fix: Guard against attribute errors in TesseractOcrModel __del__ (#1494)

This moves the initialization of the `reader` and `script_readers`
attributes to before we attempt to import tesserocr, so that when later
accessing these attributes in the garbage collection method `__del__`
the attributes exist.

This requires changing the typing of the `script_readers` dict value to
`Any` because we cannot yet reference its actual strong type, since it's
a tesserocr value.

This prevents throwing an exception during garbage collection for
cases where the TesseractOcrModel instance didn't properly initialize,
like when it throws an `ImportError` during its initializer.

Signed-off-by: Ben Browning <bbrownin@redhat.com>

Author: bbrowning

docling/models/tesseract_ocr_model.py

@@ -1,7 +1,7 @@
 import logging
 from collections.abc import Iterable
 from pathlib import Path
-from typing import Optional, Type
+from typing import Any, Optional, Type
 
 from docling_core.types.doc import BoundingBox, CoordOrigin
 from docling_core.types.doc.page import BoundingRectangle, TextCell
@@ -38,6 +38,8 @@ def __init__(
         self.options: TesseractOcrOptions
 
         self.scale = 3  # multiplier for 72 dpi == 216 dpi.
+        self.reader = None
+        self.script_readers: dict[str, Any] = {}
 
         if self.enabled:
             install_errmsg = (
@@ -84,9 +86,7 @@ def __init__(
                 "oem": tesserocr.OEM.DEFAULT,
             }
 
-            self.reader = None
             self.osd_reader = None
-            self.script_readers: dict[str, tesserocr.PyTessBaseAPI] = {}
 
             if self.options.path is not None:
                 tesserocr_kwargs["path"] = self.options.path