BUG: Fix incorrect tm_matrix in call to visitor_text (py-pdf#2060)

Supply the old tm_matrix when flushing out `text` to the `visitor_text`
in `crlf_space_check`. The new one might already be changed and
unrelated to the current text.

Also add a test for the tm_matrix and cm_matrix that are given to
`visitor_text` when extracting text.
The test computes the coordinates of three letters in different
parts of a test page based on the matrices and checks, if they are
roughly where they should be.

Fixes py-pdf#2059

pypdf/_page.py

@@ -1936,7 +1936,7 @@ def _extract_text(
             1.0,
             0.0,
             0.0,
-        ]  # will store cm_matrix * tm_matrix
+        ]  # will store previous tm_matrix
         char_scale = 1.0
         space_scale = 1.0
         _space_width: float = 500.0  # will be set correctly at first Tf

pypdf/_text_extraction/__init__.py

@@ -99,13 +99,13 @@ def crlf_space_check(
     visitor_text: Optional[Callable[[Any, Any, Any, Any, Any], None]],
     spacewidth: float,
 ) -> Tuple[str, str, List[float]]:
+    m_prev = mult(tm_prev, cm_matrix)
     m = mult(tm_matrix, cm_matrix)
     orientation = orient(m)
-    delta_x = m[4] - tm_prev[4]
-    delta_y = m[5] - tm_prev[5]
+    delta_x = m[4] - m_prev[4]
+    delta_y = m[5] - m_prev[5]
     k = math.sqrt(abs(m[0] * m[3]) + abs(m[1] * m[2]))
     f = font_size * k
-    tm_prev = m
     if orientation not in orientations:
         raise OrientationNotFoundError
     try:
@@ -117,7 +117,7 @@ def crlf_space_check(
                         visitor_text(
                             text + "\n",
                             cm_matrix,
-                            tm_matrix,
+                            tm_prev,
                             cmap[3],
                             font_size,
                         )
@@ -136,7 +136,7 @@ def crlf_space_check(
                         visitor_text(
                             text + "\n",
                             cm_matrix,
-                            tm_matrix,
+                            tm_prev,
                             cmap[3],
                             font_size,
                         )
@@ -155,7 +155,7 @@ def crlf_space_check(
                         visitor_text(
                             text + "\n",
                             cm_matrix,
-                            tm_matrix,
+                            tm_prev,
                             cmap[3],
                             font_size,
                         )
@@ -174,7 +174,7 @@ def crlf_space_check(
                         visitor_text(
                             text + "\n",
                             cm_matrix,
-                            tm_matrix,
+                            tm_prev,
                             cmap[3],
                             font_size,
                         )
@@ -187,6 +187,7 @@ def crlf_space_check(
                 text += " "
     except Exception:
         pass
+    tm_prev = tm_matrix.copy()
     return text, output, tm_prev
 
 

tests/test_text_extraction.py

@@ -55,3 +55,45 @@ def test_multi_language(visitor_text):
     assert "حَبيبي" in reader.pages[2].extract_text(visitor_text=visitor_text)
     assert "habibi" in reader.pages[3].extract_text(visitor_text=visitor_text)
     assert "حَبيبي" in reader.pages[3].extract_text(visitor_text=visitor_text)
+
+
+@pytest.mark.parametrize(("file_name", "constraints"),
+                         [
+                             ("inkscape-abc.pdf",
+                              {"A": lambda x, y:
+                               0 < x < 94 and
+                               189 < y < 283,  # In upper left
+                               "B": lambda x, y:
+                               94 < x < 189 and
+                               94 < y < 189,   # In the center
+                               "C": lambda x, y:
+                               189 < x < 283 and
+                               0 < y < 94}     # In lower right
+                              )
+                             ])
+def test_visitor_text_matrices(file_name, constraints):
+    """
+    Checks if the matrices given to the visitor_text function when calling
+    `extract_text` on the first page of `file_name` match some given constraints.
+    `constraints` is a dictionary mapping a line of text to a constraint that should
+    evaluate to `True` on its expected x,y-coordinates.
+    """
+    reader = PdfReader(RESOURCE_ROOT / file_name)
+
+    lines = []
+
+    def visitor_text(text, cm, tm, font_dict, font_size) -> None:
+        x = tm[4] * cm[0] + tm[5] * cm[2] + cm[4]  # mult(tm, cm)[4]
+        y = tm[4] * cm[1] + tm[5] * cm[3] + cm[5]  # mult(tm, cm)[5]
+        lines.append({"text": text, "x": x, "y": y})
+
+    reader.pages[0].extract_text(visitor_text=visitor_text)
+
+    for text, constraint in constraints.items():
+        matches = [li for li in lines if li["text"].strip() == text]
+        assert len(matches) <= 1, f"Multiple lines match {text}"
+        assert len(matches) >= 1, f"No lines match {text}"
+
+        x = matches[0]["x"]
+        y = matches[0]["y"]
+        assert constraint(x, y), f'Line "{text}" is wrong at x:{x}, y:{y}'