Reducing fonttools memory usage by 10MiB by passing lazy=True to TTFo…

…nt constructor (#718)

CHANGELOG.md

@@ -28,6 +28,7 @@ This can also be enabled programmatically with `warnings.simplefilter('default',
 - hyperlinks were not working on encrypted files - thanks to @andersonhc
 - unicode (non limited to ASCII) text can now be provided as metadata [#685](https://github.com/PyFPDF/fpdf2/issues/685)
 - all `TitleStyle` constructor parameters are now effectively optional
+- memory usage was reduced by 10 MiB in some cases, thanks to a small optimization in using `fonttools`
 ### Changed
 - vector images parsing is now more robust: `fpdf2` can now embed SVG files without `viewPort` or no `height` / `width`
 - bitonal images are now encoded using `CCITTFaxDecode`, reducing their size in the PDF document - thanks to @eroux

fpdf/fpdf.py

@@ -1798,7 +1798,7 @@ def add_font(self, family=None, style="", fname=None, uni="DEPRECATED"):
         if fontkey in self.fonts or fontkey in self.core_fonts:
             warnings.warn(f"Core font or font already added '{fontkey}': doing nothing")
             return
-        font = ttLib.TTFont(font_file_path, fontNumber=0)
+        font = ttLib.TTFont(font_file_path, fontNumber=0, lazy=True)
 
         scale = 1000 / font["head"].unitsPerEm
         default_width = round(scale * font["hmtx"].metrics[".notdef"][0])

fpdf/output.py

@@ -578,7 +578,7 @@ def _add_fonts(self):
                 # recalcTimestamp=False means that it doesn't modify the "modified" timestamp in head table
                 # if we leave recalcTimestamp=True the tests will break every time
                 fonttools_font = ttLib.TTFont(
-                    file=font["ttffile"], recalcTimestamp=False, fontNumber=0
+                    file=font["ttffile"], recalcTimestamp=False, fontNumber=0, lazy=True
                 )
 
                 # 1. get all glyphs in PDF

fpdf/svg.py

@@ -635,7 +635,8 @@ def from_file(cls, filename, *args, encoding="utf-8", **kwargs):
     def __init__(self, svg_text):
         self.cross_references = {}
 
-        svg_tree = parse_xml_str(svg_text)
+        # disabling bandit rule as we use defusedxml:
+        svg_tree = parse_xml_str(svg_text)  # nosec B314
 
         if svg_tree.tag not in xmlns_lookup("svg", "svg"):
             raise ValueError(f"root tag must be svg, not {svg_tree.tag}")

test/end_to_end_legacy/charmap/test_charmap.py → test/fonts/test_charmap.py

@@ -11,36 +11,45 @@
 """
 from pathlib import Path
 
+from fontTools.ttLib import TTFont
 import pytest
 
-import fpdf
-from fontTools import ttLib
+from fpdf import FPDF
 from test.conftest import assert_pdf_equal
 
 HERE = Path(__file__).resolve().parent
 
 
 @pytest.mark.parametrize(
     "font_filename",
-    ["DejaVuSans.ttf", "DroidSansFallback.ttf", "Roboto-Regular.ttf", "cmss12.ttf"],
+    [
+        font_file.name
+        for font_file in HERE.glob("*.*tf")
+        if not any(
+            exclude in font_file.stem
+            for exclude in ("Bold", "Italic", "NotoColorEmoji")
+        )
+    ],
 )
-def test_first_999_chars(font_filename, tmp_path):
-    font_path = HERE / ".." / ".." / "fonts" / font_filename
+def test_charmap_first_999_chars(font_filename, tmp_path):
+    """
+    Character Map Test
+    from PyFPDF version 1.7.2: github.com/reingart/pyfpdf/commit/2eab310cfd866ce24947c3a9d850ebda7c6d515d
+    """
+    font_path = HERE / font_filename
     font_name = font_path.stem
 
-    pdf = fpdf.FPDF()
+    pdf = FPDF()
     pdf.add_page()
-    pdf.add_font(font_name, fname=font_path)
+    pdf.add_font(fname=font_path)
     pdf.set_font(font_name, size=10)
 
-    font = ttLib.TTFont(font_path)
+    font = TTFont(font_path, lazy=True)
     cmap = font.getBestCmap()
 
-    # Create a PDF with the first 999 charters defined in the font:
-    for counter, character in enumerate(cmap, 0):
+    # Create a PDF with the first 999 characters defined in the font:
+    for counter, character in enumerate(list(cmap.keys())[:1000]):
         pdf.write(8, f"{counter:03}) {character:03x} - {character:c}", print_sh=True)
         pdf.ln()
-        if counter >= 999:
-            break
 
     assert_pdf_equal(pdf, HERE / f"charmap_first_999_chars-{font_name}.pdf", tmp_path)

test/test_perfs.py

@@ -1,15 +1,17 @@
 from pathlib import Path
 
-import memunit, pytest
+import pytest
 
 from fpdf import FPDF
 
 HERE = Path(__file__).resolve().parent
 
 
 @pytest.mark.timeout(40)
-# ensure memory usage does not get too high - this value depends on Python version:
-@memunit.assert_lt_mb(185)
+# Note: this does not combine well with memunit.
+# memory_profiler.MemTimer does not terminate properly when a signal is raised by pytest-timeout,
+# and as a consequence multiprocessing.util._exit_function becomes blocking at the end of Pytest execution,
+# (on line "calling join() for process MemTimer-1")
 def test_intense_image_rendering():
     png_file_paths = []
     for png_file_path in (HERE / "image/png_images/").glob("*.png"):