Improve performance working with unicode fonts (#907)

CHANGELOG.md

@@ -41,6 +41,7 @@ This can also be enabled programmatically with `warnings.simplefilter('default',
 ### Changed
 * [`FPDF.table()`](https://py-pdf.github.io/fpdf2/Tables.html) now raises an error when a single row is too high to be rendered on a single page
 * [`FPDF.write_html()`](https://py-pdf.github.io/fpdf2/fpdf/fpdf.html#fpdf.fpdf.FPDF.write_html): `tag_indents` can now be non-integer. Indentation of HTML elements is now independent of font size and bullet strings.
+* improved performance of font glyph selection by using functools cache
 
 ## [2.7.9] - 2024-05-17
 ### Added

fpdf/fonts.py

@@ -12,6 +12,7 @@
 from bisect import bisect_left
 from collections import defaultdict
 from dataclasses import dataclass, replace
+from functools import lru_cache
 from typing import List, Optional, Tuple, Union
 
 from fontTools import ttLib
@@ -447,8 +448,6 @@ def __init__(self, font: TTFFont, identities: List[int]):
             glyph = self.get_glyph(unicode=x)
             if glyph:
                 self._char_id_per_glyph[glyph] = int(x)
-        # This is a cache to speed things up:
-        self._char_id_per_unicode = {}
 
     def __repr__(self):
         return (
@@ -463,10 +462,9 @@ def items(self):
         for glyph, char_id in self._char_id_per_glyph.items():
             yield glyph, char_id
 
+    # pylint: disable=method-cache-max-size-none
+    @lru_cache(maxsize=None)
     def pick(self, unicode: int):
-        cache_hit = self._char_id_per_unicode.get(unicode)
-        if cache_hit:
-            return cache_hit
         glyph = self.get_glyph(unicode=unicode)
         if glyph is None and unicode not in self.font.missing_glyphs:
             self.font.missing_glyphs.append(unicode)
@@ -482,10 +480,10 @@ def pick_glyph(self, glyph):
             char_id = self._next
             self._char_id_per_glyph[glyph] = char_id
             self._next += 1
-            # Fill cache:
-            self._char_id_per_unicode[glyph.unicode] = char_id
         return char_id
 
+    # pylint: disable=method-cache-max-size-none
+    @lru_cache(maxsize=None)
     def get_glyph(
         self, glyph=None, unicode=None, glyph_name=None, glyph_width=None
     ) -> Glyph:

fpdf/fpdf.py

@@ -3437,7 +3437,7 @@ def get_fallback_font(self, char, style=""):
 
     def _parse_chars(self, text: str, markdown: bool) -> Iterator[Fragment]:
         "Split text into fragments"
-        if not markdown and not self.is_ttf_font:
+        if not markdown and not self.text_shaping and not self._fallback_font_ids:
             yield Fragment(text, self._get_current_graphics_state(), self.k)
             return
         txt_frag, in_bold, in_italics, in_underline = (

fpdf/output.py

@@ -686,6 +686,8 @@ def format_code(unicode):
                 self._add_pdf_obj(font_file_cs_obj, "fonts")
                 font_descriptor_obj.font_file2 = font_file_cs_obj
 
+                font.subset.pick.cache_clear()
+                font.subset.get_glyph.cache_clear()
                 font.close()
 
         return font_objs_per_index

test/text/test_cell.py

@@ -325,7 +325,7 @@ def test_cell_deprecated_txt_arg():
         pdf.cell(txt="Lorem ipsum Ut nostrud irure")
 
 
-@ensure_exec_time_below(seconds=24)
+@ensure_exec_time_below(seconds=18)
 @ensure_rss_memory_below(mib=1)
 def test_cell_speed_with_long_text():  # issue #907
     pdf = FPDF()

test/text_shaping/test_bidirectional.py

@@ -195,6 +195,7 @@ def test_bidi_paragraph_direction(tmp_path):
     pdf.add_page()
     pdf.add_font(family="SBL_Hbrw", fname=HERE / "SBL_Hbrw.ttf")
     pdf.set_font("SBL_Hbrw", "", 18)
+    pdf.set_text_shaping(True)
 
     text1 = "אנגלית (באנגלית: English)"  # first char is RTL
     text2 = "The test is: אנגלית (באנגלית: English)"  # first char is LTR