Preserve identified form visual parts and responsive CSS facts

php-transformer/docs/form-presentation-graph.md

@@ -0,0 +1,25 @@
+# Form Presentation Graph
+
+`generic/computed-form-presentation/v2` is an optional `generic/forms/v1` payload member. It carries bounded source-CSS facts for controls, associated labels, and identified inline SVG descendants. It is not a provider recipe and never assigns visual semantics such as "globe" or "chevron".
+
+```php
+array{
+  schema: 'generic/computed-form-presentation/v2',
+  basis: 'source_css_cascade',
+  controls: list<array{index: int, control?: Role, label?: Role, required_marker?: Role}>,
+  visual_parts: list<array{
+    id: 'control-{control order}-svg-{descendant order}', index: int,
+    kind: 'inline_svg', source_selector: string, markup: safe-svg,
+    intrinsic_size?: array{width: positive-int, height: positive-int},
+    source_css: array{state: 'known', styles: array<string,string>, provenance: list<Provenance>}|array{state: 'unknown'}
+  }>,
+  variants: list<array{index: int, role: 'control'|'label'|'required_marker'|'visual_part', part_id?: string, condition: Condition, style_patch: array<string,string>, precedence: array<string,array>, provenance: list<Provenance>}>,
+  truncated: bool, limits: array{controls: 128, rules_per_role: 32}, diagnostics: list<string>
+}
+```
+
+`visual_parts` contains only drawable SVG descendants that pass `SourceDom::isSafeInlineSvgMarkup()`. `source_selector` identifies the source descendant and `index` links it to the source control order. `source_css.state: 'unknown'` means no unconditional source-CSS facts were matched. Responsive facts can still be present in `variants`; consumers evaluate those conditions rather than treating an unknown base as missing presentation. Consumers must not infer a role or layout from SVG shape or order.
+
+`required_marker` is emitted only for the existing explicit required marker identity: an `aria-hidden="true"` span containing one to four asterisks on a required control's associated label. It records the marker's own matched CSS and variants; when it inherits the label unchanged, it safely has empty `styles` and `provenance` rather than invented source facts. Existing v1 graphs remain valid. Graphs without visual parts or an explicit required marker retain the v1 envelope; graphs with either emit v2. This contract describes source appearance, not interaction-state semantics or provider destinations.
+
+The unshipped v2 envelope also includes `visual_groups`. Each group records the source selector of the SVG parts' lowest common ancestor, a stable `id`, ordered `part_ids`, and `source_css` using the same base-fact shape as visual parts. Group responsive variants use `role: visual_group` and `group_id` rather than a control index. Consumers preserve those authored container facts independently of each SVG's intrinsic dimensions and keep provider visibility state outside that layout container.

php-transformer/src/ArtifactCompiler/ArtifactCompiler.php

@@ -2666,6 +2666,11 @@ private function stylesheetAssetsForSource(string $html, string $sourcePath, arr
                 if ( is_array($file) && ! isset($seenPaths[$file['path']]) ) {
                     $assets[] = array( 'path' => $file['path'], 'source_path' => $file['source_path'] ?? $file['path'], 'content' => $file['content'], 'source_hash' => (string) ($file['provenance']['hash'] ?? hash('sha256', $file['content']) ), 'media' => (string) ($file['media'] ?? ''), 'type' => (string) ($file['type'] ?? '') );
                     $seenPaths[$file['path']] = true;
+                } elseif ( '' !== ($content = trim(html_entity_decode($tagRecord['content'], ENT_QUOTES | ENT_HTML5, 'UTF-8'))) ) {
+                    // Generated inline-style files can be omitted at the artifact
+                    // file limit. Their source HTML was accepted independently,
+                    // so retain the authored stylesheet for source analysis.
+                    $assets[] = array( 'path' => 'inline-style-' . $inlineIndex . '.css', 'source_path' => 'inline-style', 'content' => $content, 'source_hash' => hash('sha256', $content), 'media' => $this->htmlAttribute($attributes, 'media'), 'type' => $this->htmlAttribute($attributes, 'type') );
                 }
                 continue;
             }

php-transformer/src/Css/CssAnalysisLimits.php

@@ -0,0 +1,11 @@
+<?php
+declare(strict_types=1);
+
+namespace Automattic\BlocksEngine\PhpTransformer\Css;
+
+/** Shared bounds for source-preserving CSS analysis consumers. */
+final class CssAnalysisLimits
+{
+    // Allows a complete author stylesheet while retaining a bounded parser input.
+    public const MAX_STYLESHEET_BYTES = 4194304;
+}

php-transformer/src/Css/CssRuleAnalyzer.php

@@ -80,7 +80,7 @@ private function analyzeStylesheet(string $css, string $path, string $hash, ?arr
         for ( $offset = 0, $length = strlen($css); $offset < $length; ) {
             $boundary = $this->nextRuleBoundary($css, $offset);
             if ( null === $boundary ) {
-                if ( '' !== trim(substr($css, $offset)) ) {
+                if ( $this->hasNonTrivia($css, $offset) ) {
                     $result['diagnostics'][] = 'malformed_stylesheet:' . $path;
                 }
                 return;
@@ -249,6 +249,29 @@ private function nextRuleBoundary(string $css, int $offset): ?int
         return null;
     }
 
+    /**
+     * A stylesheet may legally end with whitespace and comments, including a
+     * source-map comment. Preserve malformed-input diagnostics for every other
+     * incomplete trailing token.
+     */
+    private function hasNonTrivia(string $css, int $offset): bool
+    {
+        $state = CssSyntaxScanner::state();
+        for ($length = strlen($css); $offset < $length; ) {
+            $insideComment = $state['comment'];
+            $startsComment = ! $insideComment && '' === $state['quote'] && '/*' === substr($css, $offset, 2);
+            $next = CssSyntaxScanner::consume($css, $offset, $state);
+            if ( null === $next ) {
+                return true;
+            }
+            if (! $insideComment && ! $startsComment && ! CssSyntaxScanner::isCssWhitespace($css[$offset])) {
+                return true;
+            }
+            $offset = $next;
+        }
+        return ! CssSyntaxScanner::isComplete($state);
+    }
+
     /**
      * CSS comments disappear from selectors, but adjacent identifier-like tokens
      * need a separator to avoid changing a descendant selector into one token.

php-transformer/src/HtmlToBlocks/Classification/SourceElementClassifier.php

@@ -278,7 +278,7 @@ public function isPositiveCssLength(string $value): bool
     public function isVisibleEmptyVisualPaint(string $value): bool
     {
         $value = strtolower(trim($value));
-        if ( '' === $value || 'none' === $value || 'transparent' === $value || preg_match('/^rgba?\([^)]*,\s*0(?:\.0+)?\s*\)$/', $value) ) {
+        if ( '' === $value || 'none' === $value || 'transparent' === $value || preg_match('/^0(?:px|%)?(?:\s+0(?:px|%)?)?$/', $value) || preg_match('/^rgba?\([^)]*,\s*0(?:\.0+)?\s*\)$/', $value) ) {
             return false;
         }
 

php-transformer/src/HtmlToBlocks/Elements/FormControlMetadataBuilder.php

@@ -109,14 +109,8 @@ public function control(DOMElement $control): array
 
         if ( $control->hasAttribute('required') || 'true' === strtolower(trim(SourceDom::attr($control, 'aria-required'))) ) {
             $metadata['required'] = true;
-            if ( $labelElement instanceof DOMElement ) {
-                foreach ( $labelElement->getElementsByTagName('span') as $marker ) {
-                    $text = trim($marker->textContent ?? '');
-                    if ( 'true' === strtolower(SourceDom::attr($marker, 'aria-hidden')) && preg_match('/^\*{1,4}$/D', $text) ) {
-                        $metadata['required_text'] = $text;
-                        break;
-                    }
-                }
+            if ( ($marker = $this->requiredMarker($control)) instanceof DOMElement ) {
+                $metadata['required_text'] = trim($marker->textContent ?? '');
             }
         }
         foreach ( array( 'disabled', 'readonly', 'checked', 'multiple' ) as $attribute ) {
@@ -192,6 +186,25 @@ public function associatedLabel(DOMElement $control): ?DOMElement
         return null;
     }
 
+    /** The explicit decorative required marker also supplies provider presentation identity. */
+    public function requiredMarker(DOMElement $control): ?DOMElement
+    {
+        if ( ! $control->hasAttribute('required') && 'true' !== strtolower(trim(SourceDom::attr($control, 'aria-required'))) ) {
+            return null;
+        }
+        $label = $this->labelElement($control);
+        if ( ! $label instanceof DOMElement ) {
+            return null;
+        }
+        foreach ( $label->getElementsByTagName('span') as $marker ) {
+            $text = trim($marker->textContent ?? '');
+            if ( 'true' === strtolower(SourceDom::attr($marker, 'aria-hidden')) && preg_match('/^\*{1,4}$/D', $text) ) {
+                return $marker;
+            }
+        }
+        return null;
+    }
+
     private function labelElement(DOMElement $control): ?DOMElement
     {
         $label = $this->associatedLabel($control);

php-transformer/src/HtmlToBlocks/Elements/FormFallbackFindingBuilder.php

@@ -31,7 +31,9 @@ public function build(DOMElement $element, ?array $readableFormBlock, ?array $bi
         $controlTopology = (new FormControlTopologyBuilder())->build($element);
         $layoutGraph = (new FormLayoutGraphBuilder())->build($element, $this->context->stylesheetAssets(), $this->context->formLayoutCss());
         $presentationGraph = (new FormPresentationGraphBuilder(
-            fn (DOMElement $control, string $value): string => $this->context->resolvePresentationValue($control, $value)
+            fn (DOMElement $control, string $value): string => $this->context->resolvePresentationValue($control, $value),
+            fn (DOMElement $element): string => $this->context->sanitizeInlineSvgMarkup($element),
+            fn (DOMElement $control): ?DOMElement => $this->metadataBuilder->requiredMarker($control)
         ))->build($element, $this->context->stylesheetAssets(), $this->context->formLayoutCss());
         $boundedHtml = $this->context->boundedFallbackHtml($element);
         $replacesRuntimeIsland = null !== $bindingBlock;

php-transformer/src/HtmlToBlocks/Elements/FormFallbackFindingContext.php

@@ -18,6 +18,7 @@ final class FormFallbackFindingContext
      * @param Closure(DOMElement): array<string, mixed>                                             $classifyFallbackSubtree
      * @param Closure(array<string, mixed>, string, array<int, string>): array<string, mixed>       $blockBinding
      * @param (Closure(DOMElement, string): string)|null                                            $resolvePresentationValue
+     * @param (Closure(DOMElement): string)|null                                                    $sanitizeInlineSvgMarkup
      */
     public function __construct(
         private readonly HtmlTransformerSession $session,
@@ -26,7 +27,8 @@ public function __construct(
         private readonly Closure $sourceContext,
         private readonly Closure $classifyFallbackSubtree,
         private readonly Closure $blockBinding,
-        private readonly ?Closure $resolvePresentationValue = null
+        private readonly ?Closure $resolvePresentationValue = null,
+        private readonly ?Closure $sanitizeInlineSvgMarkup = null
     ) {
     }
 
@@ -80,6 +82,11 @@ public function resolvePresentationValue(DOMElement $element, string $value): st
         return null !== $this->resolvePresentationValue ? ($this->resolvePresentationValue)($element, $value) : $value;
     }
 
+    public function sanitizeInlineSvgMarkup(DOMElement $element): string
+    {
+        return null !== $this->sanitizeInlineSvgMarkup ? ($this->sanitizeInlineSvgMarkup)($element) : '';
+    }
+
     /** @param array<string, mixed> $finding @return array<string, mixed> */
     public function buildFallbackDiagnostic(array $finding): array
     {

php-transformer/src/HtmlToBlocks/HtmlCompilation.php

@@ -617,7 +617,8 @@ function (DOMElement $element, array &$fallbacks, bool $captureUnsupported): arr
                 fn (DOMElement $element): array => $this->sourceContext($element),
                 fn (DOMElement $element): array => $this->fallbackEmitter()->classifyFallbackSubtree($element),
                 fn (array $block, string $role, array $supersededRuntimeSelectors): array => $this->blockBinding($block, $role, $supersededRuntimeSelectors),
-                fn (DOMElement $element, string $value): string => $this->styleResolver->resolveCssVariablesInValue($value, $element)
+                fn (DOMElement $element, string $value): string => $this->styleResolver->resolveCssVariablesInValue($value, $element),
+                fn (DOMElement $element): string => $this->svgMaterializer->restoreSvgCasing($this->sanitizeInlineSvgMarkup($element))
             ),
             $this->formControlMetadataBuilder,
             $this->formSuccessPanelMetadataBuilder,
@@ -1928,7 +1929,9 @@ private function materializeAuthorStylesheet(string $html, string $staticCss, bo
             // core Group and its children. Keep authored grid/flex children as
             // direct layout items, matching the saved frontend markup.
             $beforeAuthorCssParts[] = ':root :where(.' . self::CSS_OWNED_LAYOUT_CLASS . ')>.block-editor-inner-blocks,'
-                . ':root :where(.' . self::CSS_OWNED_LAYOUT_CLASS . ')>.block-editor-inner-blocks>.block-editor-block-list__layout{display:contents}';
+                . ':root :where(.' . self::CSS_OWNED_LAYOUT_CLASS . ')>.block-editor-inner-blocks>.block-editor-block-list__layout{display:contents}'
+                // Empty Group placeholders have an additional unadorned editor wrapper.
+                . ':root .editor-styles-wrapper :where(.' . self::CSS_OWNED_LAYOUT_CLASS . ')>div:not([class]):not([id]):not([style]):has(>[data-block].wp-block-group__placeholder){display:contents}';
         }
         $layoutShellBlockName = $this->generatedBlocks()->blockName('layout-shell');
         if ( str_contains($serializedBlocks, '<!-- wp:' . $layoutShellBlockName) ) {

php-transformer/src/HtmlToBlocks/Style/FormControlTopologyBuilder.php

@@ -52,6 +52,38 @@ public function build(DOMElement $form): array
         );
     }
 
+    /**
+     * Return each control's nearest-to-farthest wrapper ancestors that contain no
+     * other form controls. Labels and other non-controls intentionally do not
+     * make a wrapper shared.
+     *
+     * @return array<int, list<DOMElement>>
+     */
+    public function exclusiveWrapperAncestors(DOMElement $form): array
+    {
+        $result = array();
+        $controls = $this->controls($form);
+        $owners = array();
+        foreach ( $controls as $control ) {
+            $depth = 0;
+            for ( $ancestor = $control->parentNode; $ancestor instanceof DOMElement && ! $ancestor->isSameNode($form) && $depth < self::MAX_DEPTH; $ancestor = $ancestor->parentNode, ++$depth ) {
+                $path = $ancestor->getNodePath();
+                $owners[$path] = ($owners[$path] ?? 0) + 1;
+            }
+        }
+        foreach ( $controls as $index => $control ) {
+            $ancestors = array();
+            $depth = 0;
+            for ( $ancestor = $control->parentNode; $ancestor instanceof DOMElement && ! $ancestor->isSameNode($form) && $depth < self::MAX_DEPTH; $ancestor = $ancestor->parentNode, ++$depth ) {
+                if ( 1 === ($owners[$ancestor->getNodePath()] ?? 0) ) {
+                    $ancestors[] = $ancestor;
+                }
+            }
+            $result[$index] = $ancestors;
+        }
+        return $result;
+    }
+
     /**
      * @param array<string, int> $controlIndexes
      * @param array<string, bool> $relevantElements