truncate bytes instead of unicode (GC-422)

beets/library.py

@@ -1152,6 +1152,7 @@ class Library(BaseLibrary):
         # Encode for the filesystem.
         if not fragment:
             subpath = bytestring_path(subpath)
+        subpath = util.truncate_path(subpath, pathmod)
 
         # Preserve extension.
         _, extension = pathmod.splitext(item.path)

beets/util/__init__.py

@@ -452,16 +452,19 @@ def sanitize_path(path, pathmod=None, replacements=None):
     if not comps:
         return ''
     for i, comp in enumerate(comps):
-        # Replace special characters.
         for regex, repl in replacements:
             comp = regex.sub(repl, comp)
-
-        # Truncate each component.
-        comp = comp[:MAX_FILENAME_LENGTH]
-
         comps[i] = comp
     return pathmod.join(*comps)
 
+def truncate_path(path, pathmod=None):
+    """Given a bytestring path or a Unicode path fragment, truncate the
+    components to a legal length.
+    """
+    pathmod = pathmod or os.path
+    comps = [c[:MAX_FILENAME_LENGTH] for c in components(path, pathmod)]
+    return pathmod.join(*comps)
+
 def sanitize_for_path(value, pathmod, key=None):
     """Sanitize the value for inclusion in a path: replace separators
     with _, etc. Doesn't guarantee that the whole path will be valid;

docs/changelog.rst

@@ -58,6 +58,8 @@ Changelog
 * Add the track mapping dictionary to the ``album_distance`` plugin function.
 * When an exception is raised while reading a file, the path of the file in
   question is now logged (thanks to Mike Kazantsev).
+* Truncate long filenames based on their *bytes* rather than their Unicode
+  *characters*, fixing situations where encoded names could be too long.
 * Fix an assertion failure when the MusicBrainz main database and search server
   disagree.
 * Fix a bug that caused the :doc:`/plugins/lastgenre` and other plugins not to