Fix handling of nonmath hyphens in mathtext.

hyphen-to-unicode-minus replacement now occurs at the parsing stage, and
thus does not need to be done again in get_unicode_index (otherwise,
text hyphens also get incorrectly substituted.

Strictly speaking, one would need either to fully deprecate
get_unicode_index and replace it by a private helper (for which `math`
is always False), or introduce a deprecation period during which users
*must* pass the `math` kwarg (any value other than False results in a
warning) and then another one that deprecates that kwarg; which is quite
long; hopefully the change is in a private-enough part of the library
that we can just do it as is.

Also do some more changes to the mathtext-svg-text-only-tests that I
missed when the feature first went in: only test svg (that's the
point of it...) and only test two fonts: a "standard" truetype font --
the default dejavusans -- and the "cm" font, which has a very specific
encoding (see the BakomaFonts class).

Author: anntzer

doc/api/next_api_changes/behavior/23270-AL.rst

@@ -0,0 +1,5 @@
+The *math* parameter of ``mathtext.get_unicode_index`` is deprecated and defaults to False
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+In math mode, ASCII hyphens (U+002D) are now replaced by unicode minus signs
+(U+2212) at the parsing stage.

doc/api/next_api_changes/deprecations/22507-AL.rst

@@ -34,7 +34,7 @@
 
 
 @_api.delete_parameter("3.6", "math")
-def get_unicode_index(symbol, math=True):  # Publicly exported.
+def get_unicode_index(symbol, math=False):  # Publicly exported.
     r"""
     Return the integer index (from the Unicode table) of *symbol*.
 
@@ -43,8 +43,8 @@ def get_unicode_index(symbol, math=True):  # Publicly exported.
     symbol : str
         A single (Unicode) character, a TeX command (e.g. r'\pi') or a Type1
         symbol name (e.g. 'phi').
-    math : bool, default: True
-        If False, always treat as a single Unicode character.
+    math : bool, default: False
+        If True (deprecated), replace ASCII hyphen-minus by Unicode minus.
     """
     # From UTF #25: U+2212 minus sign is the preferred
     # representation of the unary and binary minus sign rather than

lib/matplotlib/_mathtext.py

@@ -119,6 +119,7 @@
 # 'svgastext' tests switch svg output to embed text as text (rather than as
 # paths).
 svgastext_math_tests = [
+    r'$-$-',
 ]
 # 'lightweight' tests test only a single fontset (dejavusans, which is the
 # default) and only png outputs, in order to minimize the size of baseline
@@ -206,11 +207,10 @@ def test_mathtext_rendering(baseline_images, fontset, index, text):
 
 @pytest.mark.parametrize('index, text', enumerate(svgastext_math_tests),
                          ids=range(len(svgastext_math_tests)))
-@pytest.mark.parametrize(
-    'fontset', ['cm', 'stix', 'stixsans', 'dejavusans', 'dejavuserif'])
+@pytest.mark.parametrize('fontset', ['cm', 'dejavusans'])
 @pytest.mark.parametrize('baseline_images', ['mathtext0'], indirect=True)
 @image_comparison(
-    baseline_images=None,
+    baseline_images=None, extensions=['svg'],
     savefig_kwarg={'metadata': {  # Minimize image size.
         'Creator': None, 'Date': None, 'Format': None, 'Type': None}})
 def test_mathtext_rendering_svgastext(baseline_images, fontset, index, text):