render: retain authored webfont identity

crates/obscura-render/src/dom.rs

@@ -410,6 +410,24 @@ pub fn layout_dom_with_resources(
     viewport: (f32, f32),
     intrinsic: &HashMap<NodeId, (f32, f32)>,
     fonts: &[Vec<u8>],
+) -> DomLayout {
+    let fonts: Vec<_> = fonts
+        .iter()
+        .map(|data| crate::inline::WebFont {
+            data: data.clone(),
+            family: None,
+            weight: None,
+            italic: None,
+        })
+        .collect();
+    layout_dom_with_web_fonts(tree, viewport, intrinsic, &fonts)
+}
+
+pub(crate) fn layout_dom_with_web_fonts(
+    tree: &DomTree,
+    viewport: (f32, f32),
+    intrinsic: &HashMap<NodeId, (f32, f32)>,
+    fonts: &[crate::inline::WebFont],
 ) -> DomLayout {
     let timing = std::env::var("OBSCURA_RENDER_TIMING").is_ok();
 
@@ -467,7 +485,7 @@ pub fn layout_dom_with_resources(
     let mut taffy_tree: TaffyTree<usize> = TaffyTree::new();
     let mut id_map: HashMap<taffy::NodeId, NodeId> = HashMap::new();
     let mut words: HashMap<taffy::NodeId, (NodeId, String)> = HashMap::new();
-    let mut engine = crate::inline::TextEngine::new_with_fonts(fonts);
+    let mut engine = crate::inline::TextEngine::new_with_web_fonts(fonts);
     let mut ifc_items = IfcRegistry::default();
 
     // The document node itself is not an element; lay out from the first

crates/obscura-render/src/inline.rs

@@ -80,14 +80,31 @@ fn resolve_font_family(fam: Option<&str>) -> &'static str {
 
 #[derive(Clone)]
 struct LoadedFamily {
+    faces: Vec<LoadedFace>,
+}
+
+#[derive(Clone)]
+struct LoadedFace {
     name: Arc<str>,
-    weights: Vec<u16>,
+    min_weight: u16,
+    max_weight: u16,
+    shape_weight: u16,
+    italic: bool,
     variable: bool,
 }
 
+#[derive(Clone)]
+pub(crate) struct WebFont {
+    pub data: Vec<u8>,
+    pub family: Option<String>,
+    pub weight: Option<(u16, u16)>,
+    pub italic: Option<bool>,
+}
+
 fn resolve_loaded_font(
     fam: Option<&str>,
     requested_weight: u16,
+    requested_italic: bool,
     loaded: &HashMap<String, LoadedFamily>,
 ) -> (Arc<str>, u16) {
     if let Some(stack) = fam {
@@ -97,12 +114,35 @@ fn resolve_loaded_font(
                 .trim_matches(|c| c == '"' || c == '\'')
                 .trim();
             if let Some(family) = loaded.get(&name.to_ascii_lowercase()) {
-                let weight = if family.variable {
+                let exact_style: Vec<_> = family
-                    requested_weight
+                    .faces
+                    .iter()
+                    .filter(|face| face.italic == requested_italic)
+                    .collect();
+                let candidates: Vec<_> = if exact_style.is_empty() {
+                    family.faces.iter().collect()
                 } else {
-                    match_font_weight(requested_weight, &family.weights)
+                    exact_style
                 };
-                return (Arc::clone(&family.name), weight);
+                if let Some(face) = candidates.iter().copied().find(|face| {
+                    (face.min_weight..=face.max_weight).contains(&requested_weight)
+                }) {
+                    let weight = if face.variable {
+                        requested_weight
+                    } else {
+                        face.shape_weight
+                    };
+                    return (Arc::clone(&face.name), weight);
+                }
+                let available: Vec<_> =
+                    candidates.iter().map(|face| face.min_weight).collect();
+                let matched = match_font_weight(requested_weight, &available);
+                if let Some(face) = candidates
+                    .into_iter()
+                    .find(|face| face.min_weight == matched)
+                {
+                    return (Arc::clone(&face.name), face.shape_weight);
+                }
             }
         }
     }
@@ -316,42 +356,76 @@ impl TextEngine {
     }
 
     pub fn new_with_fonts(fonts: &[Vec<u8>]) -> Self {
-        // Build a database from our four embedded faces only. Never call
+        let fonts: Vec<_> = fonts
+            .iter()
+            .map(|data| WebFont {
+                data: data.clone(),
+                family: None,
+                weight: None,
+                italic: None,
+            })
+            .collect();
+        Self::new_with_web_fonts(&fonts)
+    }
+
+    pub(crate) fn new_with_web_fonts(fonts: &[WebFont]) -> Self {
+        // Build a database from embedded and page-provided faces. Never call
         // load_system_fonts: a host's font set would make layout differ
         // machine to machine and add a multi-millisecond startup scan.
         let mut db = cosmic_text::fontdb::Database::new();
+        let mut declarations = Vec::new();
         for bytes in [
             SANS_R, SANS_B, SANS_O, SANS_BO,
             SERIF_R, SERIF_B, SERIF_O, SERIF_BO,
             MONO_R, MONO_B, MONO_O, MONO_BO,
             FALLBACK,
         ] {
-            db.load_font_source(cosmic_text::fontdb::Source::Binary(Arc::new(bytes)));
+            for id in db.load_font_source(cosmic_text::fontdb::Source::Binary(Arc::new(bytes))) {
+                declarations.push((id, None, None, None));
+            }
         }
-        for bytes in fonts {
+        for font in fonts {
-            db.load_font_source(cosmic_text::fontdb::Source::Binary(Arc::new(
+            for id in db.load_font_source(cosmic_text::fontdb::Source::Binary(Arc::new(
-                bytes.clone(),
+                font.data.clone(),
-            )));
+            ))) {
+                declarations.push((
+                    id,
+                    font.family.clone(),
+                    font.weight,
+                    font.italic,
+                ));
+            }
         }
         let mut loaded_families = HashMap::new();
-        let faces: Vec<_> = db
+        for (id, declared_family, declared_weight, declared_italic) in declarations {
-            .faces()
+            let Some(face) = db.face(id) else { continue };
-            .map(|face| (face.id, face.families.clone(), face.weight.0))
+            let names = face.families.clone();
-            .collect();
+            let internal_name = names
-        for (id, names, weight) in faces {
+                .first()
+                .map(|(name, _)| Arc::<str>::from(name.as_str()))
+                .unwrap_or_else(|| Arc::from(FAMILY));
+            let shape_weight = face.weight.0;
+            let italic = declared_italic
+                .unwrap_or(!matches!(face.style, cosmic_text::fontdb::Style::Normal));
             let variable = db
                 .with_face_data(id, face_is_variable)
                 .unwrap_or(false);
-            for (name, _) in names {
+            let weight = declared_weight.unwrap_or((shape_weight, shape_weight));
+            let declared_names: Vec<String> = declared_family
+                .map(|name| vec![name])
+                .unwrap_or_else(|| names.into_iter().map(|(name, _)| name).collect());
+            for name in declared_names {
                 let family = loaded_families
                     .entry(name.to_ascii_lowercase())
-                    .or_insert_with(|| LoadedFamily {
+                    .or_insert_with(|| LoadedFamily { faces: Vec::new() });
-                        name: Arc::from(name.as_str()),
+                family.faces.push(LoadedFace {
-                        weights: Vec::new(),
+                    name: Arc::clone(&internal_name),
-                        variable: false,
+                    min_weight: weight.0,
-                    });
+                    max_weight: weight.1,
-                family.weights.push(weight);
+                    shape_weight,
-                family.variable |= variable;
+                    italic,
+                    variable,
+                });
             }
         }
         db.set_sans_serif_family(FAMILY);
@@ -387,6 +461,7 @@ impl TextEngine {
         let (family, weight) = resolve_loaded_font(
             base.font_family.as_deref(),
             crate::style::used_font_weight(base),
+            base.font_style_italic.unwrap_or(false),
             &self.loaded_families,
         );
         let ctx = base_span_ctx(
@@ -436,6 +511,7 @@ impl TextEngine {
         let (family, weight) = resolve_loaded_font(
             base.font_family.as_deref(),
             crate::style::used_font_weight(base),
+            base.font_style_italic.unwrap_or(false),
             &self.loaded_families,
         );
         let ctx = base_span_ctx(
@@ -775,6 +851,9 @@ fn collect_node_spans(
                     resolve_loaded_font(
                         Some(family),
                         requested_weight,
+                        style
+                            .and_then(|style| style.font_style_italic)
+                            .unwrap_or(ctx.italic),
                         loaded_families,
                     )
                 })
@@ -1224,6 +1303,45 @@ mod tests {
         assert_eq!(match_font_weight(500, &[400, 500, 700]), 500);
     }
 
+    #[test]
+    fn declared_family_selects_a_face_with_a_different_internal_name() {
+        let family = LoadedFamily {
+            faces: vec![
+                LoadedFace {
+                    name: Arc::from("Poppins"),
+                    min_weight: 400,
+                    max_weight: 400,
+                    shape_weight: 400,
+                    italic: false,
+                    variable: false,
+                },
+                LoadedFace {
+                    name: Arc::from("Poppins Medium"),
+                    min_weight: 500,
+                    max_weight: 500,
+                    shape_weight: 500,
+                    italic: false,
+                    variable: false,
+                },
+                LoadedFace {
+                    name: Arc::from("Poppins"),
+                    min_weight: 700,
+                    max_weight: 700,
+                    shape_weight: 700,
+                    italic: false,
+                    variable: false,
+                },
+            ],
+        };
+        let loaded = HashMap::from([("poppins".to_string(), family)]);
+        let medium = resolve_loaded_font(Some("Poppins, sans-serif"), 500, false, &loaded);
+        assert_eq!(medium.0.as_ref(), "Poppins Medium");
+        assert_eq!(medium.1, 500);
+        let semibold = resolve_loaded_font(Some("Poppins"), 600, false, &loaded);
+        assert_eq!(semibold.0.as_ref(), "Poppins");
+        assert_eq!(semibold.1, 700);
+    }
+
     #[test]
     fn clip_fill_only_for_transparent_clip_text() {
         let mut s = LayoutStyle::default();

crates/obscura-render/src/paint.rs

@@ -13,7 +13,7 @@ static FONT_BYTES: &[u8] = include_bytes!("../assets/liberation-sans.ttf");
 static SERIF_FONT_BYTES: &[u8] = include_bytes!("../assets/liberation-serif.ttf");
 static MONO_FONT_BYTES: &[u8] = include_bytes!("../assets/liberation-mono.ttf");
 
-use crate::layout_dom_with_resources;
+use crate::dom::layout_dom_with_web_fonts;
 
 /// Render `tree` at `viewport` (width, height) in CSS pixels to a Pixmap, or
 /// None if the viewport is zero-sized. `base_url`, when given, resolves the
@@ -36,7 +36,7 @@ pub fn paint_dom(tree: &DomTree, viewport: (f32, f32), base_url: Option<&str>) -
     // each URL is still fetched at most once.
     let intrinsic = collect_image_intrinsics(tree, base_url, &mut image_cache);
     let fonts = collect_web_fonts(tree, base_url);
-    let mut laid = layout_dom_with_resources(tree, viewport, &intrinsic, &fonts);
+    let mut laid = layout_dom_with_web_fonts(tree, viewport, &intrinsic, &fonts);
     let root_font_size = tree
         .query_selector("html")
         .ok()
@@ -953,15 +953,42 @@ fn fetch_bytes(
 /// filtering is load-bearing for performance: generated font packages commonly
 /// emit six or seven script subsets per face, while an English page needs only
 /// the subset containing ASCII.
-fn collect_web_fonts(tree: &DomTree, base_url: Option<&str>) -> Vec<Vec<u8>> {
+fn collect_web_fonts(tree: &DomTree, base_url: Option<&str>) -> Vec<crate::inline::WebFont> {
     let mut cache = std::collections::HashMap::new();
     let mut seen = std::collections::HashSet::new();
     let mut fonts = Vec::new();
+    let mut rules = Vec::new();
+
+    for nid in tree.descendants(tree.document()) {
+        let Some(node) = tree.get_node(nid) else { continue };
+        if node
+            .as_element()
+            .map(|element| element.local.as_ref() != "style")
+            .unwrap_or(true)
+        {
+            continue;
+        }
+        let css = tree.text_content(nid);
+        for face in font_face_blocks(&css) {
+            if !font_face_covers_ascii(face) {
+                continue;
+            }
+            let Some(src) = font_face_urls(face).into_iter().next() else {
+                continue;
+            };
+            rules.push((
+                font_resource_key(&src, base_url),
+                src,
+                font_face_family(face),
+                font_face_weight(face),
+                font_face_italic(face),
+            ));
+        }
+    }
 
     // Critical web fonts are normally preloaded from the document with a URL
-    // already resolved relative to the HTML. Prefer that browser-authored list:
+    // already resolved relative to the HTML. Fetch those first, while retaining
-    // CSS gathered from external stylesheets is flattened into <style> text in
+    // the matching @font-face descriptors needed for CSS family/weight lookup.
-    // the live DOM, where a relative src no longer carries its stylesheet URL.
     let mut preloads = Vec::new();
     for nid in tree.descendants(tree.document()) {
         let Some(node) = tree.get_node(nid) else { continue };
@@ -985,50 +1012,52 @@ fn collect_web_fonts(tree: &DomTree, base_url: Option<&str>) -> Vec<Vec<u8>> {
         }
     }
     for src in preloads.iter().take(16) {
-        if !seen.insert(src.clone()) {
+        let key = font_resource_key(src, base_url);
+        if !seen.insert(key.clone()) {
             continue;
         }
         if let Some(decoded) = fetch_and_decode_font(src, base_url, &mut cache) {
-            fonts.push(decoded);
+            let metadata = rules.iter().find(|rule| rule.0 == key);
+            fonts.push(crate::inline::WebFont {
+                data: decoded,
+                family: metadata.and_then(|rule| rule.2.clone()),
+                weight: metadata.and_then(|rule| rule.3),
+                italic: metadata.and_then(|rule| rule.4),
+            });
         }
     }
-    if !preloads.is_empty() {
-        return fonts;
-    }
 
-    for nid in tree.descendants(tree.document()) {
+    for (key, src, family, weight, italic) in rules {
-        let Some(node) = tree.get_node(nid) else { continue };
+        if fonts.len() >= 16 {
-        if node
+            break;
-            .as_element()
+        }
-            .map(|element| element.local.as_ref() != "style")
+        if !seen.insert(key) {
-            .unwrap_or(true)
-        {
             continue;
         }
-        let css = tree.text_content(nid);
+        if let Some(decoded) = fetch_and_decode_font(&src, base_url, &mut cache) {
-        for face in font_face_blocks(&css) {
+            fonts.push(crate::inline::WebFont {
-            if !font_face_covers_ascii(face) {
+                data: decoded,
-                continue;
+                family,
-            }
+                weight,
-            let Some(src) = font_face_urls(face).into_iter().next() else {
+                italic,
-                continue;
+            });
-            };
-            if !seen.insert(src.clone()) {
-                continue;
-            }
-            // Bound pathological pages without penalizing normal generated
-            // font packages (family x weight x style usually stays under 20).
-            if fonts.len() >= 12 {
-                return fonts;
-            }
-            if let Some(decoded) = fetch_and_decode_font(&src, base_url, &mut cache) {
-                fonts.push(decoded);
-            }
         }
     }
     fonts
 }
 
+fn font_resource_key(src: &str, base_url: Option<&str>) -> String {
+    url::Url::parse(src)
+        .ok()
+        .or_else(|| {
+            base_url
+                .and_then(|base| url::Url::parse(base).ok())
+                .and_then(|base| base.join(src).ok())
+        })
+        .map(|url| url.to_string())
+        .unwrap_or_else(|| src.to_string())
+}
+
 fn fetch_and_decode_font(
     src: &str,
     base_url: Option<&str>,
@@ -1108,6 +1137,45 @@ fn font_face_declaration<'a>(face: &'a str, name: &str) -> Option<&'a str> {
     })
 }
 
+fn font_face_family(face: &str) -> Option<String> {
+    font_face_declaration(face, "font-family")
+        .map(|family| family.trim().trim_matches(|ch| matches!(ch, '"' | '\'')).to_string())
+        .filter(|family| !family.is_empty())
+}
+
+fn font_face_weight(face: &str) -> Option<(u16, u16)> {
+    fn parse(value: &str) -> Option<u16> {
+        match value.to_ascii_lowercase().as_str() {
+            "normal" => Some(400),
+            "bold" => Some(700),
+            value => value
+                .parse::<f32>()
+                .ok()
+                .filter(|weight| weight.is_finite() && (1.0..=1000.0).contains(weight))
+                .map(|weight| weight.round() as u16),
+        }
+    }
+    let mut values = font_face_declaration(face, "font-weight")?
+        .split_ascii_whitespace()
+        .filter_map(parse);
+    let first = values.next()?;
+    let second = values.next().unwrap_or(first);
+    Some((first.min(second), first.max(second)))
+}
+
+fn font_face_italic(face: &str) -> Option<bool> {
+    font_face_declaration(face, "font-style").and_then(|style| {
+        let style = style.trim().to_ascii_lowercase();
+        if style == "normal" {
+            Some(false)
+        } else if style == "italic" || style.starts_with("oblique") {
+            Some(true)
+        } else {
+            None
+        }
+    })
+}
+
 fn font_face_covers_ascii(face: &str) -> bool {
     let Some(range) = font_face_declaration(face, "unicode-range") else {
         return true;
@@ -3187,6 +3255,8 @@ mod tests {
             }
             @font-face {
                 font-family: "Example";
+                font-style: italic;
+                font-weight: 350 650;
                 src: url(data:font/woff2;base64,d09GMg==) format("woff2"),
                      url("./example-latin.woff") format("woff");
                 unicode-range: U+??, U+2000-206F;
@@ -3196,6 +3266,9 @@ mod tests {
         assert_eq!(faces.len(), 2);
         assert!(!font_face_covers_ascii(faces[0]));
         assert!(font_face_covers_ascii(faces[1]));
+        assert_eq!(font_face_family(faces[1]).as_deref(), Some("Example"));
+        assert_eq!(font_face_weight(faces[1]), Some((350, 650)));
+        assert_eq!(font_face_italic(faces[1]), Some(true));
         assert_eq!(
             font_face_urls(faces[1]),
             vec![