Bake glyphs from a host's own face

include/tile57.h

@@ -1750,6 +1750,41 @@ tile57_status tile57_bake_glyph_sdf(tile57_assets *out, tile57_error *err);
  * place names and italic hydrography from the SDF text path. Free with
  * tile57_assets_free. */
 tile57_status tile57_bake_glyph_sdf_face(tile57_assets *out, int32_t face, tile57_error *err);
+/* tile57_bake_glyph_sdf for named codepoints, out of a font the HOST supplies:
+ * `font_bytes` is a TrueType file or a collection, of which the first face is
+ * read, and `codepoints` names the characters to rasterize.
+ *
+ * The bundled Noto Sans covers Latin, Greek and Cyrillic, and the prebaked
+ * sheets carry printable ASCII and Latin-1. A chart naming its features in
+ * another script needs glyphs no prebaked sheet can hold — CJK alone is some
+ * 20,000 — so a host bakes the characters its labels actually use, from
+ * whatever face its platform has for the script, and merges the sheet into the
+ * atlas it already has.
+ *
+ * Same output as tile57_bake_glyph_sdf: only sprite_* filled, at the same em
+ * size and spread, so the two sheets share one atlas. A codepoint the face has
+ * no glyph for is left out rather than failing the bake. Free with
+ * tile57_assets_free. */
+tile57_status tile57_bake_glyph_sdf_codepoints(tile57_assets *out,
+                                               const uint8_t *font_bytes, size_t font_len,
+                                               const uint32_t *codepoints, size_t count,
+                                               tile57_error *err);
+/* True when `font_bytes` can draw `codepoint`: the face has a glyph for it AND
+ * the rasterizer can read that glyph's outlines.
+ *
+ * A host picking a face out of its platform's font store has to ask, because
+ * the two are not the same question. Apple's UI faces cover CJK and are what
+ * CoreText hands back for Chinese text, but they carry Apple's `cidg` outlines
+ * rather than glyf or CFF, and no glyph comes out of them. A host that
+ * installed one would draw blank labels with nothing to say why. Ask with one
+ * character of the script, then install the first face that answers true. */
+tile57_status tile57_font_covers(const uint8_t *font_bytes, size_t font_len,
+                                 uint32_t codepoint, bool *out, tile57_error *err);
+/* The same question about the BUNDLED label face, so a host can tell which
+ * scripts it needs a platform font for and which are already drawn. A chart
+ * stating four languages may need a face for one of them: asking first stops a
+ * host from mapping a font for Finnish and leaving Inuktitut blank. */
+tile57_status tile57_label_font_covers(uint32_t codepoint, bool *out, tile57_error *err);
 void tile57_assets_free(tile57_assets *out);
 
 /* ---- chart-style generation ---------------------------------------------

src/capi.zig

@@ -2209,6 +2209,76 @@ export fn tile57_bake_glyph_sdf_face(out: ?*CAssets, face: i32, err: ?*CError) c
     return bakeGlyphSdf(out, face, err);
 }
 
+/// An SDF sheet for named codepoints, out of a font the HOST supplies. See
+/// tile57.h.
+export fn tile57_bake_glyph_sdf_codepoints(
+    out: ?*CAssets,
+    font_bytes: ?[*]const u8,
+    font_len: usize,
+    codepoints: ?[*]const u32,
+    count: usize,
+    err: ?*CError,
+) callconv(.c) c_int {
+    const o = out orelse return failWith(err, .badarg, "out must not be null");
+    o.* = .{};
+    const fb = font_bytes orelse return failWith(err, .badarg, "font_bytes must not be null");
+    if (font_len == 0) return failWith(err, .badarg, "font_len must not be zero");
+    const cps_c = codepoints orelse return failWith(err, .badarg, "codepoints must not be null");
+    if (count == 0) return failWith(err, .badarg, "count must not be zero");
+
+    var arena = std.heap.ArenaAllocator.init(gpa);
+    defer arena.deinit();
+    const a = arena.allocator();
+    const cps = a.alloc(u21, count) catch |e| return fail(err, e);
+    for (cps_c[0..count], cps) |src, *dst| {
+        if (src > 0x10FFFF) return failWith(err, .badarg, "codepoint above the Unicode range");
+        dst.* = @intCast(src);
+    }
+    var atlas = glyph_sdf.build(a, fb[0..font_len], cps, 32.0, 6) catch |e| return fail(err, e);
+    // Every codepoint came back with neither ink nor an advance. Either the
+    // face draws none of them, or its outlines are in a format the rasterizer
+    // cannot read. Say so: a host that took an empty sheet as success would
+    // record the characters as served and never ask for them again.
+    if (atlas.glyphs.count() == 0)
+        return failWith(err, .unsupported, "the face draws none of those codepoints");
+    return finishGlyphSdf(o, a, &atlas, err);
+}
+
+/// True when the BUNDLED label face can draw `codepoint`. See tile57.h.
+export fn tile57_label_font_covers(codepoint: u32, out: ?*bool, err: ?*CError) callconv(.c) c_int {
+    const o = out orelse return failWith(err, .badarg, "out must not be null");
+    o.* = false;
+    if (codepoint > 0x10FFFF) return failWith(err, .badarg, "codepoint above the Unicode range");
+    var arena = std.heap.ArenaAllocator.init(gpa);
+    defer arena.deinit();
+    const one = [_]u21{@intCast(codepoint)};
+    const ft = @import("render").font;
+    var atlas = glyph_sdf.build(arena.allocator(), ft.notosans, &one, 32.0, 6) catch |e| return fail(err, e);
+    o.* = atlas.glyphs.count() > 0;
+    return OK;
+}
+
+/// True when `font_bytes` can draw `codepoint`. See tile57.h.
+export fn tile57_font_covers(
+    font_bytes: ?[*]const u8,
+    font_len: usize,
+    codepoint: u32,
+    out: ?*bool,
+    err: ?*CError,
+) callconv(.c) c_int {
+    const o = out orelse return failWith(err, .badarg, "out must not be null");
+    o.* = false;
+    const fb = font_bytes orelse return failWith(err, .badarg, "font_bytes must not be null");
+    if (font_len == 0) return failWith(err, .badarg, "font_len must not be zero");
+    if (codepoint > 0x10FFFF) return failWith(err, .badarg, "codepoint above the Unicode range");
+    var arena = std.heap.ArenaAllocator.init(gpa);
+    defer arena.deinit();
+    const one = [_]u21{@intCast(codepoint)};
+    var atlas = glyph_sdf.build(arena.allocator(), fb[0..font_len], &one, 32.0, 6) catch |e| return fail(err, e);
+    o.* = atlas.glyphs.count() > 0;
+    return OK;
+}
+
 fn bakeGlyphSdf(out: ?*CAssets, face: i32, err: ?*CError) c_int {
     const o = out orelse return failWith(err, .badarg, "out must not be null");
     o.* = .{};
@@ -2223,9 +2293,16 @@ fn bakeGlyphSdf(out: ?*CAssets, face: i32, err: ?*CError) c_int {
     };
     const cps = glyph_sdf.defaultCodepoints(a) catch |e| return fail(err, e);
     var atlas = glyph_sdf.build(a, font, cps, 32.0, 6) catch |e| return fail(err, e);
+    return finishGlyphSdf(o, a, &atlas, err);
+}
+
+/// Encode a built atlas into *out. `a` is the arena the atlas lives in; the two
+/// buffers that leave are duped into gpa, which is what tile57_assets_free
+/// releases.
+fn finishGlyphSdf(o: *CAssets, a: std.mem.Allocator, atlas: *glyph_sdf.Atlas, err: ?*CError) c_int {
     const png = (atlas.encodePng(a) catch |e| return fail(err, e)) orelse
         return failWith(err, .internal, "glyph atlas PNG encode produced nothing");
-    const json = glyphMetricsJson(a, &atlas) catch |e| return fail(err, e);
+    const json = glyphMetricsJson(a, atlas) catch |e| return fail(err, e);
     o.sprite_png = (gpa.dupe(u8, png) catch |e| {
         tile57_assets_free(o);
         return fail(err, e);

src/sprite/glyph.zig

@@ -9,6 +9,14 @@ const Allocator = std.mem.Allocator;
 // C glue (src/sprite/svgraster.c): stb_truetype SDF + the shared PNG encoder.
 extern fn tg_glyph_sdf(font: [*]const u8, font_len: c_int, cp: c_int, em_px: f32, pad: c_int, onedge: c_int, dist_scale: f32, w: *c_int, h: *c_int, xoff: *c_int, yoff: *c_int, advance: *f32) callconv(.c) ?[*]u8;
 extern fn tg_glyph_free(p: ?[*]u8) callconv(.c) void;
+extern fn tg_glyph_present(font: [*]const u8, font_len: c_int, cp: c_int) callconv(.c) c_int;
+
+/// True when `font` has a glyph of its own for `cp`. A codepoint the face does
+/// not map resolves to .notdef, which has an advance and usually a box, so the
+/// metrics a rasterizer returns cannot answer this on their own.
+pub fn present(font: []const u8, cp: u21) bool {
+    return tg_glyph_present(font.ptr, @intCast(font.len), @intCast(cp)) != 0;
+}
 extern fn tg_png_encode(rgba: [*]const u8, w: c_int, h: c_int, out_len: *c_int) callconv(.c) ?[*]u8;
 extern fn tg_svg_free(p: ?*anyopaque) callconv(.c) void;
 
@@ -90,6 +98,16 @@ pub fn build(a: Allocator, font: []const u8, cps: []const u21, em_px: f32, pad:
             .h = @as(f32, @floatFromInt(h)) / em_px,
             .advance = adv / em_px,
         };
+        // A codepoint the face does not map resolves to .notdef, whose box
+        // would be baked as if it were the character. And a face whose
+        // outlines the rasterizer cannot read at all (Apple's variable UI
+        // faces carry `cidg` rather than glyf or CFF) yields neither ink nor
+        // an advance, which a host would pack into its atlas and lay out as
+        // nothing — a blank label that says the glyph arrived. Skip both: an
+        // absent codepoint is what "this face cannot draw it" means. A space
+        // has a glyph of its own and advances, so it survives.
+        if (!present(font, cp)) continue;
+        if (sdf == null and adv <= 0) continue;
         try cells.append(a, .{ .cp = cp, .w = @intCast(@max(w, 0)), .h = @intCast(@max(h, 0)), .sdf = sdf, .gi = gi });
     }
 

src/sprite/svgraster.c

@@ -47,6 +47,19 @@ unsigned char *tg_glyph_sdf(const unsigned char *font, int font_len, int cp,
 }
 void tg_glyph_free(unsigned char *p) { stbtt_FreeSDF(p, NULL); }
 
+// Whether the face has a glyph of its own for `cp`. stbtt_FindGlyphIndex
+// answers 0 for a codepoint the cmap does not map, which is .notdef — a real
+// glyph with a real advance and, in most faces, a visible box. Metrics alone
+// therefore cannot tell a character the face draws from one it does not, and a
+// caller that trusted them would bake a row of tofu and record it as the text.
+// 1 when the face has the glyph, 0 when it does not or cannot be read.
+int tg_glyph_present(const unsigned char *font, int font_len, int cp) {
+    (void)font_len;
+    stbtt_fontinfo f;
+    if (!stbtt_InitFont(&f, font, stbtt_GetFontOffsetForIndex(font, 0))) return 0;
+    return stbtt_FindGlyphIndex(&f, cp) != 0;
+}
+
 // Rasterize a flattened SVG (viewBox normalized to "0 0 W H") at `scale` device
 // px per user unit. Forces even-odd winding on every shape — the S-101 danger
 // glyphs are compound paths whose inner subpath is a hole, and nonzero winding