ENH: implement font fallback for PDF

efiring · Aug 11, 2022 · Aug 7, 2022 · Aug 9, 2022 · Jul 28, 2021 · Aug 13, 2021
commit c5fd8804204ee715ee008c35f96d6e95f8dfcc29
diff --git a/lib/matplotlib/_text_helpers.py b/lib/matplotlib/_text_helpers.py
@@ -9,7 +9,7 @@
 
 
 LayoutItem = dataclasses.make_dataclass(
-    "LayoutItem", ["char", "glyph_idx", "x", "prev_kern"])
+    "LayoutItem", ["ft_object", "char", "glyph_idx", "x", "prev_kern"])
 
 
 def warn_on_missing_glyph(codepoint):
@@ -57,12 +57,18 @@ def layout(string, font, *, kern_mode=KERNING_DEFAULT):
     """
     x = 0
     prev_glyph_idx = None
+    char_to_font = font._get_fontmap(string)
+    base_font = font
     for char in string:
+        # This has done the fallback logic
+        font = char_to_font.get(char, base_font)
         glyph_idx = font.get_char_index(ord(char))
-        kern = (font.get_kerning(prev_glyph_idx, glyph_idx, kern_mode) / 64
-                if prev_glyph_idx is not None else 0.)
+        kern = (
+            base_font.get_kerning(prev_glyph_idx, glyph_idx, kern_mode) / 64
+            if prev_glyph_idx is not None else 0.
+        )
         x += kern
         glyph = font.load_glyph(glyph_idx, flags=LOAD_NO_HINTING)
-        yield LayoutItem(char, glyph_idx, x, kern)
+        yield LayoutItem(font, char, glyph_idx, x, kern)
         x += glyph.linearHoriAdvance / 65536
         prev_glyph_idx = glyph_idx
diff --git a/lib/matplotlib/backends/_backend_pdf_ps.py b/lib/matplotlib/backends/_backend_pdf_ps.py
@@ -137,8 +137,8 @@ def _get_font_afm(self, prop):
         return _cached_get_afm_from_fname(fname)
 
     def _get_font_ttf(self, prop):
-        fname = font_manager.findfont(prop)
-        font = font_manager.get_font(fname)
+        fnames = font_manager.fontManager._find_fonts_by_props(prop)
+        font = font_manager.get_font(fnames)
         font.clear()
         font.set_size(prop.get_size_in_points(), 72)
         return font
diff --git a/lib/matplotlib/backends/backend_pdf.py b/lib/matplotlib/backends/backend_pdf.py
@@ -32,7 +32,9 @@
     RendererBase)
 from matplotlib.backends.backend_mixed import MixedModeRenderer
 from matplotlib.figure import Figure
-from matplotlib.font_manager import findfont, get_font
+from matplotlib.font_manager import (
+    findfont, get_font, fontManager as _fontManager
+)
 from matplotlib._afm import AFM
 from matplotlib.ft2font import (FIXED_WIDTH, ITALIC, LOAD_NO_SCALE,
                                 LOAD_NO_HINTING, KERNING_UNFITTED, FT2Font)
@@ -925,20 +927,28 @@ def fontName(self, fontprop):
         """
 
         if isinstance(fontprop, str):
-            filename = fontprop
+            filenames = [fontprop]
         elif mpl.rcParams['pdf.use14corefonts']:
-            filename = findfont(
-                fontprop, fontext='afm', directory=RendererPdf._afm_font_dir)
+            filenames = _fontManager._find_fonts_by_props(
+                fontprop, fontext='afm', directory=RendererPdf._afm_font_dir
+            )
         else:
-            filename = findfont(fontprop)
-
-        Fx = self.fontNames.get(filename)
-        if Fx is None:
-            Fx = next(self._internal_font_seq)
-            self.fontNames[filename] = Fx
-            _log.debug('Assigning font %s = %r', Fx, filename)
-
-        return Fx
+            filenames = _fontManager._find_fonts_by_props(fontprop)
+        first_Fx = None
+        for fname in filenames:
+            Fx = self.fontNames.get(fname)
+            if not first_Fx:
+                first_Fx = Fx
+            if Fx is None:
+                Fx = next(self._internal_font_seq)
+                self.fontNames[fname] = Fx
+                _log.debug('Assigning font %s = %r', Fx, fname)
+                if not first_Fx:
+                    first_Fx = Fx
+
+        # find_fontsprop's first value always adheres to
+        # findfont's value, so technically no behaviour change
+        return first_Fx
 
     def dviFontName(self, dvifont):
         """
@@ -1204,7 +1214,6 @@ def get_char_width(charcode):
                 width = font.load_char(
                     s, flags=LOAD_NO_SCALE | LOAD_NO_HINTING).horiAdvance
                 return cvt(width)
-
             with warnings.catch_warnings():
                 # Ignore 'Required glyph missing from current font' warning
                 # from ft2font: here we're just building the widths table, but
@@ -2389,22 +2398,27 @@ def draw_text(self, gc, x, y, s, prop, angle, ismath=False, mtext=None):
         # the regular text show command (TJ) with appropriate kerning between
         # chunks, whereas multibyte characters use the XObject command (Do).
         else:
-            # List of (start_x, [prev_kern, char, char, ...]), w/o zero kerns.
+            # List of (ft_object, start_x, [prev_kern, char, char, ...]),
+            # w/o zero kerns.
             singlebyte_chunks = []
-            # List of (start_x, glyph_index).
+            # List of (ft_object, start_x, glyph_index).
             multibyte_glyphs = []
             prev_was_multibyte = True
+            prev_font = font
             for item in _text_helpers.layout(
                     s, font, kern_mode=KERNING_UNFITTED):
                 if _font_supports_glyph(fonttype, ord(item.char)):
-                    if prev_was_multibyte:
-                        singlebyte_chunks.append((item.x, []))
+                    if prev_was_multibyte or item.ft_object != prev_font:
+                        singlebyte_chunks.append((item.ft_object, item.x, []))
+                        prev_font = item.ft_object
                     if item.prev_kern:
-                        singlebyte_chunks[-1][1].append(item.prev_kern)
-                    singlebyte_chunks[-1][1].append(item.char)
+                        singlebyte_chunks[-1][2].append(item.prev_kern)
+                    singlebyte_chunks[-1][2].append(item.char)
                     prev_was_multibyte = False
                 else:
-                    multibyte_glyphs.append((item.x, item.glyph_idx))
+                    multibyte_glyphs.append(
+                        (item.ft_object, item.x, item.glyph_idx)
+                    )
                     prev_was_multibyte = True
             # Do the rotation and global translation as a single matrix
             # concatenation up front
@@ -2414,10 +2428,12 @@ def draw_text(self, gc, x, y, s, prop, angle, ismath=False, mtext=None):
                              -math.sin(a), math.cos(a),
                              x, y, Op.concat_matrix)
             # Emit all the 1-byte characters in a BT/ET group.
-            self.file.output(Op.begin_text,
-                             self.file.fontName(prop), fontsize, Op.selectfont)
+
+            self.file.output(Op.begin_text)
             prev_start_x = 0
-            for start_x, kerns_or_chars in singlebyte_chunks:
+            for ft_object, start_x, kerns_or_chars in singlebyte_chunks:
+                ft_name = self.file.fontName(ft_object.fname)
+                self.file.output(ft_name, fontsize, Op.selectfont)
                 self._setup_textpos(start_x, 0, 0, prev_start_x, 0, 0)
                 self.file.output(
                     # See pdf spec "Text space details" for the 1000/fontsize
@@ -2429,8 +2445,10 @@ def draw_text(self, gc, x, y, s, prop, angle, ismath=False, mtext=None):
                 prev_start_x = start_x
             self.file.output(Op.end_text)
             # Then emit all the multibyte characters, one at a time.
-            for start_x, glyph_idx in multibyte_glyphs:
-                self._draw_xobject_glyph(font, fontsize, glyph_idx, start_x, 0)
+            for ft_object, start_x, glyph_idx in multibyte_glyphs:
+                self._draw_xobject_glyph(
+                    ft_object, fontsize, glyph_idx, start_x, 0
+                )
             self.file.output(Op.grestore)
 
     def _draw_xobject_glyph(self, font, fontsize, glyph_idx, x, y):

diff --git a/lib/matplotlib/tests/baseline_images/test_backend_pdf/multi_font_type3.pdf b/lib/matplotlib/tests/baseline_images/test_backend_pdf/multi_font_type3.pdf
diff --git a/lib/matplotlib/tests/baseline_images/test_backend_pdf/multi_font_type42.pdf b/lib/matplotlib/tests/baseline_images/test_backend_pdf/multi_font_type42.pdf
diff --git a/lib/matplotlib/tests/test_backend_pdf.py b/lib/matplotlib/tests/test_backend_pdf.py
@@ -9,7 +9,9 @@
 import pytest
 
 import matplotlib as mpl
-from matplotlib import pyplot as plt, rcParams
+from matplotlib import (
+    pyplot as plt, rcParams, font_manager as fm
+)
 from matplotlib.cbook import _get_data_path
 from matplotlib.ft2font import FT2Font
 from matplotlib.font_manager import findfont, FontProperties
@@ -383,3 +385,29 @@ def test_glyphs_subset():
 
     # since both objects are assigned same characters
     assert subfont.get_num_glyphs() == nosubfont.get_num_glyphs()
+
+
+@image_comparison(["multi_font_type3.pdf"])
+def test_multi_font_type3():
+    fp = fm.FontProperties(family=["WenQuanYi Zen Hei"])
+    if Path(fm.findfont(fp)).name != "wqy-zenhei.ttc":
+        pytest.skip("Font may be missing")
+
+    plt.rc('font', family=['DejaVu Sans', 'WenQuanYi Zen Hei'], size=27)
+    plt.rc('pdf', fonttype=3)
+
+    fig = plt.figure()
+    fig.text(0.15, 0.475, "There are 几个汉字 in between!")
+
+
+@image_comparison(["multi_font_type42.pdf"])
+def test_multi_font_type42():
+    fp = fm.FontProperties(family=["WenQuanYi Zen Hei"])
+    if Path(fm.findfont(fp)).name != "wqy-zenhei.ttc":
+        pytest.skip("Font may be missing")
+
+    plt.rc('font', family=['DejaVu Sans', 'WenQuanYi Zen Hei'], size=27)
+    plt.rc('pdf', fonttype=42)
+
+    fig = plt.figure()
+    fig.text(0.15, 0.475, "There are 几个汉字 in between!")