Performance improvements

Author: VikParuchuri

pdftext/pdf/chars.py

@@ -85,8 +85,13 @@ def word_break():
             word_break()
             continue
 
-        # we break on any change in font info
-        if any(char['font'][k] != word['font'][k] for k in ['name', 'flags', 'size', 'weight']):
+        # we break on any change in font info - optimized comparison
+        char_font = char['font']
+        word_font = word['font']
+        if (char_font['name'] != word_font['name'] or 
+            char_font['flags'] != word_font['flags'] or 
+            char_font['size'] != word_font['size'] or 
+            char_font['weight'] != word_font['weight']):
             word_break()
             continue
 
@@ -99,17 +104,19 @@ def word_break():
         word['bbox'] = word['bbox'].merge(char['bbox'])
         word['chars'].append(char)
 
-    # deduplicate words
-    seen = {}
+    # deduplicate words - use tuple keys instead of strings
+    seen = set()
     deduped = []
     for word in words:
         # Round the bbox coordinates
         bbox = word['bbox'].bbox
-        bbox = [round(x, 0) for x in bbox]
+        bbox_rounded = tuple(round(x, 0) for x in bbox)
 
-        key = f"{bbox}-{word['text']}-{word['rotation']}-{word['font']['name']}-{word['font']['flags']}-{word['font']['size']}-{word['font']['weight']}"
+        key = (bbox_rounded, word['text'], word['rotation'], 
+               word['font']['name'], word['font']['flags'], 
+               word['font']['size'], word['font']['weight'])
         if key not in seen:
-            seen[key] = True
+            seen.add(key)
             deduped.append(word)
 
     return [char for word in deduped for char in word['chars']]

pdftext/pdf/pages.py

@@ -132,8 +132,11 @@ def span_break():
 def get_lines(spans: Spans) -> Lines:
     lines: Lines = []
     line: Line = None
+    line_text: str = ""
 
     def line_break():
+        global line_text
+        line_text = ""
         lines.append({"spans": [span], "bbox": span["bbox"], "rotation": span["rotation"]})
 
     for span in spans:
@@ -144,8 +147,13 @@ def line_break():
             line_break()
             continue
 
-        # we break if the previous span ends with a linebreak or hyphenation
-        if any(line["spans"][-1]["text"].endswith(suffix) for suffix in ["\n", "\x02"]):
+        # we break if the previous span ends with a linebreak, and the line has text
+        if line["spans"][-1]["text"].endswith("\n") and line_text.strip():
+            line_break()
+            continue
+
+        # we break if the current line ends with a hyphen
+        if line["spans"][-1]["text"].endswith("\x02"):
             line_break()
             continue
 
@@ -159,6 +167,7 @@ def line_break():
             continue
 
         line["spans"].append(span)
+        line_text += span["text"]
         line["bbox"] = line["bbox"].merge(span["bbox"])
 
     return lines

pdftext/schema.py

@@ -56,12 +56,14 @@ def y_end(self):
         return self.bbox[3]
 
     def merge(self, other: Bbox) -> Bbox:
-        x_start = self.x_start if self.x_start < other.x_start else other.x_start
-        y_start = self.y_start if self.y_start < other.y_start else other.y_start
-        x_end = self.x_end if self.x_end > other.x_end else other.x_end
-        y_end = self.y_end if self.y_end > other.y_end else other.y_end
-
-        return Bbox([x_start, y_start, x_end, y_end])
+        self_bbox = self.bbox
+        other_bbox = other.bbox
+        return Bbox([
+            min(self_bbox[0], other_bbox[0]),
+            min(self_bbox[1], other_bbox[1]),
+            max(self_bbox[2], other_bbox[2]),
+            max(self_bbox[3], other_bbox[3])
+        ])
 
     def overlap_x(self, other: Bbox):
         return max(0, min(self.bbox[2], other.bbox[2]) - max(self.bbox[0], other.bbox[0]))