Select an image by similarity if multiple images are globbed by `.. image:: filename.*`

Until this point, it is used the last image returned from glob.glob().
The order of glob.glob() is determined by filesystem.
After this, Sphinx uses filename similarity to determine the image.

CHANGES

@@ -105,6 +105,7 @@ Features added
 * #1853: support custom text splitter on html search with ``language='ja'``.
 * #2320: classifier of glossary terms can be used for index entries grouping key.
   The classifier also be used for translation. See also :ref:`glossary-directive`.
+* Select an image by similarity if multiple images are globbed by ``.. image:: filename.*``
 
 Bugs fixed
 ----------

sphinx/environment.py

@@ -899,15 +899,18 @@ class BuildEnvironment:
             # set imgpath as default URI
             node['uri'] = rel_imgpath
             if rel_imgpath.endswith(os.extsep + '*'):
+                globbed = {}
                 for filename in glob(full_imgpath):
                     new_imgpath = relative_path(path.join(self.srcdir, 'dummy'),
                                                 filename)
                     try:
                         mimetype = guess_mimetype(filename)
-                        candidates[mimetype] = new_imgpath
+                        globbed.setdefault(mimetype, []).append(new_imgpath)
                     except (OSError, IOError) as err:
                         self.warn_node('image file %s not readable: %s' %
                                        (filename, err), node)
+                for key, files in iteritems(globbed):
+                    candidates[key] = sorted(files, key=len)[0]  # select by similarity
             else:
                 candidates['*'] = rel_imgpath