@verajs/renderer 0.2.0 → 0.2.2

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/README.md CHANGED
@@ -1,6 +1,6 @@
1
1
  # @verajs/renderer
2
2
 
3
- The DOM renderer for VeraJS — <!--size:renderer.gzip-->4.38 KB<!--/size:renderer.gzip--> gzipped,
3
+ The DOM renderer for VeraJS — <!--size:renderer.gzip-->4.58 KB<!--/size:renderer.gzip--> gzipped,
4
4
  no dependencies, no build step required.
5
5
 
6
6
  Tagged templates parse once and clone; every render after the first walks only the value slots, so
@@ -138,7 +138,8 @@ What a child position does with each kind of value. These match lit-html exactly
138
138
 
139
139
  | Value | Renders |
140
140
  | --- | --- |
141
- | string, number, `true`, `false`, `0` | as text — `${cond && 'x'}` puts the word `false` on the page |
141
+ | string, number, `0` | as text |
142
+ | `true`, `false` | as text — **and development says so**; see below |
142
143
  | `null`, `undefined` | nothing |
143
144
  | a template result | the template, updated in place while its shape holds |
144
145
  | an array or iterable | each entry in order; key them with `keyed()` |
@@ -148,6 +149,22 @@ What a child position does with each kind of value. These match lit-html exactly
148
149
  Strings render as **text**, always. There is no path by which an interpolated value becomes markup
149
150
  — see [Trusted HTML](#trusted-html-and-why-there-is-no-unsafehtml).
150
151
 
152
+ ### A boolean child renders the word, and JSX is the exception
153
+
154
+ `${items.length > 0 && html\`…\`}` is the ordinary conditional idiom, and when the test fails the
155
+ whole expression is `false` — which renders the text **"false"** on the page. The value is
156
+ legitimate and nothing throws, so **development names it** at the binding rather than leaving it to
157
+ be found by looking at the page. Write the form that renders nothing:
158
+
159
+ ```js
160
+ html`<p>${cond ? html`<em>yes</em>` : null}</p>` // or: ${(cond && html`…`) || null}
161
+ ```
162
+
163
+ **In JSX the same code needs no change** — `@verajs/jsx` compiles a boolean child away, which is
164
+ React's rule and where React expectations live. That is the **one** value semantic on which JSX and
165
+ a hand-written template differ, and this warning is what meets JSX-shaped code pasted into a
166
+ template. `0` still renders in both, exactly as in React: the rule is about booleans, not falsiness.
167
+
151
168
  A DOM node renders as itself, which is how a template holds something another library owns:
152
169
 
153
170
  ```js
@@ -283,6 +300,13 @@ Native `<slot>` needs a shadow root. This entry teaches the renderer to distribu
283
300
  component's own children into the `<slot name="…">` positions of its template, so **one component
284
301
  works in both modes** — users write `<div slot="title">` exactly as they would against shadow DOM.
285
302
 
303
+ **Wire it at your app entry, before anything renders.** A template resolves this once, at
304
+ construction, and is interned per call site for the life of the page — so wiring `slots` after a
305
+ component has already rendered does not reach that component, and it keeps showing its fallback
306
+ content while the children the host was given sit beside it as stray markup. Development names it
307
+ when it happens, but the rule is cheaper than the diagnostic: wire it in the same call as the
308
+ renderer, `wire([renderer, slots])`.
309
+
286
310
  Given this markup:
287
311
 
288
312
  ```html
@@ -649,6 +673,70 @@ Keys are strings carrying sigils, so TypeScript cannot check them against the el
649
673
  That is a genuine step down from written bindings, and the trade for names that are not known until
650
674
  runtime.
651
675
 
676
+ ### `props()` — a typed bag of property bindings
677
+
678
+ For the common case where every key is a **property**, `props()` removes both the sigils and the
679
+ typing gap:
680
+
681
+ ```js
682
+ import { props } from '@verajs/renderer/spread';
683
+
684
+ html`<calendar-day ${props({ date, events })}></calendar-day>`
685
+ ```
686
+
687
+ ```jsx
688
+ <calendar-day {...props({ date, events })}></calendar-day>
689
+ <calendar-day date={date} events={events} /> // JSX only: bare props ARE props on a component tag
690
+ ```
691
+
692
+ One function, both surfaces — JSX compiles `{...x}` to `spread(x)`, and `spread()` recognises an
693
+ already-branded result, so the two spellings are literally the same call. In JSX the bag is
694
+ optional altogether: on a dash-named tag a bare prop compiles to the `.name` binding directly
695
+ (`@verajs/jsx`'s README has the two attribute carve-outs), so `props()` is the template's
696
+ spelling and the bag for names not known until runtime. It exists because an
697
+ attribute is always a string: an array, a `Date` or a store can only reach a custom element as a
698
+ property — and while vera's JSX accepts the sigil spelling `.date={d}`, TSX's type-checker refuses
699
+ it (TS1003), so the bag is the typed path.
700
+
701
+ Three rules, each earned:
702
+
703
+ - **Keys are property names, never sigils** — `props({ date })` binds `.date`. Events and boolean
704
+ attributes keep their own spellings (`@click`/`onClick`, `?disabled`); this bag is properties
705
+ only, by definition.
706
+ - **Prefer keys spelled conditionally over bags that change shape**:
707
+ `props({ date: loaded ? date : null })` — key present from the first render — over
708
+ `props(loaded ? { date } : {})`. Both work (a key that appears later on a component is adopted
709
+ live, and one that disappears restores what the element held), but a stable shape means values
710
+ update in place, the same reason templates prefer `?hidden` over swapped subtrees.
711
+ - **A type argument makes the bag checked**: `props<CalendarDay>({ dat })` is a compile error
712
+ naming the misspelling — the checking that sigil-keyed spread genuinely cannot have.
713
+
714
+ ### How a component receives them
715
+
716
+ Nothing to declare on the other side — no `static properties`, no props argument. A component that
717
+ calls `init()` finds every bound property on `this`, **reactively**: reading `this.date` in a
718
+ render tracks it, the parent's next commit re-renders, and a store or `ref()` passed through stays
719
+ live because it arrives by identity. The full reception contract — including what happens when the
720
+ component's module loads *after* the parent rendered — is documented with `init()` in
721
+ `@verajs/core`'s README; this package's half is the delivery:
722
+
723
+ - **A property is not an attribute.** `date="…"` is an attribute — always a string, visible in
724
+ markup. `.date=${…}`/`props({ date })` is a property — any value, by identity, invisible to
725
+ `getAttribute`. Components receive **properties**; attributes are for CSS hooks and static
726
+ markup.
727
+ - **Delivery survives lazy definition.** A property bound before the element's module runs would
728
+ otherwise be destroyed by the class's field initializers at upgrade (at ES2022,
729
+ `item?: T` emits a real `item;` that runs during upgrade). The renderer records what nothing
730
+ received yet, and `init()` re-applies it — so a bound value outranks a class default, in both
731
+ field spellings. Elements that never call `init()` keep the development warning and the
732
+ `declare` advice instead.
733
+ - **Platform and foreign elements are untouched.** A property with an accessor anywhere —
734
+ `.title`, a Lit-style element, anything that already receives it — is delivered plainly and
735
+ never recorded.
736
+ - **SSR delivers, never serializes.** Under `@verajs/ssr`, a property bound on a rendered
737
+ component tag reaches that child's server render by identity — the markup never carries it, and
738
+ an unregistered tag passes through for the client to handle.
739
+
652
740
  ### Removing a key
653
741
 
654
742
  A key that disappears between renders **restores what the element held before the binding existed**.
@@ -688,7 +776,7 @@ a pair of totals — the totals move with every change to this package and the d
688
776
  which is the mistake this line already made once. `llms.txt` and this file disagreed about the figure
689
777
  for a while, at 16 B and 8 B respectively, and both were wrong. Nothing regenerates it, so it is
690
778
  dated; re-measure the same way if it matters.
691
- The entry itself is **<!--size:spread.gzip-->1.28 KB<!--/size:spread.gzip-->** gzipped, and only apps
779
+ The entry itself is **<!--size:spread.gzip-->1.60 KB<!--/size:spread.gzip-->** gzipped, and only apps
692
780
  that import it pay for that.
693
781
 
694
782
  Runtime is at parity with writing the bindings out: both do one comparison per binding per render,
@@ -730,7 +818,17 @@ each other in the same corner, and the second one's teardown would stop profilin
730
818
  which kept repainting a frozen report.
731
819
 
732
820
  Full API: `startProfiling()`, `stopProfiling()`, `getReport()`, `isProfiling()`, `profile(fn)`,
733
- `formatReport(report)`, `showProfiler(options?)`.
821
+ `formatReport(report)`, `showProfiler(options?)`. The first four are `profile()` unrolled, for a
822
+ session that does not fit one callback — a long-running tab, a REPL:
823
+
824
+ ```js
825
+ import { startProfiling, stopProfiling, getReport, isProfiling, formatReport } from '@verajs/renderer/profiler';
826
+
827
+ startProfiling();
828
+ // … interact with the app for as long as you like …
829
+ if (isProfiling()) console.log(formatReport(getReport())); // read mid-flight without stopping
830
+ stopProfiling(); // freezes the report
831
+ ```
734
832
 
735
833
  This costs production nothing, and there is nothing to strip: the instrumentation sits behind a
736
834
  `__DEV__` constant the build folds to `false`, so `vera-renderer.min.js` is byte-identical whether
@@ -830,7 +928,7 @@ life of three defects.
830
928
  writing tests for it.
831
929
  - HTML only. There is no `svg`/`mathml` equivalent yet.
832
930
 
833
- <!--size:tag.gzip-->1.94 KB<!--/size:tag.gzip--> gzipped, which includes `/spread` — the factory
931
+ <!--size:tag.gzip-->2.20 KB<!--/size:tag.gzip--> gzipped, which includes `/spread` — the factory
834
932
  needs it to apply props whose names it cannot know. Additive, like `/spread` and unlike the other
835
933
  entries: it inlines no renderer internals, so it is safe alongside any of them.
836
934
 
@@ -839,7 +937,14 @@ They are the table this entry uses to map React's names, and `@verajs/jsx` carri
839
937
  deliberately — the two are build-time and runtime, and a shared package would be a dependency where
840
938
  a test does the job. `tests/jsx-name-mapping.test.mjs` asserts the two agree on every key, and it can
841
939
  only do that against the built artifact, which is why they are exported at all. Read them if you are
842
- writing something that has to agree with both; do not build on them.
940
+ writing something that has to agree with both; do not build on them:
941
+
942
+ ```js
943
+ import { jsxName, BOOLEAN_ATTRIBUTES } from '@verajs/renderer/tag';
944
+
945
+ jsxName('className'); // 'class' — the runtime half of the React-name mapping
946
+ BOOLEAN_ATTRIBUTES.has('disabled'); // true — the names the tag entry toggles rather than assigns
947
+ ```
843
948
 
844
949
  ## Extending it — `_$apply$` and `_$child$`
845
950
 
@@ -980,7 +1085,7 @@ Directives other renderers ship, and what replaces them here.
980
1085
  | `ifDefined()` | built in — `null`/`undefined` remove an attribute |
981
1086
  | `classMap()` / `styleMap()` | build the string: `class="base ${extra}"` |
982
1087
  | `guard()` | reactivity already skips unchanged work |
983
- | `until()`, `asyncReplace()` | render a loading state and re-render from an effect — or nine lines against [`_$child$`](#extending-it--apply-and-child) |
1088
+ | `until()`, `asyncReplace()` | render a loading state and re-render from an effect — or nine lines against [`_$child$`](#extending-it--_apply-and-_child) |
984
1089
  | `unsafeHTML()` | `.innerHTML=${trusted}`, above |
985
1090
  | `literal()` / `static-html` | [`/tag`](#verajsrenderertag) — and a tag doubles as a JSX component |
986
1091
  | `live()` | [`!name`](#name--a-live-property) — a sigil, for the case that needs it |
@@ -1,3 +1,39 @@
1
+ /**
2
+ * The renderer's cross-file and public types — one `types.ts` at the import-graph root, per
3
+ * CODE-PRINCIPLES §1.
4
+ *
5
+ * **What is here and what deliberately is not.** Everything in this file is a pure data shape: it
6
+ * names no runtime value, so it can sit upstream of every module in the package and nothing here
7
+ * can pull a class into the import graph. The renderer's remaining cross-file types — `Item`,
8
+ * `ListStrategy`, `KeyedResult` — name the `Instance` and `ChildPart` CLASSES in their definitions,
9
+ * which puts them structurally downstream of `renderer.ts`; they stay beside those classes, and
10
+ * §1's "one file at the import-graph root" records that limit rather than pretending otherwise.
11
+ * Restating a class's shape here to satisfy the letter of the rule would create exactly the twin
12
+ * §1 forbids.
13
+ *
14
+ * Moving `ProfileReport` and `OverlayOptions` here dissolved a real cycle: `profiler.ts` imported
15
+ * `OverlayOptions` from `overlay.ts` while `overlay.ts` imported `ProfileReport` back from
16
+ * `profiler.ts`. Type-only, so it was erased and never a runtime cycle — but it is the shape this
17
+ * file's position in the graph exists to make impossible.
18
+ */
19
+ /**
20
+ * What a template tag (`html`, `svg`, `mathml`) produces — the compiled shape every renderer entry
21
+ * point accepts, and the one type a consumer normally names.
22
+ *
23
+ * `strings` is the template literal's own `strings` array and is the TEMPLATE'S IDENTITY: the
24
+ * engine interns it per call site, so two literals with identical text are two templates and
25
+ * rendering one after the other is a teardown, not an update. Anything building these by hand must
26
+ * hand back the same array to get an update.
27
+ */
28
+ type TemplateResult = {
29
+ /** 1 = html, 2 = svg, 3 = mathml — the markers core's built-in tags produce. */
30
+ _$litType$?: number;
31
+ strings: TemplateStringsArray;
32
+ values: unknown[];
33
+ /** Set by `keyed()`; drives keyed list reconciliation. */
34
+ key?: unknown;
35
+ };
36
+
1
37
  /**
2
38
  * @verajs/renderer — a ground-up, keyed, template-identity renderer.
3
39
  *
@@ -57,14 +93,7 @@
57
93
  * deliberately do not match the pattern: the first is an interop wire format, the second is called
58
94
  * by the DOM itself.
59
95
  */
60
- type TemplateResult = {
61
- /** 1 = html, 2 = svg, 3 = mathml — the markers core's built-in tags produce. */
62
- _$litType$?: number;
63
- strings: TemplateStringsArray;
64
- values: unknown[];
65
- /** Set by `keyed()`; drives keyed list reconciliation. */
66
- key?: unknown;
67
- };
96
+
68
97
  /**
69
98
  * Preserves the DOM of templates a child position toggles away from, instead of destroying it —
70
99
  * form values, scroll positions and media playback survive the round trip (lit-html calls this
@@ -1 +1 @@
1
- {"version":3,"file":"vera-renderer-hydrate.d.ts","sources":["../../src/renderer.ts","../../src/hydrate.ts"],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA0DG;AAEH,KAAA,cAAS;AAYT;;;;;;;;;;;;;;;AAeG;AACH;AAEI;;AAGJ;;;AAGG;AACH,cAAA,IAAA;AACA,QAAM,cAAgB;AACtB;;;;;;;;;;;;;;;;;;ACjFG;AACH;AAwBA,cAAa;AAGb;AACA;AAEA;;;;AAIG;AACH;;;;","names":[]}
1
+ {"version":3,"file":"vera-renderer-hydrate.d.ts","sources":["../../src/types.ts","../../src/renderer.ts","../../src/hydrate.ts"],"mappings":"AAAA;;;;;;;;;;;;;;;;;AAiBG;;;;;;;;;;;;;;;;;;;ACjBH;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA0DG;;;;;;;;;;;;;;;;AAwBA;AACH;AAEI,kBAAgC;sBACvB;AAEb;;;;;;;;;;;;;;;;;;ACtEG;AACH;AAyBA,cAAa;AAGb;AACA;AAEA;;;;AAIG;AACH;;;;","names":[]}
@@ -763,6 +763,7 @@ const getTemplate = (result) => {
763
763
  templateCache.set(result.strings, (template = new Template(result)));
764
764
  return template;
765
765
  };
766
+ /** Anything bound to a live position: commits values[index..], returns the next value index. */
766
767
  const IGNORED_PART = { _commit: (_values, index) => index + 1 };
767
768
  /** Never equal to any user value, so the first commit always runs. */
768
769
  const UNSET = {};
@@ -800,6 +801,143 @@ const LIVE = 3;
800
801
  const EVENT = 4; // @name
801
802
  /** A binding that must never write — see the `__proto__` refusal in the constructor. */
802
803
  const REFUSED = -1;
804
+ /** `.name` on a custom element nothing receives yet — records into `_$props$`, then becomes
805
+ * `PROPERTY`. A distinct kind keeps the steady-state property path free of any adoption check. */
806
+ const PROP_ADOPT = 6;
807
+ /**
808
+ * The `PROP_ADOPT` commit, kept OUT of `_commit` so that function's size — and the JIT layout of
809
+ * its hot `PROPERTY`/`ATTR` branches — is unchanged from before this feature. `_commit` carries one
810
+ * dispatch line; all of this lives here, off every part's steady state.
811
+ *
812
+ * One question decides everything: **after the write, does anything receive this property?** An
813
+ * accessor anywhere on the chain — the platform's own (`title`, `hidden`), a component's declared
814
+ * one, or the pair core's `init()` defines — means the write just landed in real hands: no record,
815
+ * and the part rejoins `PROPERTY` for good. The walk is one step when the write created an own data
816
+ * property (nobody consumed it) and stops at the first descriptor either way, and it is what keeps
817
+ * a platform property on a lazy tag out of the record — recording `.title` would end with the drain
818
+ * shadowing `HTMLElement.prototype.title` and breaking it.
819
+ *
820
+ * An unreceived write is recorded under the sigiled expando **`_$props$`**, because two arrival
821
+ * orders both need it: on a LAZY tag the value sits as a plain own property that the class's field
822
+ * initializers will overwrite at upgrade (the clobber the `PROPERTY` branch only *detects*), and on
823
+ * an EAGER one the value survives but is indistinguishable from the component's own fields by the
824
+ * time `init()` runs. The record answers both — core's `init()` drains it, re-applying values
825
+ * unconditionally, which repairs BOTH spellings of the clobber (`item;` and `item = default`): it
826
+ * never asks whether a clobber happened, it asserts a bound value outranks a class default, which
827
+ * is what props mean. Element-carried because `./spread.ts` writes the same properties from a
828
+ * separate bundle with no shared registry — the `_$`-sigiled member is the cross-bundle channel,
829
+ * and the recorder there is this one's deliberate twin.
830
+ *
831
+ * Returns the part's NEXT kind: `PROPERTY` once something receives the property or the element is
832
+ * upgraded (its record is final — `init()` drains it in the same tick) or has no realm; `REFUSED`
833
+ * for a getter with no setter, so the binding goes inert instead of the flipped plain write
834
+ * throwing on the next commit; `0` to stay adopting — a dash-named element that never upgrades and
835
+ * never receives keeps recording, so its record stays current rather than staling. The one
836
+ * retention this leaves: a defined non-vera element with a plain data property keeps its FIRST
837
+ * committed values in `_$props$` for its lifetime — bounded, one generation, and the part itself
838
+ * already retains the current generation in `_committed`.
839
+ */
840
+ const commitAdopt = (element, name, value) => {
841
+ const el = element;
842
+ /**
843
+ * The walk comes BEFORE the write, because what the walk finds decides whether writing is even
844
+ * legal: a SETTER receives the value (and a setter that throws is the component's own error), a
845
+ * GETTER with no setter cannot — the old order assigned first, which in strict mode threw a raw
846
+ * TypeError out of the render for the eager spelling of exactly the case the drain refuses by
847
+ * name for the lazy one. Same contested state, one rule, both arrival orders — and the server's
848
+ * `deliverProperty` applies it too. The dev warning defers to `_$adopt$` where it exists: an
849
+ * initialized component's drain already speaks, and two voices for one binding is noise.
850
+ */
851
+ let carrier = el;
852
+ while (carrier !== null) {
853
+ const desc = Object.getOwnPropertyDescriptor(carrier, name);
854
+ if (desc !== undefined) {
855
+ if (desc.set !== undefined) {
856
+ el[name] = value;
857
+ return PROPERTY;
858
+ }
859
+ if (desc.get !== undefined) {
860
+ /** An initialized component's receiver owns the refusal (and its channel); for anything
861
+ * else this is the only door, so it says so itself. Either way the binding goes INERT —
862
+ * flipping to `PROPERTY` would make the next commit's plain write throw. */
863
+ if (el._$adopt$ !== undefined)
864
+ el._$adopt$(name, value);
865
+ else if (true )
866
+ console.warn(`[vera] renderer: <${element.localName}> declares \`${name}\` as a getter with no ` +
867
+ `setter — the value bound by \`.${name}=\${…}\` cannot be delivered and the binding ` +
868
+ `is ignored. Add a setter, or stop binding it.`);
869
+ return REFUSED;
870
+ }
871
+ break; // a data property — an own field, or an inherited default: nothing receives it
872
+ }
873
+ carrier = Object.getPrototypeOf(carrier);
874
+ }
875
+ el[name] = value;
876
+ /**
877
+ * An initialized component receives LIVE: `init()` left `_$adopt$` on the element, and a key it
878
+ * has not adopted yet goes through that door — a record here would never be read again, because
879
+ * the drain already ran. This is how a hydrated child gets the props its parent commits after
880
+ * the child settled, and how a spread bag's conditional key arrives reactive on a live element.
881
+ */
882
+ if (el._$adopt$ !== undefined) {
883
+ el._$adopt$(name, value);
884
+ return PROPERTY;
885
+ }
886
+ const record = (el._$props$ ??= {});
887
+ const firstRecording = !Object.hasOwn(record, name);
888
+ record[name] = value;
889
+ /**
890
+ * Upgrade is read off the PROTOTYPE, never off `el.constructor`: a spread bag's keys are runtime
891
+ * data, so a bag carrying a key named `constructor` writes an own property that shadows the real
892
+ * one, and a check that reads it would then misjudge every later commit. No property write can
893
+ * move an element's prototype — the one assignment that could, `__proto__`, both twins refuse.
894
+ * The realm comes from the element, as always.
895
+ */
896
+ const win = element.ownerDocument.defaultView;
897
+ const upgraded = win === null || Object.getPrototypeOf(el) !== win.HTMLElement.prototype;
898
+ /**
899
+ * **The clobber detector, for the element the drain will never reach.** A component that calls
900
+ * `init()` has the record re-applied, so for it the pre-upgrade window is repaired; an element
901
+ * that never drains — a plain custom element with a class field — still loses the bound value to
902
+ * the field initializer at upgrade, exactly as before this feature, and still deserves the
903
+ * warning. Told apart by OWNERSHIP, not by value: after the definition arrives (and, for a
904
+ * connected element, `connectedCallback` and the drain have run synchronously inside `define()`),
905
+ * an own accessor means the drain took the property and there is nothing to report. Identity
906
+ * against the COMMITTED value would lie twice — a drained value reads back through the store's
907
+ * proxy, and a value superseded by a later pre-upgrade commit differs without anything being
908
+ * wrong — so the comparison is against the RECORD, which later recordings keep current, and the
909
+ * subscription is installed once per (element, property), on the first recording. The registry
910
+ * is the element's own realm's. Development only; production carries no check, no message, no
911
+ * `whenDefined` subscription.
912
+ */
913
+ if (true && win !== null && !upgraded && firstRecording) {
914
+ const tag = element.localName;
915
+ win.customElements.whenDefined(tag).then(() => {
916
+ /** Ownership is asked of the whole CHAIN: the drain's own accessor, a class's prototype
917
+ * pair the value was handed to, or a get-only surface the drain refused by name — every
918
+ * one means the property has an owner and this closure has nothing left to report. */
919
+ let carrier = el;
920
+ let owned = false;
921
+ while (carrier !== null) {
922
+ const desc = Object.getOwnPropertyDescriptor(carrier, name);
923
+ if (desc !== undefined) {
924
+ owned = desc.get !== undefined || desc.set !== undefined;
925
+ break;
926
+ }
927
+ carrier = Object.getPrototypeOf(carrier);
928
+ }
929
+ if (!owned && el[name] !== record[name])
930
+ console.warn(`[vera] renderer: the value bound by \`.${name}=\${…}\` on <${tag}> was replaced ` +
931
+ `while the element upgraded. A class field is the usual cause: at ES2022 ` +
932
+ `\`${name}?: …\` emits \`${name};\`, which runs during upgrade and overwrites ` +
933
+ `whatever was set beforehand — write it \`declare ${name}?: …\` instead, which ` +
934
+ `emits nothing. A component that calls init() adopts bound properties automatically ` +
935
+ `and never sees this; this element did not. Ignore this if the component replaced ` +
936
+ `the value on purpose.`);
937
+ });
938
+ }
939
+ return upgraded ? PROPERTY : 0;
940
+ };
803
941
  /**
804
942
  * Calls an element ref, and survives one that throws.
805
943
  *
@@ -890,6 +1028,19 @@ class AttrPart {
890
1028
  kind = EVENT;
891
1029
  realName = name.slice(2).toLowerCase();
892
1030
  }
1031
+ /**
1032
+ * **A property write to a custom element starts as `PROP_ADOPT`**, a distinct kind so the
1033
+ * steady-state `PROPERTY` commit path stays byte-for-byte unchanged — zero added instruction
1034
+ * for every `.value=` on a built-in. Every dash-named element starts here, defined or not,
1035
+ * because the record has to exist for BOTH arrival orders: an eager component's first commit
1036
+ * lands after upgrade but before `init()` (one commit records, then the part flips), and a
1037
+ * lazy one's land before upgrade (the part records until the definition arrives). `commitAdopt`
1038
+ * decides which per commit and flips the part back to `PROPERTY` the moment something real
1039
+ * receives the property, so the window is one commit for a defined element and the pre-upgrade
1040
+ * stretch for a lazy one.
1041
+ */
1042
+ if (kind === PROPERTY && element.localName.includes('-'))
1043
+ kind = PROP_ADOPT;
893
1044
  this._kind = kind;
894
1045
  this._name = realName;
895
1046
  this._element = element;
@@ -1002,13 +1153,29 @@ class AttrPart {
1002
1153
  if (kind === LIVE) {
1003
1154
  this._committed = value;
1004
1155
  /**
1005
- * **Except while adopting.** Hydration reaches a DOM a person may already have used, and the
1006
- * click that checked a radio happened before any handler existed to tell the store about it —
1007
- * so re-asserting the server's choice here would throw the interaction away and nothing would
1008
- * ever put it back. Recorded, not written, exactly as the other form properties are; the
1009
- * first state-driven render after that applies live semantics normally.
1156
+ * **Except while adopting — on a form control.** Hydration reaches a DOM a person may
1157
+ * already have used, and the click that checked a radio happened before any handler existed
1158
+ * to tell the store about it — so re-asserting the server's choice here would throw the
1159
+ * interaction away and nothing would ever put it back. Recorded, not written, exactly as
1160
+ * the other form properties are; the first state-driven render after that applies live
1161
+ * semantics normally.
1162
+ *
1163
+ * **A COMPONENT tag is the exception to the exception**: there is no user-editable carrier
1164
+ * behind `!prop` on a component — it is a property delivery spelled with `!` — and the
1165
+ * server DELIVERED it to the child's render, so yielding here dropped the one copy the
1166
+ * client would ever get and hydration regressed the child's content to its prop-less state.
1167
+ * Found by the hydration fixture on its first run. The write is unconditional and routes
1168
+ * through the element like any component prop: an accessor receives it, an initialized
1169
+ * component adopts it live.
1010
1170
  */
1011
- if (!(true && adopting)) {
1171
+ if (true && adopting) {
1172
+ /** Routed through `commitAdopt`, not written raw: the child hydrated BEFORE this commit,
1173
+ * so only the adoption door re-runs its render. The part stays `LIVE` — the return is
1174
+ * deliberately dropped — so live semantics resume on the next state-driven render. */
1175
+ if (this._element.localName.includes('-'))
1176
+ commitAdopt(this._element, this._name, value);
1177
+ }
1178
+ else {
1012
1179
  const liveTarget = this._element;
1013
1180
  if (liveTarget[this._name] !== value)
1014
1181
  liveTarget[this._name] = value;
@@ -1088,52 +1255,16 @@ class AttrPart {
1088
1255
  this._committed = value;
1089
1256
  return index + this._slots;
1090
1257
  }
1091
- target[name] = value;
1092
1258
  /**
1093
- * Detection only, and deliberately not repair.
1094
- *
1095
- * A property set on a custom element that has not upgraded yet lands as an own property on
1096
- * the instance. When the definition arrives — lazily imported, code-split, or a module that
1097
- * simply had not run — `customElements.define` upgrades synchronously and the class's field
1098
- * initializers execute. At target ES2022 a field declaration is a `[[Define]]`, so
1099
- * `item?: T` emits `item;` and overwrites the bound value with `undefined`. Nothing throws;
1100
- * it reads as broken reactivity.
1101
- *
1102
- * Repairing it was tried and removed. Re-applying when the slot came back `undefined`
1103
- * covered `item?: T` but not `item = someDefault`, which overwrites with the default and
1104
- * never looks clobbered — so the repair was silently partial, and made the bug intermittent
1105
- * across two spellings of the same mistake. It also cost 74 B in every app while leaving
1106
- * `declare` mandatory anyway, because a property assigned imperatively is unrecoverable:
1107
- * the renderer never saw it, and by the time `init()` runs the value is already gone.
1108
- *
1109
- * So this is `true `-only and production pays nothing — no check, no message, no
1110
- * `whenDefined` subscription. Lit reached the same place from the other direction and
1111
- * throws (`lit.dev/msg/class-field-shadowing`); a warning suffices here because Vera has no
1112
- * prototype accessors to shadow permanently — the damage is one lost value, not a property
1113
- * that never updates again.
1259
+ * No un-upgraded custom element can be behind this write: a dash-named property part is
1260
+ * born `PROP_ADOPT` and only flips here once something receives the property, so the
1261
+ * pre-upgrade window — the clobber, its repair for components, and the `whenDefined`
1262
+ * detector for elements that never drain — lives entirely in `commitAdopt` above. History
1263
+ * worth keeping: an earlier repair was tried HERE and removed as silently partial (it
1264
+ * re-applied on `undefined`, so `item = someDefault` never looked clobbered); the record
1265
+ * the drain re-applies unconditionally is what answered that objection.
1114
1266
  */
1115
- if (true ) {
1116
- const tag = this._element.localName;
1117
- if (tag.indexOf('-') > 0 && !customElements.get(tag)) {
1118
- customElements.whenDefined(tag).then(() => {
1119
- /**
1120
- * States what was observed rather than diagnosing it. A class field is the usual
1121
- * cause by a wide margin, but a component that assigns the property itself during
1122
- * upgrade produces the same observation, and that is a legitimate — if confusing —
1123
- * thing to do, since the renderer's dirty check will not re-apply the bound value on
1124
- * the next render either.
1125
- */
1126
- if (target[name] !== value)
1127
- console.warn(`[vera] renderer: the value bound by \`.${name}=\${…}\` on <${tag}> was replaced ` +
1128
- `while the element upgraded. A class field is the usual cause: at ES2022 ` +
1129
- `\`${name}?: …\` emits \`${name};\`, which runs during upgrade and overwrites ` +
1130
- `whatever was set beforehand — write it \`declare ${name}?: …\` instead, which ` +
1131
- `emits nothing. A plain field also wipes properties assigned imperatively, ` +
1132
- `which cannot be detected at all. Ignore this if the component replaced the ` +
1133
- `value on purpose.`);
1134
- });
1135
- }
1136
- }
1267
+ target[name] = value;
1137
1268
  }
1138
1269
  else if (kind === BOOLEAN) {
1139
1270
  /** Unconditional for the same reason: `<b hidden ?hidden=${false}>` must end up not hidden. */
@@ -1160,6 +1291,11 @@ class AttrPart {
1160
1291
  this._element.addEventListener(this._name, this);
1161
1292
  this._handler = value ?? null;
1162
1293
  }
1294
+ else if (kind === PROP_ADOPT) {
1295
+ const next = commitAdopt(this._element, this._name, value);
1296
+ if (next !== 0)
1297
+ this._kind = next;
1298
+ }
1163
1299
  else if (value != null) {
1164
1300
  /**
1165
1301
  * Element ref: `<input ${myRef} />`. A function is called with the element; an object gets
@@ -1422,6 +1558,38 @@ const isTemplateResult = (value) => value.strings !== undefined;
1422
1558
  * with a text node — so the steady state is compare-and-assign on `.data`. This part carries only
1423
1559
  * that, upgrading itself to a full ChildPart the first time it sees null, a template, or an array.
1424
1560
  */
1561
+ /**
1562
+ * **A boolean in a child position renders the WORD, and almost nobody means that.**
1563
+ *
1564
+ * `${items.length > 0 && html`…`}` is the ordinary conditional idiom, and when the test is false
1565
+ * the whole expression is `false` — which becomes the text `false` on the page. Nothing throws;
1566
+ * the value is legitimate; only the intent is wrong. lit-html does the same and this renderer
1567
+ * matches it deliberately (anything not nullish renders), so the BEHAVIOUR stays and the mistake
1568
+ * is named where it happens instead of being found by looking at the page.
1569
+ *
1570
+ * `@verajs/jsx` compiles this case away — in JSX a boolean child becomes nothing, React's rule,
1571
+ * because that is where React expectations live. **That is the one value semantic on which JSX and
1572
+ * a hand-written template differ**, which makes this the warning that meets JSX-shaped code pasted
1573
+ * into a template.
1574
+ *
1575
+ * Called from BOTH child sinks, because a text position is a `TextPart` until something upgrades
1576
+ * it and a boolean never triggers that upgrade — putting the check in `ChildPart` alone left it
1577
+ * silent for exactly the common case. One function, two call sites, per the standing rule that a
1578
+ * deliberate duplication is a fix's second address.
1579
+ *
1580
+ * Both sinks gate this on the value CHANGING — not merely on reaching a commit — so an unchanged
1581
+ * `false` is reported once and not once per render. `ChildPart` needed that stated explicitly: its
1582
+ * dirty check sits below, and calling this above it spammed every pass.
1583
+ */
1584
+ const warnBooleanChild = (value) => {
1585
+ if (typeof value !== 'boolean')
1586
+ return;
1587
+ console.warn(`[vera] renderer: a child position was given \`${value}\`, which renders as the word ` +
1588
+ `"${value}" — the usual cause is \`\${cond && …}\` with a false \`cond\`.\n` +
1589
+ `Write \`\${cond ? … : null}\`, or \`\${(cond && …) || null}\`; \`null\` and \`undefined\` are ` +
1590
+ `the values that render nothing. If you meant to display the boolean, say so with ` +
1591
+ `\`\${String(value)}\` and this goes quiet.`);
1592
+ };
1425
1593
  class TextPart {
1426
1594
  _text;
1427
1595
  _value = '';
@@ -1477,6 +1645,8 @@ class TextPart {
1477
1645
  return index + 1;
1478
1646
  }
1479
1647
  if (value !== this._value) {
1648
+ if (true )
1649
+ warnBooleanChild(value);
1480
1650
  this._value = value;
1481
1651
  this._text.data = value;
1482
1652
  }
@@ -1761,6 +1931,15 @@ class ChildPart {
1761
1931
  return;
1762
1932
  }
1763
1933
  if (typeof value !== 'object') {
1934
+ /**
1935
+ * Gated on the value CHANGING, which is the same condition the commit below uses. Called
1936
+ * before that check it fired on every render of an unchanged `false` — a channel that
1937
+ * repeats is as useless as one that stays silent, and the comment on `warnBooleanChild`
1938
+ * claimed this was already true. `_value` holds whatever was last committed here, so a part
1939
+ * arriving from a template state differs and reports once.
1940
+ */
1941
+ if (true && this._value !== value)
1942
+ warnBooleanChild(value);
1764
1943
  if (this._mode === TEXT) {
1765
1944
  if (this._value !== value) {
1766
1945
  this._value = value;