Merge pull request #99 from pelson/feature/mathjax3

Add mathjax v3 support

Author: pelson

xkcd-script/font/xkcd-script.otf

@@ -0,0 +1,25 @@
+# xkcd-script font generation pipeline
+
+Each script runs in order inside the `fontbuilder` Docker image; `run.sh` orchestrates them and accepts an optional starting step (`./run.sh 5` skips pt1–pt4).
+
+## Stages
+
+| # | Script | What it does |
+|---|---|---|
+| 1 | `pt1_character_extraction.py` | Extract character strokes from `handwriting_minimal.png` (scikit-image). |
+| 2 | `pt2_character_classification.py` | Cluster strokes into lines (k-means, fixed seed). |
+| 3 | `pt3_ppm_to_svg.py` | Convert per-character PPM → SVG via `potrace`. |
+| 4 | `pt4_additional_sources.py` | Trace extra glyphs from comic panels and `extras/`. |
+| 5 | `pt5_svg_to_font.py` | Import SVG glyphs into a FontForge SFD; apply stroke normalisation, weight nudges, math-symbol imports. |
+| 6 | `pt6_derived_chars.py` | Build derived/composed glyphs: diacritics, ligatures, Greek, IPA, combining marks, math cmap aliases (U+1D400 block via altuni). |
+| 7 | `pt7_font_properties.py` | Apply kerning, GPOS anchors, pin CFF hints and OS/2 metrics. Output is `xkcd-script-pt7.sfd` — the **base** font used for everything downstream. |
+| 8 | `pt8_derivatives.py` | Orchestrator. Runs each `pt8X_*.py` derivative step in turn. |
+| 9 | `pt9_gen_reprod_font.py` | Scrub the SFD for reproducibility, freeze CFF charstrings, generate committed binaries (otf/ttf/woff). |
+
+## Derivatives (pt8)
+
+`pt7` produces a single kitchen-sink base SFD with everything — Latin, Greek, math symbols and aliases, ligatures, combining marks. Each `pt8X_<name>.py` reads that base and either writes its own derivative SFD or extracts data from it to splice elsewhere; `pt8_derivatives.py` runs them with `runpy`.
+
+Today there is no live derivative font: the sole entry, `pt8a_mathjax3.py`, only extracts extensible-glyph outline data into `../xkcd-mathjax3.js`. The display-sized large operators that used to live in a separate mathjax3 WOFF are now stylistic alternates in the base font (`ss01`).
+
+Because derivatives can only subtract or overlay what pt7 already has, **everything plausibly useful belongs in pt7**. Don't pre-strip pt7 for size — that loses the subtractive option.

xkcd-script/font/xkcd-script.sfd

@@ -72,17 +72,23 @@ def _clean_potrace_svg(raw_svg_path, clean_svg_path):
     os.remove(clean_svg_path + '.sfd')
 
 
-def extract_symbol(arr, y0, y1, x0, x1, name, exclude=None):
+def extract_symbol(arr, y0, y1, x0, x1, name, exclude=None, pad=0):
     """Crop glyph region, upsample, binarise, run potrace, clean, save SVG.
 
     exclude: optional list of (y0, y1, x0, x1) regions in full-image coordinates
              to blank out (set to background) before potrace, for removing
              artefacts that cannot be separated by tightening the main crop.
+    pad: white-pixel border added around the crop before upsampling.  Use
+         when the source PNG is tightly cropped and ink touches the edge —
+         potrace otherwise produces edge artefacts where contours run into
+         the canvas boundary.
     """
     crop = arr[y0:y1, x0:x1].copy()
     if exclude:
         for ey0, ey1, ex0, ex1 in exclude:
             crop[ey0 - y0:ey1 - y0, ex0 - x0:ex1 - x0] = 255
+    if pad:
+        crop = np.pad(crop, pad, mode='constant', constant_values=255)
     upsample = SPECIALUPSAMPLE.get(name, UPSAMPLE)
     big = Image.fromarray(crop).resize(
         (crop.shape[1] * upsample, crop.shape[0] * upsample),
@@ -175,14 +181,18 @@ def extract_symbol(arr, y0, y1, x0, x1, name, exclude=None):
     ('right_half_arrow', '2343_mathematical_symbol_fight_2x__right_half_arrow'),      # ⇀ U+21C0 source
     ('right_lim_arrow', '2343_mathematical_symbol_fight_2x__right_lim_arrow'),        # → U+2192 source
     ('triangle', '2343_mathematical_symbol_fight_2x__triangle'),              # △ U+25B3 source
+    ('circled_times', '2034_equations_2x__circled_times'),                    # ⊗ U+2297 source
+    ('sqrt_vertical', 'sqrt_vertical'),                                       # √ tall surd, unencoded glyph `radical.tall`
+    ('braceleft_tall', '2435_geothmetic_meandian_2x__brace__shortened'),      # { tall brace for MathJax stretchy assembly
+    ('parenleft_tall', '2059_modified_bayes_theorem_2x__lparen'),             # ( tall paren for MathJax stretchy assembly (\binom, \left(, pmatrix); right paren is mirrored in pt5
 ]
 
 print('Extracting hand-drawn extras...')
 for name, filename in EXTRAS:
     src_path = os.path.join(EXTRAS_DIR, f'{filename}.png')
     arr_extra = np.array(Image.open(src_path).convert('L'))
     h, w = arr_extra.shape
-    extract_symbol(arr_extra, 0, h, 0, w, name)
+    extract_symbol(arr_extra, 0, h, 0, w, name, pad=10)
 
 
 # ---------------------------------------------------------------------------