ENH: support font fallback for Agg renderer

Co-authored-by: Aitik Gupta <aitikgupta@gmail.com>
Co-authored-by: Elliott Sales de Andrade <quantum.analyst@gmail.com>

Author: 2 people

.circleci/config.yml

@@ -55,6 +55,7 @@ commands:
               texlive-latex-recommended \
               texlive-pictures \
               texlive-xetex \
+              ttf-wqy-zenhei \
               graphviz \
               fonts-crosextra-carlito \
               fonts-freefont-otf \

doc/api/ft2font.rst

@@ -0,0 +1,8 @@
+**********************
+``matplotlib.ft2font``
+**********************
+
+.. automodule:: matplotlib.ft2font
+   :members:
+   :undoc-members:
+   :show-inheritance:

doc/api/index.rst

@@ -120,6 +120,7 @@ Alphabetical list of modules:
    figure_api.rst
    font_manager_api.rst
    fontconfig_pattern_api.rst
+   ft2font.rst
    gridspec_api.rst
    hatch_api.rst
    image_api.rst

doc/users/next_whats_new/font_fallback.rst

@@ -0,0 +1,27 @@
+Font Fallback in Agg
+--------------------
+
+It is now possible to specify a list of fonts families and the Agg renderer
+will try them in order to locate a required glyph.
+
+.. plot::
+   :caption: Demonstration of mixed English and Chinese text with font fallback.
+   :alt: The phrase "There are 几个汉字 in between!" rendered in various fonts.
+   :include-source: True
+
+   import matplotlib.pyplot as plt
+
+   text = "There are 几个汉字 in between!"
+
+   plt.rcParams["font.size"] = 20
+   fig = plt.figure(figsize=(4.75, 1.85))
+   fig.text(0.05, 0.85, text, family=["WenQuanYi Zen Hei"])
+   fig.text(0.05, 0.65, text, family=["Noto Sans CJK JP"])
+   fig.text(0.05, 0.45, text, family=["DejaVu Sans", "Noto Sans CJK JP"])
+   fig.text(0.05, 0.25, text, family=["DejaVu Sans", "WenQuanYi Zen Hei"])
+
+   plt.show()
+
+
+This currently only works with the Agg backend, but support for the vector
+backends is planned for Matplotlib 3.7.

lib/matplotlib/backends/backend_agg.py

@@ -31,7 +31,7 @@
 from matplotlib import _api, cbook
 from matplotlib.backend_bases import (
     _Backend, FigureCanvasBase, FigureManagerBase, RendererBase)
-from matplotlib.font_manager import findfont, get_font
+from matplotlib.font_manager import fontManager as _fontManager, get_font
 from matplotlib.ft2font import (LOAD_FORCE_AUTOHINT, LOAD_NO_HINTING,
                                 LOAD_DEFAULT, LOAD_NO_AUTOHINT)
 from matplotlib.mathtext import MathTextParser
@@ -272,7 +272,7 @@ def _prepare_font(self, font_prop):
         """
         Get the `.FT2Font` for *font_prop*, clear its buffer, and set its size.
         """
-        font = get_font(findfont(font_prop))
+        font = get_font(_fontManager._find_fonts_by_props(font_prop))
         font.clear()
         size = font_prop.get_size_in_points()
         font.set_size(size, self.dpi)

lib/matplotlib/font_manager.py

@@ -167,11 +167,6 @@
 ]
 
 
-@lru_cache(64)
-def _cached_realpath(path):
-    return os.path.realpath(path)
-
-
 def get_fontext_synonyms(fontext):
     """
     Return a list of file extensions that are synonyms for
@@ -1354,7 +1349,110 @@ def get_font_names(self):
         """Return the list of available fonts."""
         return list(set([font.name for font in self.ttflist]))
 
-    @lru_cache()
+    def _find_fonts_by_props(self, prop, fontext='ttf', directory=None,
+                             fallback_to_default=True, rebuild_if_missing=True):
+        """
+        Find font families that most closely match the given properties.
+
+        Parameters
+        ----------
+        prop : str or `~matplotlib.font_manager.FontProperties`
+            The font properties to search for. This can be either a
+            `.FontProperties` object or a string defining a
+            `fontconfig patterns`_.
+
+        fontext : {'ttf', 'afm'}, default: 'ttf'
+            The extension of the font file:
+
+            - 'ttf': TrueType and OpenType fonts (.ttf, .ttc, .otf)
+            - 'afm': Adobe Font Metrics (.afm)
+
+        directory : str, optional
+            If given, only search this directory and its subdirectories.
+
+        fallback_to_default : bool
+            If True, will fallback to the default font family (usually
+            "DejaVu Sans" or "Helvetica") if none of the families were found.
+
+        rebuild_if_missing : bool
+            Whether to rebuild the font cache and search again if the first
+            match appears to point to a nonexisting font (i.e., the font cache
+            contains outdated entries).
+
+        Returns
+        -------
+        list[str]
+            The paths of the fonts found
+
+        Notes
+        -----
+        This is an extension/wrapper of the original findfont API, which only
+        returns a single font for given font properties. Instead, this API
+        returns an dict containing multiple fonts and their filepaths
+        which closely match the given font properties.  Since this internally
+        uses the original API, there's no change to the logic of performing the
+        nearest neighbor search.  See `findfont` for more details.
+
+        """
+
+        rc_params = tuple(tuple(rcParams[key]) for key in [
+            "font.serif", "font.sans-serif", "font.cursive", "font.fantasy",
+            "font.monospace"])
+
+        prop = FontProperties._from_any(prop)
+
+        fpaths = []
+        for family in prop.get_family():
+            cprop = prop.copy()
+
+            # set current prop's family
+            cprop.set_family(family)
+
+            # do not fall back to default font
+            try:
+                fpaths.append(
+                    self._findfont_cached(
+                        cprop, fontext, directory,
+                        fallback_to_default=False,
+                        rebuild_if_missing=rebuild_if_missing,
+                        rc_params=rc_params,
+                    )
+                )
+            except ValueError:
+                if family in font_family_aliases:
+                    _log.warning(
+                        "findfont: Generic family %r not found because "
+                        "none of the following families were found: %s",
+                        family,
+                        ", ".join(self._expand_aliases(family))
+                    )
+                else:
+                    _log.warning(
+                        'findfont: Font family \'%s\' not found.', family
+                    )
+
+        # only add default family if no other font was found and
+        # fallback_to_default is enabled
+        if not fpaths:
+            if fallback_to_default:
+                dfamily = self.defaultFamily[fontext]
+                cprop = prop.copy()
+                cprop.set_family(dfamily)
+                fpaths.append(
+                    self._findfont_cached(
+                        cprop, fontext, directory,
+                        fallback_to_default=True,
+                        rebuild_if_missing=rebuild_if_missing,
+                        rc_params=rc_params,
+                    )
+                )
+            else:
+                raise ValueError("Failed to find any font, and fallback "
+                                 "to the default font was disabled.")
+
+        return fpaths
+
+    @lru_cache(1024)
     def _findfont_cached(self, prop, fontext, directory, fallback_to_default,
                          rebuild_if_missing, rc_params):
 
@@ -1447,9 +1545,19 @@ def is_opentype_cff_font(filename):
 
 
 @lru_cache(64)
-def _get_font(filename, hinting_factor, *, _kerning_factor, thread_id):
+def _get_font(font_filepaths, hinting_factor, *, _kerning_factor, thread_id):
+    first_fontpath, *rest = font_filepaths
     return ft2font.FT2Font(
-        filename, hinting_factor, _kerning_factor=_kerning_factor)
+        first_fontpath, hinting_factor,
+        _fallback_list=[
+            ft2font.FT2Font(
+                fpath, hinting_factor,
+                _kerning_factor=_kerning_factor
+            )
+            for fpath in rest
+        ],
+        _kerning_factor=_kerning_factor
+    )
 
 
 # FT2Font objects cannot be used across fork()s because they reference the same
@@ -1461,16 +1569,51 @@ def _get_font(filename, hinting_factor, *, _kerning_factor, thread_id):
     os.register_at_fork(after_in_child=_get_font.cache_clear)
 
 
-def get_font(filename, hinting_factor=None):
+@lru_cache(64)
+def _cached_realpath(path):
     # Resolving the path avoids embedding the font twice in pdf/ps output if a
     # single font is selected using two different relative paths.
-    filename = _cached_realpath(filename)
+    return os.path.realpath(path)
+
+
+@_api.rename_parameter('3.6', "filepath", "font_filepaths")
+def get_font(font_filepaths, hinting_factor=None):
+    """
+    Get an `.ft2font.FT2Font` object given a list of file paths.
+
+    Parameters
+    ----------
+    font_filepaths : Iterable[str, Path, bytes], str, Path, bytes
+        Relative or absolute paths to the font files to be used.
+
+        If a single string, bytes, or `pathlib.Path`, then it will be treated
+        as a list with that entry only.
+
+        If more than one filepath is passed, then the returned FT2Font object
+        will fall back through the fonts, in the order given, to find a needed
+        glyph.
+
+    Returns
+    -------
+    `.ft2font.FT2Font`
+
+    """
+    if isinstance(font_filepaths, (str, Path, bytes)):
+        paths = (_cached_realpath(font_filepaths),)
+    else:
+        paths = tuple(_cached_realpath(fname) for fname in font_filepaths)
+
     if hinting_factor is None:
         hinting_factor = rcParams['text.hinting_factor']
-    # also key on the thread ID to prevent segfaults with multi-threading
-    return _get_font(filename, hinting_factor,
-                     _kerning_factor=rcParams['text.kerning_factor'],
-                     thread_id=threading.get_ident())
+
+    return _get_font(
+        # must be a tuple to be cached
+        paths,
+        hinting_factor,
+        _kerning_factor=rcParams['text.kerning_factor'],
+        # also key on the thread ID to prevent segfaults with multi-threading
+        thread_id=threading.get_ident()
+    )
 
 
 def _load_fontmanager(*, try_read_cache=True):

lib/matplotlib/tests/test_ft2font.py

@@ -0,0 +1,78 @@
+from pathlib import Path
+import io
+
+import pytest
+
+from matplotlib import ft2font
+from matplotlib.testing.decorators import check_figures_equal
+import matplotlib.font_manager as fm
+import matplotlib.pyplot as plt
+
+
+def test_fallback_errors():
+    file_name = fm.findfont('DejaVu Sans')
+
+    with pytest.raises(TypeError, match="Fallback list must be a list"):
+        # failing to be a list will fail before the 0
+        ft2font.FT2Font(file_name, _fallback_list=(0,))
+
+    with pytest.raises(
+            TypeError, match="Fallback fonts must be FT2Font objects."
+    ):
+        ft2font.FT2Font(file_name, _fallback_list=[0])
+
+
+def test_ft2font_positive_hinting_factor():
+    file_name = fm.findfont('DejaVu Sans')
+    with pytest.raises(
+            ValueError, match="hinting_factor must be greater than 0"
+    ):
+        ft2font.FT2Font(file_name, 0)
+
+
+def test_fallback_smoke():
+    fp = fm.FontProperties(family=["WenQuanYi Zen Hei"])
+    if Path(fm.findfont(fp)).name != "wqy-zenhei.ttc":
+        pytest.skip("Font wqy-zenhei.ttc may be missing")
+
+    fp = fm.FontProperties(family=["Noto Sans CJK JP"])
+    if Path(fm.findfont(fp)).name != "NotoSansCJK-Regular.ttc":
+        pytest.skip("Noto Sans CJK JP font may be missing.")
+
+    plt.rcParams['font.size'] = 20
+    fig = plt.figure(figsize=(4.75, 1.85))
+    fig.text(0.05, 0.45, "There are 几个汉字 in between!",
+             family=['DejaVu Sans', "Noto Sans CJK JP"])
+    fig.text(0.05, 0.25, "There are 几个汉字 in between!",
+             family=['DejaVu Sans', "WenQuanYi Zen Hei"])
+    fig.text(0.05, 0.65, "There are 几个汉字 in between!",
+             family=["Noto Sans CJK JP"])
+    fig.text(0.05, 0.85, "There are 几个汉字 in between!",
+             family=["WenQuanYi Zen Hei"])
+
+    # TODO enable fallback for other backends!
+    for fmt in ['png', 'raw']:  # ["svg", "pdf", "ps"]:
+        fig.savefig(io.BytesIO(), format=fmt)
+
+
+@pytest.mark.parametrize('family_name, file_name',
+                         [("WenQuanYi Zen Hei",  "wqy-zenhei.ttc"),
+                          ("Noto Sans CJK JP", "NotoSansCJK-Regular.ttc")]
+                         )
+@check_figures_equal(extensions=["png"])
+def test_font_fallback_chinese(fig_test, fig_ref, family_name, file_name):
+    fp = fm.FontProperties(family=[family_name])
+    if Path(fm.findfont(fp)).name != file_name:
+        pytest.skip(f"Font {family_name} ({file_name}) is missing")
+
+    text = ["There are", "几个汉字", "in between!"]
+
+    plt.rcParams["font.size"] = 20
+    test_fonts = [["DejaVu Sans", family_name]] * 3
+    ref_fonts = [["DejaVu Sans"], [family_name], ["DejaVu Sans"]]
+
+    for j, (txt, test_font, ref_font) in enumerate(
+            zip(text, test_fonts, ref_fonts)
+    ):
+        fig_ref.text(0.05, .85 - 0.15*j, txt, family=ref_font)
+        fig_test.text(0.05, .85 - 0.15*j, txt, family=test_font)