id": "table-sort-inputs-no-longer-a-model",
|
|
133
|
+
"kind": "removed-api",
|
|
134
|
+
"old": "TableComponent.sortKey.set()",
|
|
135
|
+
"new": null,
|
|
136
|
+
"note": "ds-table's sortKey and sortDir changed from model() to an input() plus a matching output() (sortKeyChange / sortDirChange), so the component can tell a host's write apart from its own — the fix for the lost sort indicator. TEMPLATES ARE UNAFFECTED: [(sortKey)] and [(sortDir)] still work, one-way [sortKey] with (sortChange) still works, and a static sortKey=\"project\" still works. WHAT BREAKS is TypeScript access to the signal: table.sortKey.set(...) and table.sortDir.set(...) no longer exist, because an input is read-only. ACTION: keep the sort in your own state and bind it in — which every server-paged list already does — or listen to (sortChange) and write your own signal. Reading table.sortKey() still works and returns what the host bound.",
|
|
137
|
+
"searchHints": [
|
|
138
|
+
"sortKey.set",
|
|
139
|
+
"sortDir.set",
|
|
140
|
+
"sortKeyChange",
|
|
141
|
+
"sortDirChange",
|
|
142
|
+
"ds-table",
|
|
143
|
+
"sortChange"
|
|
144
|
+
]
|
|
145
|
+
},
|
|
146
|
+
{
|
|
147
|
+
"id": "outline-button-fill-becomes-transparent",
|
|
148
|
+
"kind": "behavior",
|
|
149
|
+
"old": "ds-button[variant=outline]",
|
|
150
|
+
"new": null,
|
|
151
|
+
"note": "The outlined variants no longer paint an opaque surface: ds-button variant=\"outline\", ds-icon-button variant=\"outlined\" and the outlined half of ds-split-button now default their fill to `transparent` instead of `--surface-default`. WHY: an opaque white fill is invisible on a white page and wrong on every tinted one - on a --surface-secondary card, a table toolbar or an alert the button read as a sticker laid over the surface rather than a control drawn on it. Material 3 and Bootstrap both specify a transparent container for this variant. WHO IS AFFECTED: only buttons placed on a non-white ground; on the default page surface the rendering is identical. On ds-button and ds-icon-button the disabled state is transparent too (mixing a transparent fill toward --surface-disabled would yield a 72%-alpha box), so the border and label carry it; a DISABLED outlined ds-split-button is the exception and still paints --surface-muted, because its outline rules are gated on :not(:disabled) - unchanged by this release. Hover fills differ per component and all stay opaque: --c-subtitle on ds-button, --surface-secondary on ds-icon-button and ds-split-button. ACTION if a button sits on a photo, a gradient or any busy ground and needed the opaque fill: set --ds-button-bg-color: var(--surface-default) on that instance (or --ds-icon-button-bg-color / --ds-split-button-outline-bg-color). ALSO SCOPED HERE: The same default move applies to ds-icon-button[variant=\"outlined\"] and ds-split-button[variant=\"outline\"].",
|
|
152
|
+
"searchHints": [
|
|
153
|
+
"variant=\"outline\"",
|
|
154
|
+
"variant=\"outlined\"",
|
|
155
|
+
"--ds-button-bg-color",
|
|
156
|
+
"ds-split-button",
|
|
157
|
+
"--ds-split-button-outline-bg-color"
|
|
158
|
+
]
|
|
159
|
+
},
|
|
160
|
+
{
|
|
161
|
+
"id": "input-floating-removed",
|
|
162
|
+
"kind": "removed-api",
|
|
163
|
+
"old": "ds-input[floating]",
|
|
164
|
+
"new": null,
|
|
165
|
+
"note": "The `floating` input on ds-input is REMOVED, with no replacement. It was half-built: the label was drawn inside a taller control, but it never animated out of the field on focus, which is the whole point of a floating label -- so an empty field read as a permanently mislabelled one. REMOVE the attribute; the label then renders above the field like every other DS field, which is the format the rest of the library already uses. Note the height change that comes with it: a floating ds-input was --hit-touch (48px at default density) and it now takes the normal size scale (40px at `md`, --field-h-md -> --hit-cozy), so a row built around one may need its own height re-checked. Also gone with it: the five internal tokens that only this variant read -- --field-h-floating, --field-floating-label-fs, --field-floating-label-lh, --field-floating-label-top and --field-floating-pad-t. Overriding any of them is now inert; DELETE-LIST. Unrelated and UNTOUCHED: --ds-elevation-floating, which is the elevation role for toasts.",
|
|
166
|
+
"searchHints": [
|
|
167
|
+
"floating",
|
|
168
|
+
"[floating]",
|
|
169
|
+
"floating]=\"true\"",
|
|
170
|
+
"ds-input",
|
|
171
|
+
"floating-label",
|
|
172
|
+
"--field-h-floating",
|
|
173
|
+
"--field-floating-label-fs",
|
|
174
|
+
"--field-floating-label-lh",
|
|
175
|
+
"--field-floating-label-top",
|
|
176
|
+
"--field-floating-pad-t"
|
|
177
|
+
]
|
|
178
|
+
},
|
|
179
|
+
{
|
|
180
|
+
"id": "field-invalid-class-to-data-field-state",
|
|
181
|
+
"kind": "selector-rename",
|
|
182
|
+
"old": "ds-color-picker.invalid",
|
|
183
|
+
"new": "ds-color-picker[data-field-state='error']",
|
|
184
|
+
"note": "Validation state on a field's control is now one shared rule keyed off a `data-field-state` attribute on the host ('error' | 'warning' | 'success'), replacing three different spellings. The dead `.invalid` HOST CLASS is removed from ds-color-picker and ds-otp-input, and the `.invalid` class on ds-multi-select's .field-trigger with it -- nothing styled them any more. ACTION: a CSS override selecting `.invalid` on any of those three now matches nothing; rewrite it as [data-field-state='error']. The .error class is STILL EMITTED - on the wrapper by ds-input and ds-textarea, on the box by ds-select, ds-date-picker, ds-time-picker and ds-autocomplete - and .success by ds-input alone, the only control with a success state. Your selectors still match, but the library no longer PAINTS through them, so a rule written to stack on top of a DS rule now stands alone. NOT AFFECTED, and important because .invalid is one of this item's searchHints: ds-slider still binds [class.invalid] on its HOST and still paints through it (:host(.invalid) .fill / .thumb), and ds-field-group has its own [invalid] input and its own class. Leave both alone.",
|
|
185
|
+
"searchHints": [
|
|
186
|
+
"ds-color-picker.invalid",
|
|
187
|
+
"ds-otp-input.invalid",
|
|
188
|
+
"ds-multi-select .invalid",
|
|
189
|
+
".invalid",
|
|
190
|
+
"invalid]",
|
|
191
|
+
"data-field-state",
|
|
192
|
+
"ds-slider"
|
|
193
|
+
]
|
|
194
|
+
},
|
|
195
|
+
{
|
|
196
|
+
"id": "max-visible-chips-number-is-a-ceiling",
|
|
197
|
+
"kind": "behavior",
|
|
198
|
+
"old": "ds-autocomplete[maxVisibleChips]",
|
|
199
|
+
"new": null,
|
|
200
|
+
"note": "CHANGED SEMANTICS of an existing binding: a NUMBER passed to maxVisibleChips on ds-autocomplete or ds-multi-select is now a CEILING that is then measured, not an exact count. [maxVisibleChips]=\"2\" means 'at most two, and fewer when the trigger is too narrow to hold them beside the +N more marker and whatever follows'. It used to mean 'exactly two, whatever the width', which is what made a capped trigger on a phone run out of room and wrap the text input onto a second line - the growth the cap exists to prevent. ACTION: nothing for most, and a wide trigger still shows exactly N. You are affected if you relied on a numeric cap to guarantee a fixed chip COUNT regardless of width - that guarantee is deliberately gone, and the trigger's single-line HEIGHT is guaranteed instead. null remains the only mode that never measures. NOTE the stale wording still in the wild: the input's own JSDoc in both components, and the generated docs API table built from it, still say 'A NUMBER caps at exactly that many' - the manifest and CHANGELOG are correct, those are not. ALSO SCOPED HERE: Applies identically to ds-multi-select[maxVisibleChips].",
|
|
201
|
+
"searchHints": [
|
|
202
|
+
"maxVisibleChips",
|
|
203
|
+
"ds-autocomplete",
|
|
204
|
+
"ds-multi-select",
|
|
205
|
+
"chips"
|
|
206
|
+
]
|
|
207
|
+
},
|
|
208
|
+
{
|
|
209
|
+
"id": "max-visible-chips-zero-inverted",
|
|
210
|
+
"kind": "behavior",
|
|
211
|
+
"old": "ds-autocomplete[maxVisibleChips]",
|
|
212
|
+
"new": null,
|
|
213
|
+
"note": "INVERTED VALUE, silent: maxVisibleChips of 0 used to mean 'show every chip' (the old code read 'max && max > 0 ? slice(0, max) : all', and 0 is falsy) and now means 'show NO chips', putting the entire selection behind the '+N more' marker. Nothing throws; the trigger just empties. Applies to ds-autocomplete and ds-multi-select alike. DECISION RULE, not a find/replace: replace every 0 that can reach this input - a literal, or an expression that evaluates to 0 such as a count, a config value or a '?? 0' default - with the BINDING [maxVisibleChips]=\"null\". A grep for the literal misses the computed cases, so read the expression feeding the input, not just the template text. TRAP, verified in source: a STATIC maxVisibleChips=\"null\" is NOT equivalent and reintroduces the bug. The transform is 'raw === 'fit' ? 'fit' : raw == null ? null : numberAttribute(raw)', so the string 'null' is not == null, so it falls through to numberAttribute('null') -> NaN, and Math.min(Math.max(NaN, 0), total) -> NaN -> zero chips again. Only the property binding passes a real null. ALSO SCOPED HERE: Applies identically to ds-multi-select[maxVisibleChips].",
|
|
214
|
+
"searchHints": [
|
|
215
|
+
"maxVisibleChips",
|
|
216
|
+
"maxVisibleChips]=\"0\"",
|
|
217
|
+
"maxVisibleChips=\"0\"",
|
|
218
|
+
"ds-autocomplete",
|
|
219
|
+
"ds-multi-select",
|
|
220
|
+
"?? 0"
|
|
221
|
+
]
|
|
222
|
+
},
|
|
223
|
+
{
|
|
224
|
+
"id": "nullable-boolean-inputs-treat-null-as-unset",
|
|
225
|
+
"kind": "behavior",
|
|
226
|
+
"old": "ds-table[stickyScrollbar]",
|
|
227
|
+
"new": null,
|
|
228
|
+
"note": "CHANGED MEANING of a null/undefined BINDING on the inputs that swapped booleanAttribute for a nullable transform so they could fall through to DS_CONFIG: ds-table bordered and stickyScrollbar, ds-badge withDot, and dividers on ds-modal and ds-modal-card. booleanAttribute(null) is false, so [bordered]=\"maybeNull\" used to render as OFF; the same binding now reads as 'unset' and resolves config-then-built-in. For bordered / stickyScrollbar / withDot the built-in is false, so only an app that also sets the matching provideDsConfig section sees a change. For dividers the built-in is TRUE, so [dividers]=\"someNullable\" flips from no-rules to rules the moment the expression yields null or undefined. ACTION: audit bindings whose expression is optional - `obj?.flag`, `input()` with no default, a `| async` before it resolves - and coerce them yourself (`[dividers]=\"!!maybe\"`). A literal [bordered]=\"false\", a bare attribute and a static bordered=\"false\" are all unaffected. ALSO SCOPED HERE: The same swap applies to ds-table[bordered], ds-badge[withDot], ds-modal[dividers] and ds-modal-card[dividers] - check all five, not just the scoped one.",
|
|
229
|
+
"searchHints": [
|
|
230
|
+
"[dividers]",
|
|
231
|
+
"[bordered]",
|
|
232
|
+
"[stickyScrollbar]",
|
|
233
|
+
"[withDot]",
|
|
234
|
+
"?.",
|
|
235
|
+
"| async"
|
|
236
|
+
]
|
|
237
|
+
},
|
|
238
|
+
{
|
|
239
|
+
"id": "bottom-sheet-stops-swallowing-inside-clicks",
|
|
240
|
+
"kind": "behavior",
|
|
241
|
+
"old": "ds-bottom-sheet",
|
|
242
|
+
"new": null,
|
|
243
|
+
"note": "ds-bottom-sheet no longer stops click propagation out of the sheet. It used to carry (click)=\"stop($event)\" on the panel purely to keep an inside click from reaching the backdrop's own click listener; that listener is gone, replaced by the pointerdown+pointerup pair that fixes the mouse-up dismissal, so the guard went with it. The sheet renders INLINE rather than through an overlay portal, so its clicks now bubble to whatever ancestor of yours wraps it. ACTION: if you have a (click) handler on an element that contains a ds-bottom-sheet - a card, a row, a section that opens something - it will now fire for clicks made inside the sheet. Move the handler, or stop propagation on your own side. The `stop` member is also gone from the class, which matters only to a subclass that overrode it.",
|
|
244
|
+
"searchHints": [
|
|
245
|
+
"ds-bottom-sheet",
|
|
246
|
+
"(click)",
|
|
247
|
+
"stopPropagation",
|
|
248
|
+
"BottomSheetComponent",
|
|
249
|
+
"stop("
|
|
250
|
+
]
|
|
251
|
+
},
|
|
252
|
+
{
|
|
253
|
+
"id": "error-string-alone-now-paints-the-control",
|
|
254
|
+
"kind": "behavior",
|
|
255
|
+
"old": "ds-input[error]",
|
|
256
|
+
"new": null,
|
|
257
|
+
"note": "An 'error' string now marks the field invalid on its own, across the fourteen field components that route through resolveFieldMessage. Before, error=\"...\" coloured only the message and left the control neutral; it now drives the error border, the tinted fill, the error focus ring and aria-invalid - border and fill on the nine BORDERED fields, aria-invalid and the message state on the other five. ACTION: nothing if you bound [invalid] alongside it. You are affected if you bound 'error' ALONE and relied on the box staying neutral: move that text to 'warning' for an advisory message, or to 'hint', and leave 'error' empty. ALSO CHANGED, same entry: the ladder is now error -> warning -> success, so a ds-input carrying both [success]=\"true\" and a warning string paints the WARNING, where it used to paint success and ignore the warning.",
|
|
258
|
+
"searchHints": [
|
|
259
|
+
"error=\"",
|
|
260
|
+
"[error]",
|
|
261
|
+
"ds-input",
|
|
262
|
+
"ds-select",
|
|
263
|
+
"ds-date-picker",
|
|
264
|
+
"warning=\"",
|
|
265
|
+
"[success]"
|
|
266
|
+
]
|
|
267
|
+
},
|
|
268
|
+
{
|
|
269
|
+
"id": "universal-focus-ring-base-rule",
|
|
270
|
+
"kind": "behavior",
|
|
271
|
+
"old": ":focus-visible",
|
|
272
|
+
"new": null,
|
|
273
|
+
"note": "A single zero-specificity base rule - :where(:focus-visible) { outline: none; box-shadow: var(--ds-control-focus-shadow) } - now gives the DS focus ring to ANY focusable element that does not style focus itself, including an app's own bare <button>, a link in projected content, and the close buttons inside ds-alert / ds-banner / ds-drawer. Before, those fell through to the browser default. ACTION only if you deliberately relied on an element having NO focus indicator: give it box-shadow: none explicitly. Note the rule also sets outline: none, so an element that was relying on the BROWSER's outline loses it - declare your own outline if that is what you wanted. Written with :where() so every component rule and every app rule still out-ranks it, and the data-focus switch is unaffected.",
|
|
274
|
+
"searchHints": [
|
|
275
|
+
":focus-visible",
|
|
276
|
+
":focus",
|
|
277
|
+
"outline: none",
|
|
278
|
+
"box-shadow: none",
|
|
279
|
+
"--ds-control-focus-shadow"
|
|
280
|
+
]
|
|
281
|
+
},
|
|
282
|
+
{
|
|
283
|
+
"id": "backdrop-dismissal-needs-press-and-release",
|
|
284
|
+
"kind": "behavior",
|
|
285
|
+
"old": "ds-modal",
|
|
286
|
+
"new": null,
|
|
287
|
+
"note": "ds-modal and ds-bottom-sheet no longer dismiss on a mouse-UP that lands on the backdrop after a press that started INSIDE the dialog - selecting text and letting go past the edge, or overshooting a slider. Dismissal now requires the press AND the release to land on the backdrop itself, both with the primary button. Nothing to do in application code. THE ONE CONSUMER-CODE BREAK IS IN TESTS: a test that dismisses with backdropElement.click() must now dispatch pointerdown + pointerup on the backdrop, each with button: 0. A real mouse or touch has always sent both. ds-drawer and the ModalService confirm were never affected - their scrim is a SIBLING of the panel rather than a wrapper - and are unchanged.",
|
|
288
|
+
"searchHints": [
|
|
289
|
+
"backdrop",
|
|
290
|
+
".click()",
|
|
291
|
+
"dispatchEvent",
|
|
292
|
+
"pointerdown",
|
|
293
|
+
"pointerup",
|
|
294
|
+
"ds-modal",
|
|
295
|
+
"ds-bottom-sheet"
|
|
296
|
+
]
|
|
297
|
+
},
|
|
298
|
+
{
|
|
299
|
+
"id": "modal-card-dialog-role-default-becomes-null",
|
|
300
|
+
"kind": "behavior",
|
|
301
|
+
"old": "ds-modal-card[dialogRole]",
|
|
302
|
+
"new": null,
|
|
303
|
+
"note": "CHANGED DEFAULT: dialogRole on ds-modal-card was input<'dialog'|'alertdialog'>('dialog') and is now input<'dialog'|'alertdialog'|null>(null), so 'unset' is distinguishable from a deliberate 'dialog'. What RENDERS is unchanged while no status is set - unset still resolves to role=\"dialog\". What changed is that a status of 'error' or 'warning' now resolves to 'alertdialog' unless you set dialogRole explicitly, and that reading dialogRole() back returns null rather than 'dialog' when nothing was bound. NOTE ds-modal has NO dialogRole input at all, so <ds-modal status=\"error\"> becomes an alertdialog with nothing to override it; use ds-modal-card directly if you need the role back.",
|
|
304
|
+
"searchHints": [
|
|
305
|
+
"dialogRole",
|
|
306
|
+
"ds-modal-card",
|
|
307
|
+
"alertdialog",
|
|
308
|
+
"role=\"dialog\""
|
|
309
|
+
]
|
|
310
|
+
},
|
|
311
|
+
{
|
|
312
|
+
"id": "avatar-tint-chroma-to-saturation",
|
|
313
|
+
"kind": "removed-api",
|
|
314
|
+
"old": "avatarTint(seed, { chroma })",
|
|
315
|
+
"new": "avatarTint(seed, { saturation })",
|
|
316
|
+
"note": "avatarTint now produces hsl(H 50% 62%) with a pinned white foreground, replacing oklch(62% 0.14 H) with an auto-chosen black/white foreground. THREE option changes: (1) `chroma` is REMOVED -- use `saturation`, an HSL percentage, default 50; a leftover `chroma` is now silently ignored, so this one does not fail the build. (2) `lightness` keeps its name AND its default of 62, but now means HSL lightness rather than OKLCH lightness -- the same number produces a different colour, so re-pick any override rather than assuming it carried across. (3) `foreground` now defaults to '#fff' instead of picking black-or-white by luminance; pass it explicitly for anything else. `avoidHues` and the seed hashing are unchanged, so a given seed keeps its hue. ACCESSIBILITY, deliberate and accepted: white initials on the lighter hues measure as little as 1.65:1, below the 4.5:1 WCAG AA asks of text -- all 29 sampled hues in the 20-300 range fall short. If your avatars carry initials as the ONLY identifier, override `lightness` (about 50) and keep white, which measures 5.3-6.6:1.",
|
|
317
|
+
"searchHints": [
|
|
318
|
+
"avatarTint",
|
|
319
|
+
"chroma",
|
|
320
|
+
"AvatarTintOptions",
|
|
321
|
+
"oklch",
|
|
322
|
+
"avoidHues",
|
|
323
|
+
"lightness",
|
|
324
|
+
"foreground"
|
|
325
|
+
]
|
|
326
|
+
},
|
|
327
|
+
{
|
|
328
|
+
"id": "radio-group-stops-handling-enter",
|
|
329
|
+
"kind": "behavior",
|
|
330
|
+
"old": "ds-radio-group",
|
|
331
|
+
"new": null,
|
|
332
|
+
"note": "REMOVED KEY HANDLING: ds-radio-group no longer handles Enter. It used to treat Enter as 'select the focused radio' and call preventDefault(), which is neither native radio behaviour nor what WAI-ARIA APG asks for, and inside a form it silently removed the keyboard's only way to submit. Space still selects, and the arrow keys still move the selection. ACTION: nothing for almost everyone, and forms get their Enter-to-submit back for free. You are affected only if you documented or tested Enter as the way to pick a radio in this group - retarget that to Space. A test driving Enter at a ds-radio-group now falls through to whatever the form does with it.",
|
|
333
|
+
"searchHints": [
|
|
334
|
+
"ds-radio-group",
|
|
335
|
+
"keydown.enter",
|
|
336
|
+
"'Enter'",
|
|
337
|
+
"radio",
|
|
338
|
+
"submit"
|
|
339
|
+
]
|
|
340
|
+
}
|
|
341
|
+
],
|
|
342
|
+
"informational": [
|
|
343
|
+
"ds-autocomplete and ds-multi-select cap their chips by MEASURED width: one line holds as many as the trigger is wide and the rest collapse into '+N more', re-split whenever the trigger resizes. This is now the DEFAULT (see the chips-cap-by-width-by-default item) - [maxVisibleChips]=\"null\" restores \"show everything\", and a NUMBER adds an upper bound on top of the same measuring rather than pinning an exact count. Choose by what has to stay predictable: \"fit\" lets the count follow the width, a number caps it as well, and null is the only mode that shows every chip and the only one that never measures.",
|
|
344
|
+
"The measurement behind maxVisibleChips=\"fit\" reads the TRIGGER's width, never the chip row's own - a row cannot measure a box whose size the cut changes. Widths are cached per selection and a geometry guard skips re-measuring when neither width nor count moved, so a page of pickers measures once, not once per chip. [maxVisibleChips]=\"null\" opts out of measuring entirely.",
|
|
345
|
+
"DELETE-LIST: if you wrote a ResizeObserver, a container query or a hand-tuned per-breakpoint maxVisibleChips number to keep a chip trigger on one line -> maxVisibleChips=\"fit\" replaces it, and it re-splits on any width change rather than at the breakpoints you listed.",
|
|
346
|
+
"DELETE-LIST: if you capped a chip trigger's height with max-height plus overflow:hidden to stop it growing a line per selection -> that clips a chip mid-glyph and hides the count; maxVisibleChips (a number, or \"fit\") drops whole chips and reports how many are hidden.",
|
|
347
|
+
"ds-sidenav's brand row fixes, none of which need action: an astral first character (emoji, rarer CJK) no longer paints as tofu in the letter-mark; the mark is aria-hidden, so a screen reader reads the wordmark once instead of announcing the glyph then the name; and a logoLetter set WITHOUT a brand string now renders its tile at all, where an outer gate used to drop the whole lockup.",
|
|
348
|
+
"DELETE-LIST: if you constrained a logo projected into ds-sidenav's [brand] slot with your own max-height (or a wrapper sized to hold it) -> --ds-sidenav-brand-max-height now does it, defaulting to none so nothing changes until you set it.",
|
|
349
|
+
"env(safe-area-inset-*) reads 0 unless YOUR index.html sets viewport-fit=cover in its viewport meta — a notched device alone is not enough, and this library cannot set it for you. ds-auth-layout pads by all four insets and ds-bottom-sheet by the bottom one, so both are plain-padded until you opt in; the docs site now sets it and both component pages state it. ds-bottom-sheet's env() also gained the `, 0px` fallback its siblings had: an unknown env() variable invalidates the entire declaration, which would have taken that sheet's horizontal padding with it.",
|
|
350
|
+
"The ds-select and ds-autocomplete panels now resolve --ds-dropdown-panel-radius, defaulting to the --ds-radius-panel role so nothing moves until you set it. TWO panels, not three: ds-multi-select's panel hardcodes border-radius: var(--radius-md) on its own .panel element, which is not the .dropdown-panel the shared rule targets, so neither this token nor --ds-radius-panel reaches it - it stays at 8px whatever you set, and that is a gap rather than a decision. This is the lever 0.9.0 notes promised as --ds-select-panel-radius, which never existed - that note is corrected in migrations/0.9.0.json. Deliberately panel-level rather than per-component: retuning ONE picker is not a thing the library encourages, and --ds-radius-panel moves modal, popover and date-picker with it. Scope it with a selector - ds-select { --ds-dropdown-panel-radius: 8px } - and only that panel moves, since the token is forwarded onto the portaled pane.",
|
|
351
|
+
"DELETE-LIST: if you wrote a global rule against .ds-dropdown-panel or .cdk-overlay-pane to change a dropdown corner -> --ds-dropdown-panel-radius does it, and scopes to one picker when written under a component selector.",
|
|
352
|
+
"A single-select ds-autocomplete no longer keeps the text typed to filter. Since 0.9.0 that text was also INVISIBLE - the filter input collapses to zero width once a chosen label is on show - while it still narrowed the list, so reopening a settled picker offered one option of four with nothing on screen to explain it. Cleared now whenever the panel closes over a displayed label, which covers both picking and dismissing after typing over a settled value. Multi-select is unchanged on purpose: its input never collapses, so the text stays visible and useful for the next pick. No action - query was component-private and could not be reset from an app.",
|
|
353
|
+
"DELETE-LIST: if you queried a ds-autocomplete inner <input> and dispatched a synthetic input event to clear leftover filter text after a pick -> the component clears it itself now, and clear() has been public since 0.9.0 for emptying value and text together.",
|
|
354
|
+
"The global .link class now resets a <button>: appearance, padding, border, background, font-family, font-size, line-height and text-align. It was written for anchors and touched none of them, yet ds-button warns variant=\"text\" off in favour of .link and an action that is not navigation has to be a button - so a button.link rendered at 13.3px inside 14px text with 1px 6px of padding and an outset border. font-weight is untouched so .quiet/.standalone still decide it, and display stays inline-block. Scoped with :where(), so the reset sits at .link specificity and a consumer rule still wins on source order.",
|
|
355
|
+
"DELETE-LIST: if you have a local .link-button class, or a padding/border/background/font reset on button.link in your overrides -> .link ships it now. mm-broker carried exactly those four lines after replacing three near-duplicate local classes with \"link sm standalone\".",
|
|
356
|
+
"ds-stat gained --ds-stat-value-white-space and --ds-stat-value-font-size, both defaulting to what the tile already rendered (nowrap, and h2/h4/h1 by size), so no existing tile moves. They exist because a value that is a PHRASE - a date range, a status - spilled straight past the tile edge under the nowrap that suits a number - the card sets no overflow, so it is not clipped, it escapes. The size hook now applies at every size input: the sm and lg variants move this token instead of restating font-size on .value, where they out-ranked any override and made such a hook work at md only.",
|
|
357
|
+
"DELETE-LIST: if you projected into ds-stat [dsStatValue] only to set white-space: normal or a smaller font-size on a text value -> the two hooks do it on the ordinary value input, and the tile keeps its geometry either way. Keep the slot for a value that is genuinely another component, like a badge.",
|
|
358
|
+
"ds-search takes [debounce] (milliseconds, default 0). At 0 it writes value on every keystroke exactly as before, so nothing changes until you ask. THREE things deliberately skip the timer, which is what a wrapped debounceTime usually gets wrong: emptying the field, the Clear button, and Enter - the last flushing a keystroke still in flight so `submitted` carries what the reader can see. Emptying also cancels a pending keystroke, so a fast type-then-clear cannot land the stale query afterwards. What the input SHOWS is never delayed at any setting, and a host writing value mid-typing wins and cancels the timer.",
|
|
359
|
+
"DELETE-LIST: if you wrapped ds-search to add a debounce - typically 200ms on typing with an immediate path when the field empties -> [debounce] does both, plus the Enter flush. mm-broker carried exactly that wrapper for exactly those two behaviours.",
|
|
360
|
+
"ds-filter-bar no longer clips the focus ring of its first and last inline filter. The measured row was overflow: hidden with each slot sized to exactly its declared width, so the row box started flush with the first control while the ring is an OUTSET shadow (3px on fields, 4px on controls) - interior filters were fine in the 12px gap, the outer two were cut off as you tabbed onto them. Now overflow: clip with a 4px clip margin; clip still forbids scrolling, so the width the collapse arithmetic reads is unchanged. Nothing to do, and it was not fixable from an app: the row is internal and padding it from outside would have changed that measured width.",
|
|
361
|
+
"ds-filter-bar takes triggerSize (sm|md|lg, default md) for its own \"+N\" trigger. The bar cannot infer it: dsFilterItem carries a WIDTH and the filters are arbitrary projected controls whose size it never sees, so a row of sm filters got an md trigger standing 8px taller under align-items: flex-end. DELETE-LIST: if you pinned --ds-button-height / --ds-button-font-size / --ds-button-pad-x under ds-filter-bar to force the trigger smaller -> triggerSize=\"sm\" replaces all three.",
|
|
362
|
+
"If you must hold two trailing controls in ONE component, the Filter bar page now shows the three declarations its :host owes (display: flex; align-items: flex-end; gap: var(--ds-filter-bar-gap, var(--space-3))) - the bar's own trailing rule, with the gap read from the existing public token so the copy cannot drift. The usual REASON for pairing them is also gone: a bar-level Clear no longer has to share a component with the search to cancel an armed debounce, because ds-search cancels its own pending keystroke when the host writes value.",
|
|
363
|
+
"DS_CONFIG gained a modal section: provideDsConfig({ modal: { dividers: false } }) sets the default for ds-modal and ds-modal-card app-wide, or on a route or subtree for the ELEMENTS, with an explicit [dividers] still winning. ONE LIMIT: ModalService is providedIn root and builds the confirm injector from the root, so only a ROOT-level provideDsConfig reaches a ModalService confirm - a route- or component-scoped one does not. It reaches where an attribute cannot - a ModalService confirm renders a ds-modal-card with no ds-modal ancestor to bind - which is why a per-element input was not enough. Internally [dividers] is now boolean | null so \"unset\" stays distinguishable from an explicit false; existing bindings are unaffected.",
|
|
364
|
+
"DELETE-LIST: if you repeat [dividers]=\"false\" on every modal (mm-broker has 41), or reach into .modal > .header|.body|.actions with !important because hiding the rules left the padding stacked -> provideDsConfig({ modal: { dividers: false } }) replaces both. Which padding collapses is deliberate: the BODY drops to zero and the header/actions keep theirs, because the body is the scroll container and a gap in its padding would scroll away and let content meet the title.",
|
|
365
|
+
"ds-tabs takes rememberLastTab (off by default), which reopens the tab a screen was last left on. Stored per path in sessionStorage under ds-tabs:{path}:{param}; inert without urlParam, which supplies the names because a remembered INDEX would repoint the moment a tab is inserted. A link always wins - the query parameter is read first and memory only fills the gap the first tab deliberately leaves. Every storage access is guarded, so Safari private mode simply means the strip stops remembering. DELETE-LIST: a helper that kept the last tab per path in sessionStorage can go.",
|
|
366
|
+
"DS_CONFIG gained a brand section: provideDsConfig({ brand: { name, logoLetter, tagline } }) answers ds-sidenav and ds-auth-layout together, since both take the same three and an app answers them identically on every screen. Each field resolves independently - an element input beats the config per FIELD, so overriding the name does not mean restating the letter, and brand=\"\" drops the wordmark while a logoLetter in the config still paints its tile (clear both for no brand row). The two-tone wordmark is deliberately NOT in the config: [dsBrandName] stays the accent's home, because a config string would have to be split apart to find the accent, and the projection wins over both. Shape: brand, logoLetter and tagline are now string | null with a null default so \"unset\" is distinguishable from an explicit empty; the input names did not change and template bindings are unaffected.",
|
|
367
|
+
"DELETE-LIST: if you wrote an adapter component that injects a brand object and fans it out into brand/logoLetter/tagline bindings on each auth screen -> provideDsConfig({ brand: … }) does it, and the accent stays a [dsBrandName] projection instead of a brandAccent string plus an indexOf split.",
|
|
368
|
+
"ds-avatar takes a `name`, which replaces the two bindings a caller used to wire by hand: it derives the initials AND seeds the tint. The tint hashes the NAME rather than the initials - two letters collide wholesale, so Alice Brown and Adam Bell would otherwise share a colour (measured: same \"AB\", hues 198 and 356). An explicit `initials` still wins, and `seed` still overrides the hash input when identity is an id or an email rather than the display name.",
|
|
369
|
+
"The rule turning a name into initials is REPLACEABLE rather than configurable by mode: [initialsRule] on one avatar or table, provideDsConfig({ avatar: { initials } }) for the app. The built-in stays initialsFrom - first two words - and is documented as a Latin-script guess, because reading a name is a cultural convention rather than a UI decision: a Spanish reader wants MG from \"Maria Jose Garcia Lopez\", an insurer wants BC from \"Blue Cross Blue Shield\". A library that shipped person/organisation modes would be claiming to know both, and would be wrong in ways it cannot see. DELETE-LIST: local initials helpers stay yours, but the ds-avatar wiring around them collapses to [name] plus one [initialsRule].",
|
|
370
|
+
"ds-multi-select and ds-autocomplete render an option's `icon` before its label at 16px, which previously only ds-select did. The three option types now extend one exported DsOption<T> - they were three near-identical interfaces, which is why the icon field existed on one picker for a release - and each keeps its own exported name, so no import changes. Nothing to do: options without an icon render as before. Only ds-select repeats the icon in its TRIGGER, because its trigger is a button showing one label; the autocomplete trigger is a text input whose leading slot belongs to [dsAutocompletePrefix], and multi-select chips stay text-only so the label keeps the width.",
|
|
371
|
+
"provideDsConfig gained five sections, so an app can state a whole-app opinion once instead of on every element: control.size (button, icon-button, input, textarea, select, autocomplete, search, date/time picker, number stepper, checkbox, radio, switch, field-group), card.elevation, card.padding, table.stickyScrollbar, table.bordered, icon.size and badge.withDot. Precedence is input -> config -> built-in, and NO built-in default moved: an app that provides nothing renders exactly as it did. DELETE-LIST: the attribute you repeat identically on every element - one consumer binds [stickyScrollbar]=\"true\" on all 35 of its tables and size=\"sm\" on 146 buttons.",
|
|
372
|
+
"DS_CONFIG.control.size is NOT the same axis as [data-density]. The density tier rewrites hit targets and padding in CSS for a route subtree; control.size picks which of the three declared sizes (sm/md/lg) a control renders at. An app can set both, and they compose.",
|
|
373
|
+
"MEASURED IN A CONSUMER, worth checking on your own side: 46 of its ds-icon elements bind [size]=\"16\", which is already the ds-icon default - dead bindings, not a gap. Config or no config, those can go.",
|
|
374
|
+
"ds-modal and ds-bottom-sheet no longer dismiss when the mouse is RELEASED on the backdrop after a press that started inside the dialog (selecting text and letting go past the edge, or overshooting a slider). Their backdrop wraps the dialog, so the browser fires click at the nearest common ancestor - the backdrop itself - which the target===currentTarget guard could not tell apart. Dismissal now requires the press AND the release to land on the backdrop. Nothing to do on adoption. ONE EXCEPTION: a test that dismisses a dialog with backdropElement.click() must now dispatch pointerdown + pointerup on the backdrop; a real mouse or touch always sends both. ds-drawer and the ModalService confirm were never affected - their scrim is a sibling of the panel - and are unchanged.",
|
|
375
|
+
"ds-table gained one projection slot, [toolbar], for controls that act on THAT table (a search box, a status filter, a refresh) — the row the export button already lived in, which until now existed only while exportButton was on. Project several controls and the row lays them out (flex, wrap, 8px gap); project a single element of your own and it takes the full row width, so your own layout is the only one in play. Visually additive: the row COLLAPSES when nothing filled it and exportButton is off, so a table that uses neither is pixel-identical. It is not DOM-identical - <div class=\"table-toolbar\"><div class=\"toolbar-slot\"> is now rendered unconditionally and hidden with display:none, so a ds-table > :first-child selector or a DOM snapshot sees it. Page-level filters still belong in a ds-filter-bar above the table.",
|
|
376
|
+
"ds-table now exposes runExport(), exporting() and exportStatus() publicly, so an Export button placed anywhere (page header, filter bar, the new toolbar slot) runs the same path as the built-in one: the exportFetcher call, the download, the in-flight state and the live-region announcement. DELETE-LIST: a consumer button wired to exportCsv() on a SERVER-PAGED table was exporting one page — exportCsv serialises the rows the table holds. Switch it to runExport() and pass exportFetcher.",
|
|
377
|
+
"ds-table no longer renders its empty row while loading is anything but false. Previously only loading=\"skeleton\" suppressed it, so an overlay reload with zero rows painted the empty state under the spinner. Nothing to do on adoption, and you can delete the @if(!loading) guard around your own empty template — worth checking whether you have one at all: of one consumer 14 overlay-loading tables, 9 shipped \"no rows\" copy during a load.",
|
|
378
|
+
"ds-table no longer loses its sort indicator when a host answers a cleared sort with its own default. Angular writes an input only when the bound expression changed (it diffs against the last value it wrote, which an internal .set() never updates), so whichever half of the host default equalled the last bound value was skipped: a host defaulting to desc had its desc dropped and the header marked nothing over sorted rows, while a host defaulting to asc was restored — the reason the same three clicks worked on one screen and not another. The marked state is now a linkedSignal over BOTH inputs. Nothing to do on adoption; you can delete any workaround that re-bound the sort after a clear.",
|
|
379
|
+
"ds-table accepts [loading]=\"true\" and resolves the treatment itself: skeleton rows while filteredData is empty, overlay once rows are on screen. The existing values are untouched — false, 'skeleton' and 'overlay' behave exactly as before, so nothing changes on adoption. DELETE-LIST: the computed every list screen writes beside its table, items().length > 0 ? 'overlay' : 'skeleton' — one consumer feature carried three copies of it. Pin 'skeleton' or 'overlay' only where you mean to override the rule.",
|
|
380
|
+
"Disabled rows and triggers are no longer dimmed twice. ds-select, ds-date-picker, ds-time-picker, ds-menu-item and the shared dropdown row each applied --text-disabled AND opacity 0.6 on top of it, which measured 1.68:1 against the surface where the colour alone gives 2.53:1 (an enabled control is 17.32:1). The opacity is gone from those five and the colour carries the state. A disabled calendar day and ds-toggle-button, where opacity is the ONLY signal, moved from --opacity-muted (0.6) to --opacity-disabled (0.5). DELETE-LIST: any override that retunes --opacity-muted to compensate for a washed-out disabled state — that token is a de-emphasis level for hints and secondary text and now carries no disabled state at all. NOTE for anyone who tried the same trick on ds-button or ds-icon-button: those never read --opacity-muted, so such an override was always a no-op — their disabled state is a set of oklch(from accent ...) colours.",
|
|
381
|
+
"ds-button and ds-icon-button use ONE disabled-label recipe for every variant now: oklch(from accent 70% 0.035 h), replacing the 58% base plus the 52% overrides that tonal and outline carried. Reason, measured across the full variant x colour x state matrix: the old lightness sat where a saturated accent already is, so on primary and error the disabled label differed from the live one by 1.16-1.60:1 on four of the five variants (neutral was fine because its enabled label is nearly black). It is now 2.40-2.71:1 on the coloured variants, and up to 6.46:1 on neutral. The disabled label is FAINTER against its own background than before (2.1-2.7:1 where it was 3.4-5.2) - that is the intended trade, and disabled text is exempt from the WCAG contrast floor. Nothing to do on adoption; --ds-button-label-color-disabled and --ds-icon-button-icon-color-disabled still override per component. DELETE-LIST: an override that retunes --opacity-muted on ds-button or ds-icon-button - neither component has ever read that token, so it was inert.",
|
|
382
|
+
"components.md — the API digest copied into dist/ds for a consumer's agents — was two renames behind: 46 of the 92 --ds-* tokens it named no longer existed (the bg->bg-color and fg->text/label/icon-color renames, plus --ds-pagehdr-* -> --ds-page-header-*), ds-button was documented with its pre-2026-06 single `variant` axis rather than color x variant, and the topbar section described a --ds-topbar-margin-b that has never existed in the shipped component. All names now verified against src/lib. If your agents or your notes were built from an older copy of this file, re-read it after the bump: token names you may have copied from it are dead.",
|
|
383
|
+
"Focus indicators are now universal: a single base rule, :where(:focus-visible) { outline: none; box-shadow: var(--ds-control-focus-shadow) }, gives the DS ring to anything focusable that does not style focus itself — the close buttons in ds-alert / ds-banner / ds-drawer, links in projected content, and an app's own bare <button>. Before this they fell through to the browser default (a black outline: auto in Chrome). Zero specificity via :where(), so every component rule and every app rule still wins, and the data-focus switch (ring / field / edge / outline) behaves exactly as before. ACTION only if you relied on an element having NO focus indicator: set box-shadow: none on it explicitly.",
|
|
384
|
+
"AutocompleteOption now extends the shared DsOption shape, so it gains `disabled?: boolean`, which it never had before — and ds-autocomplete acts on it: the row carries the .is-disabled class and aria-disabled, ignores the click, and the arrow keys skip it (as does the initial active row when the panel opens or the filter re-runs). Behaviourally this matches ds-select and ds-multi-select, which already honoured the flag. The row greys out as well - measured in the open panel at rgb(155,163,180) against rgb(22,26,40) for an enabled row.",
|
|
385
|
+
"TableColumn.width now accepts the literal 'content' in addition to a pixel number: the column sizes to its widest cell and stops wrapping, leaving the remaining width to the other columns. Purely additive — a numeric width and an unset width behave exactly as before. DELETE-LIST: a `::ng-deep ds-table { .table th, .table td { width: 1%; white-space: nowrap } }` override, which is the idiom this replaces. Notes: dragging a resizable column's edge overrides the mode (a drag is a deliberate pixel width), and `truncate` on a 'content' column has no clip width to work against, so it does nothing rather than clipping at the 240px default.",
|
|
386
|
+
"maxVisibleChips as a NUMBER is now a ceiling rather than an exact count on ds-autocomplete and ds-multi-select: at most N chips, and fewer when the trigger is too narrow to hold them beside the '+N more' marker and whatever follows (the text input, the chevron). Reported from a phone: [maxVisibleChips]=\"2\" showed two chips, ran out of width and wrapped the text input onto a second line — the growth the cap exists to prevent. Nothing to change on adoption; a wide trigger still shows exactly N. null is untouched and remains the only mode that does not measure. If you relied on a numeric cap to guarantee a fixed chip COUNT regardless of width, that guarantee is gone by design — the trigger's height is the thing being guaranteed instead.",
|
|
387
|
+
"ds-field-group segments now centre their value vertically. Every segment except ds-button pinned its value to the top of the row: the group stretches the segment, but the host is a block, so the height: 100% each control set on its bordered box resolved against `auto` and the box kept its own content height (a 19.6px line box inside a 40px row at md). Nothing to change on adoption. ACTION only if you had a ::ng-deep override compensating for it -- e.g. ds-field-group ds-input .control { align-items: center } or a padding-top nudge -- which is now double-applied; delete it.",
|
|
388
|
+
"An `error` string now marks the field invalid on its own, across all 14 field components that route through resolveFieldMessage (ds-input, ds-textarea, ds-select, ds-autocomplete, ds-multi-select, ds-date-picker, ds-time-picker, ds-color-picker, ds-otp-input, ds-slider, ds-checkbox, ds-radio, ds-radio-group, ds-switch). Before, error=\"...\" coloured only the message and left the control neutral; it now drives the error border, fill, focus ring and aria-invalid, which is what the input always documented. Nothing to do if you bound [invalid] alongside it. ACTION only if you bound `error` on its own and WANTED a neutral box: move that text to `warning` (advisory) or to `hint`.",
|
|
389
|
+
"A `warning` string now paints the control, not only the message line: a --warning-text border (#8A7400, 4.59:1 on a white field) on all nine bordered fields -- ds-input, ds-textarea, ds-select, ds-autocomplete, ds-multi-select, ds-date-picker, ds-time-picker, ds-color-picker, ds-otp-input. Success keeps its border-only treatment and error keeps border + tinted fill, so the three states stay ranked. Two visual corrections ride along and need no action: an invalid ds-autocomplete drew a near-black border because it read a --border-error token that has never existed, and ds-multi-select / ds-color-picker / ds-otp-input had the error border but not the fill their siblings had. Every per-component override token (--ds-input-bg-color-error, --ds-select-focus-shadow-error, and the rest) still applies.",
|
|
390
|
+
"ds-otp-input no longer leaves two characters in one cell when a second character is typed into an already-filled cell. The value the component reported was always correct -- only the rendered cell was stale -- so nothing to change on adoption. ACTION only if you worked around it by reading the cell inputs from the DOM instead of the component's value.",
|
|
391
|
+
"ds-password-rules: its screen-reader-only per-row text ('- not met') was rendering VISIBLY, because the .visually-hidden class it used was defined by no stylesheet -- ds-table, ds-date-picker and ds-stepper each declared their own private copy. One shared mixin now backs all four. The leaked text is also what wrapped every label in the two-column layout. Separately, the pending glyph changed from `minus` to a new `circle` icon and the row now transitions its colour over --duration-base. Nothing to change on adoption. ACTION only if you targeted the pending row's glyph by name, or added your own .visually-hidden definition to work around the leak -- that override is now redundant (DELETE-LIST).",
|
|
392
|
+
"The migration manifests themselves changed: 0.9.0.json gained a class-member-collision item it never had (fifteen classes plus four newly exported symbols), 0.8.0.json now names the private saveLayout/restoreLayout that actually stopped a consumer build, and 0.6.1 / 0.6.2 / 0.9.1 were written for the first time. Every released version now ships a manifest, so the auto-update planner no longer falls back to re-deriving a plan from changelog prose on those three. 0.6.2 turned out to carry something actionable nobody had in machine-readable form: a ds-bottom-sheet with nothing focusable inside it stopped closing on Escape.",
|
|
393
|
+
"ds-modal gained three public title tokens for the new statuses - --ds-modal-title-text-color-warning, -success and -info, defaulting to --warning-text / --success-text / --info-text - joining the -error that already existed. Additive: nothing moves until you set one.",
|
|
394
|
+
"The [empty] slot is now unreachable during ANY load, not just a skeleton one: the empty row is withheld whenever loading is not false, and the projection lives inside that branch. Copy you want a reader to see WHILE a load runs has to come from your own page, above or around the table.",
|
|
395
|
+
"runExport() without an exportFetcher falls through to exportCsv() synchronously and never touches exporting/exportStatus - so an external Export button bound to [loading]=\"table.exporting()\" shows no progress in that configuration. Bind exportFetcher if you want the in-flight state.",
|
|
396
|
+
"rememberLastTab is inert without urlParam AND outside a routed context; the key is ds-tabs:{path}:{param}, so two strips on one path stay apart. Only a user click or keypress writes the memory - a host driving selectedIndex writes neither the URL nor the memory, as before.",
|
|
397
|
+
"ds-search's Clear button now follows the text ON SCREEN rather than the debounced value, so at a non-zero [debounce] it appears on the first keystroke instead of after the delay.",
|
|
398
|
+
"On a resizableColumns ds-table the first drag OR resize keystroke anywhere locks a measured pixel width onto every column that has none, TableColumn width:'content' included. Home on a column's own handle hands that one back; the column panel's Reset to default button and clearing the public two-way columnWidths hand back all of them.",
|
|
399
|
+
"ds-time-picker in mode=\"single\" no longer renders a Done button: the panel closes itself once the user has picked BOTH an hour and a minute during that opening, and Now spans the full width of the footer beneath the columns. mode=\"range\" is unchanged -- it keeps Done and does not auto-close. Separately, hovering an already-selected hour or minute no longer paints it with the neutral hover fill (a specificity defect that left selected white text on pale grey, most visible on touch where the last-tapped cell keeps :hover). ACTION only if you scripted the panel closing via the Done button in single mode -- there is no such button now -- or overrode `.time-footer .time-action` widths, which now stretch when Done is absent.",
|
|
400
|
+
"The collapsed ds-sidenav no longer sets align-items: center on itself, which had stopped its projected children from stretching: every ds-sidenav-item shrank to its 20px icon inside a 40px rail, so the active tile rendered 20x40 (a lozenge, with the orange active rail down one side) and the alert dot sat above the glyph's middle. Items now fill the rail column, and the dot is anchored to the glyph's top-right corner through a new --sidenav-collapsed-glyph token rather than to the tile's corner. Nothing to change on adoption. ACTION only if you projected your own element into a collapsed ds-sidenav and relied on it sizing to its content -- it now stretches to the full column width -- or if you added a ::ng-deep width override to work around the squashed items, which is now double-applied (DELETE-LIST).",
|
|
401
|
+
"ds-timeline-item now pulls its node column up by half the difference between the heading's line box and the node size, so the marker's centre lands on the heading's rather than 4.2px below it. Each item is that much shorter at the top. Nothing to change on adoption. ACTION only if you compensated for the old offset with your own margin on ds-timeline-item or its node -- that override now double-applies (DELETE-LIST).",
|
|
402
|
+
"Two follow-ups to the shared field-state work, both fixed before release. (1) A field carrying its own error/warning/success state inside a ds-field-group repainted its background, tinting one segment mid-row; the three state rules now carry :not(.in-group), so the group keeps drawing the row's single treatment. (2) ds-field-group's own [invalid] read var(--border-error), a token that has never been declared, so it fell through to currentColor and drew a near-black border; it now resolves through a new public --ds-field-group-border-color-error (default --error). Nothing to change on adoption. ACTION only if you had a ::ng-deep override forcing the group's invalid border back to red, or suppressing a segment's fill inside a group -- both are now redundant (DELETE-LIST).",
|
|
403
|
+
"ds-card no longer cancels Enter and Space for everything inside it. The card bound (keydown.enter)/(keydown.space) with an unconditional preventDefault() on its <article>, and keydown bubbles, so EVERY card - not just elevation=\"interactive\" ones - killed both keys for its whole subtree: Enter in a password field submitted nothing, Space did not toggle a checkbox, Enter on a link did not follow it, and a space could not be typed into a text field. Both handlers now run only on an interactive card and only when the key landed on the card itself. Nothing to do; screens built inside a ds-card simply start working.",
|
|
404
|
+
"ds-table's onRowKey no longer consumes Enter and Space for every target inside an interactive row, so a row's own select checkbox toggles with Space and a cell's button activates with Enter. DELETE-LIST: a stopPropagation() you added on a control inside a clickable row to stop the row stealing its key - the library carried exactly that workaround on its own row-actions button and it is no longer needed.",
|
|
405
|
+
"ds-menu's type-ahead no longer swallows printable characters coming from a text-entry control, so a search or filter input projected into a menu can be typed into. Keys from ordinary menu items still drive the type-ahead. Nothing to do."
|
|
406
|
+
]
|
|
407
|
+
}
|