Avoid int overflows when converting units from/to double

Fix #2262.

tests/draw/test_text.py

@@ -875,3 +875,28 @@ def test_otb_font(assert_pixels):
         }
       </style>
       AaA''')
+
+
+def test_huge_justification(assert_pixels):
+    # Test regression: https://github.com/Kozea/WeasyPrint/issues/2262
+    assert_pixels('''
+        ____
+        _RR_
+        _RR_
+        ____
+    ''', '''
+      <style>
+        @page {
+          size: 4px 4px;
+          margin: 1px;
+        }
+        body {
+          color: red;
+          font-family: weasyprint;
+          font-size: 2px;
+          line-height: 1;
+          text-align: justify-all;
+          width: 100000px;
+        }
+      </style>
+      A B''')

weasyprint/css/computed_values.py

@@ -6,7 +6,7 @@
 from tinycss2.color3 import parse_color
 
 from ..logger import LOGGER
-from ..text.ffi import ffi, pango, units_to_double
+from ..text.ffi import FROM_UNITS, ffi, pango
 from ..text.line_break import Layout, first_line_metrics
 from ..urls import get_link_attribute
 from .properties import INITIAL_VALUES, ZERO_PIXELS, Dimension
@@ -815,9 +815,9 @@ def character_ratio(style, character):
     logical_extents = ffi.new('PangoRectangle *')
     pango.pango_layout_line_get_extents(line, ink_extents, logical_extents)
     if character == 'x':
-        measure = -units_to_double(ink_extents.y)
+        measure = -ink_extents.y * FROM_UNITS
     else:
-        measure = units_to_double(logical_extents.width)
+        measure = logical_extents.width * FROM_UNITS
     ffi.release(ink_extents)
     ffi.release(logical_extents)
 

weasyprint/draw/text.py

@@ -7,7 +7,7 @@
 
 from ..images import RasterImage, SVGImage
 from ..matrix import Matrix
-from ..text.ffi import ffi, pango, units_from_double, units_to_double
+from ..text.ffi import FROM_UNITS, TO_UNITS, ffi, pango
 from ..text.fonts import get_hb_object_data
 from ..text.line_break import get_last_word_end
 from .border import draw_line
@@ -89,7 +89,7 @@ def draw_first_line(stream, textbox, text_overflow, block_ellipsis, matrix):
         assert textbox.pango_layout.max_width is not None
         max_width = textbox.pango_layout.max_width
         pango.pango_layout_set_width(
-            textbox.pango_layout.layout, units_from_double(max_width))
+            textbox.pango_layout.layout, int(max_width * TO_UNITS))
         if text_overflow == 'ellipsis':
             pango.pango_layout_set_ellipsize(
                 textbox.pango_layout.layout, pango.PANGO_ELLIPSIZE_END)
@@ -194,12 +194,11 @@ def draw_first_line(stream, textbox, text_overflow, block_ellipsis, matrix):
                 pango.pango_font_get_glyph_extents(
                     pango_font, glyph, stream.ink_rect, stream.logical_rect)
                 font.widths[glyph] = int(round(
-                    units_to_double(stream.logical_rect.width * 1000) /
-                    font_size))
+                    stream.logical_rect.width * 1000 * FROM_UNITS / font_size))
 
             # Set kerning, word spacing, letter spacing.
             kerning = int(
-                font.widths[glyph] - units_to_double(width * 1000) / font_size + offset)
+                font.widths[glyph] + offset - width * 1000 * FROM_UNITS / font_size)
             if kerning:
                 string += f'>{kerning}<'
 
@@ -237,9 +236,8 @@ def draw_first_line(stream, textbox, text_overflow, block_ellipsis, matrix):
                     pango.pango_font_get_glyph_extents(
                         pango_font, glyph, stream.ink_rect,
                         stream.logical_rect)
-                    f = units_to_double(
-                        (-stream.logical_rect.y - stream.logical_rect.height))
-                    f = f / font_size - font_size
+                    f = -stream.logical_rect.y - stream.logical_rect.height
+                    f = f * FROM_UNITS / font_size - font_size
                     emojis.append([image, font, a, d, x_advance, f])
 
             x_advance += (font.widths[glyph] + offset - kerning) / 1000

weasyprint/pdf/fonts.py

@@ -12,7 +12,7 @@
 
 from ..logger import LOGGER, capture_logs
 from ..text.constants import PANGO_STRETCH_PERCENT
-from ..text.ffi import ffi, harfbuzz, harfbuzz_subset, pango, units_to_double
+from ..text.ffi import FROM_UNITS, ffi, harfbuzz, harfbuzz_subset, pango
 from ..text.fonts import get_hb_object_data, get_pango_font_hb_face
 
 
@@ -121,7 +121,7 @@ def clean(self, cmap, hinting):
                     self.description)
                 self.variations['wght'] = weight
             if 'opsz' not in self.variations:
-                self.variations['opsz'] = units_to_double(self.font_size)
+                self.variations['opsz'] = self.font_size * FROM_UNITS
             if 'slnt' not in self.variations:
                 slnt = 0
                 if self.style == 1:

weasyprint/text/ffi.py

@@ -463,8 +463,9 @@ def _dlopen(ffi, *names, allow_fail=False):
 
 gobject.g_type_init()
 
-units_to_double = pango.pango_units_to_double
-units_from_double = pango.pango_units_from_double
+# Call once to avoid int overflows.
+TO_UNITS = pango.pango_units_from_double(1)
+FROM_UNITS = pango.pango_units_to_double(1)
 
 
 def unicode_to_char_p(string):

weasyprint/text/fonts.py

@@ -17,8 +17,7 @@
     CAPS_KEYS, EAST_ASIAN_KEYS, FONTCONFIG_STRETCH, FONTCONFIG_STYLE,
     FONTCONFIG_WEIGHT, LIGATURE_KEYS, NUMERIC_KEYS, PANGO_STRETCH, PANGO_STYLE)
 from .ffi import (  # isort:skip
-    ffi, fontconfig, gobject, harfbuzz, pango, pangoft2, unicode_to_char_p,
-    units_from_double)
+    TO_UNITS, ffi, fontconfig, gobject, harfbuzz, pango, pangoft2, unicode_to_char_p)
 
 
 def _check_font_configuration(font_config):  # pragma: no cover
@@ -317,7 +316,7 @@ def get_font_description(style):
     pango.pango_font_description_set_stretch(font_description, font_stretch)
     font_weight = style['font_weight']
     pango.pango_font_description_set_weight(font_description, font_weight)
-    font_size = units_from_double(style['font_size'])
+    font_size = int(style['font_size'] * TO_UNITS)
     pango.pango_font_description_set_absolute_size(font_description, font_size)
     if style['font_variation_settings'] != 'normal':
         string = ','.join(

weasyprint/text/line_break.py

@@ -6,12 +6,9 @@
 import pyphen
 
 from .constants import LST_TO_ISO, PANGO_WRAP_MODE
+from .ffi import FROM_UNITS, TO_UNITS, ffi, gobject, pango, pangoft2, unicode_to_char_p
 from .fonts import font_features, get_font_description
 
-from .ffi import (  # isort:skip
-    ffi, gobject, pango, pangoft2, unicode_to_char_p, units_from_double,
-    units_to_double)
-
 
 def line_size(line, style):
     """Get logical width and height of the given ``line``.
@@ -21,8 +18,8 @@ def line_size(line, style):
     """
     logical_extents = ffi.new('PangoRectangle *')
     pango.pango_layout_line_get_extents(line, ffi.NULL, logical_extents)
-    width = units_to_double(logical_extents.width)
-    height = units_to_double(logical_extents.height)
+    width = logical_extents.width * FROM_UNITS
+    height = logical_extents.height * FROM_UNITS
     ffi.release(logical_extents)
     if style['letter_spacing'] != 'normal':
         width += style['letter_spacing']
@@ -52,7 +49,7 @@ def first_line_metrics(first_line, text, layout, resume_at, space_collapse,
         length = first_line.length if first_line is not None else 0
 
     width, height = line_size(first_line, style)
-    baseline = units_to_double(pango.pango_layout_get_baseline(layout.layout))
+    baseline = pango.pango_layout_get_baseline(layout.layout) * FROM_UNITS
     layout.deactivate()
     return layout, length, resume_at, width, height, baseline
 
@@ -107,15 +104,15 @@ def setup(self, context, style):
                 pango.pango_context_get_metrics(
                     pango_context, font_description, self.language),
                 pango.pango_font_metrics_unref)
-            self.ascent = units_to_double(
+            self.ascent = FROM_UNITS * (
                 pango.pango_font_metrics_get_ascent(metrics))
-            self.underline_position = units_to_double(
+            self.underline_position = FROM_UNITS * (
                 pango.pango_font_metrics_get_underline_position(metrics))
-            self.strikethrough_position = units_to_double(
+            self.strikethrough_position = FROM_UNITS * (
                 pango.pango_font_metrics_get_strikethrough_position(metrics))
-            self.underline_thickness = units_to_double(
+            self.underline_thickness = FROM_UNITS * (
                 pango.pango_font_metrics_get_underline_thickness(metrics))
-            self.strikethrough_thickness = units_to_double(
+            self.strikethrough_thickness = FROM_UNITS * (
                 pango.pango_font_metrics_get_strikethrough_thickness(metrics))
         else:
             self.ascent = None
@@ -179,7 +176,7 @@ def add_attr(start, end, spacing):
                 pango.pango_attr_list_change(attr_list, attr)
 
             if letter_spacing:
-                letter_spacing = units_from_double(letter_spacing)
+                letter_spacing = int(letter_spacing * TO_UNITS)
                 add_attr(0, len(bytestring), letter_spacing)
 
             if word_spacing:
@@ -189,8 +186,7 @@ def add_attr(start, end, spacing):
                     text, bytestring = unicode_to_char_p(self.text)
                     pango.pango_layout_set_text(self.layout, text, -1)
 
-                space_spacing = (
-                    units_from_double(word_spacing) + letter_spacing)
+                space_spacing = int(word_spacing * TO_UNITS + letter_spacing)
                 position = bytestring.find(b' ')
                 # Pango gives only half of word-spacing on boundaries
                 boundary_positions = (0, len(bytestring) - 1)
@@ -243,8 +239,7 @@ def create_layout(text, style, context, max_width, justification_spacing):
     # signed integer. Treat bigger values same as None: unconstrained width.
     text_wrap = style['white_space'] in ('normal', 'pre-wrap', 'pre-line')
     if max_width is not None and text_wrap and max_width < 2 ** 21:
-        pango.pango_layout_set_width(
-            layout.layout, units_from_double(max(0, max_width)))
+        pango.pango_layout_set_width(layout.layout, int(max(0, max_width) * TO_UNITS))
 
     layout.set_text(text)
     return layout
@@ -473,10 +468,8 @@ def split_first_line(text, style, context, max_width, justification_spacing,
         # instance) from keeping their shape when wrapped on the next line with
         # pango layout. Maybe insert Unicode shaping characters in text?
         layout.set_text(text)
-        pango.pango_layout_set_width(
-            layout.layout, units_from_double(max_width))
-        pango.pango_layout_set_wrap(
-            layout.layout, PANGO_WRAP_MODE['WRAP_CHAR'])
+        pango.pango_layout_set_width(layout.layout, int(max_width * TO_UNITS))
+        pango.pango_layout_set_wrap(layout.layout, PANGO_WRAP_MODE['WRAP_CHAR'])
         first_line, index = layout.get_first_line()
         resume_index = index or first_line.length
         if resume_index >= len(text.encode()):