Otherwise, it uses the system's locale database (and produces
significantly worse results).
+This module tries very hard not to fail loudly. It tends to ignore
+most Unicode recoding errors, and will eventually fall back to the C
+locale or raw codepoint-based collation. If you would like loud
+failure, you can use the collate.strings module and the individual
+Collators directly.
+
Trivial Use:
------------
strings = read_strings(...)