If the `python-levenshtein` 3rd-party package is installed, it will improve the calculation time. refs #1426.

CHANGES

@@ -22,6 +22,8 @@ Incompatible changes
   method to make the `any` role work properly.
 * gettext builder: gettext doesn't emit uuid information to generated pot files
   by default. Please set ``True`` to `gettext_uuid` to emit uuid information.
+  Additionally, if the ``python-levenshtein`` 3rd-party package is installed,
+  it will improve the calculation time.
 * gettext builder: disable extracting/apply 'index' node by default. Please set
   'index' to :confval:`gettext_enables` to enable extracting index entries.
 

doc/config.rst

@@ -430,6 +430,10 @@ documentation on :ref:`intl` for details.
    * Calculate similarity between new msgids and previously saved old msgids.
      This calculation take many time.
 
+   If you need a speed for the calculation, you can use ``python-levenshtein``
+   3rd-party package written in C by using
+   :command:`pip install python-levenshtein`.
+
    The default is ``False``.
 
    .. versionadded:: 1.3

sphinx/versioning.py

@@ -16,6 +16,11 @@ from itertools import product
 from six import iteritems
 from six.moves import range, zip_longest
 
+try:
+    import Levenshtein
+    IS_SPEEDUP = True
+except ImportError:
+    IS_SPEEDUP = False
 
 # anything below that ratio is considered equal/changed
 VERSIONING_RATIO = 65
@@ -109,7 +114,11 @@ def get_ratio(old, new):
     """
     if not all([old, new]):
         return VERSIONING_RATIO
-    return levenshtein_distance(old, new) / (len(old) / 100.0)
+
+    if IS_SPEEDUP:
+        return Levenshtein.distance(old, new) / (len(old) / 100.0)
+    else:
+        return levenshtein_distance(old, new) / (len(old) / 100.0)
 
 
 def levenshtein_distance(a, b):