Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
runtime: Document container ID charset and uniqueness domain
Allow the runtime to use it's own scheme, but let the caller use UUIDs if they want. Jonathan asked for clarification as part of opencontainers#87, but didn't suggest a particular approach [1]. When we discussed it in the 2015-08-26 meeting [2], the consensus was to just allow everything. With container IDs like 'a/b/c' leading to state entries like '/var/oci/containers/a/b/c/state.json'. But that could get ugly with container IDs that contain '../' etc. And perhaps there are some filesystems out there that cannot represent non-ASCII characters (actually, I'm not even sure what charset our JSON is in ;). I'd rather pick this minimal charset which can handle UUIDs, and make life easy for runtime implementers and safe for bundle consumers at a slight cost of flexibility for bundle-authors. There was some confusion on the list about what "ASCII letters" meant [3], so I've explicitly listed the allowed character ranges. Here's a Python 3 script that shows the associated Unicode logic: import unicodedata # http://www.unicode.org/reports/tr44/tr44-4.html#GC_Values_Table category = { 'Ll': 'lowercase letter', 'Lu': 'uppercase letter', 'Nd': 'decimal number', 'Pd': 'dash punctuation', } for i in range(1<<7): char = chr(i) abbr = unicodedata.category(char) if abbr[0] in ['L', 'N'] or abbr == 'Pd': cat = category[abbr] print('{:02x} {} {}'.format(i, char, cat)) [1]: opencontainers#87 (comment) [2]: https://github.com/opencontainers/specs/wiki/Meeting-Minutes-2015-08-26 [3]: https://groups.google.com/a/opencontainers.org/d/msg/dev/P9gZBYhiqDE/-ptpOcQ5FwAJ Message-Id: <7ec9cff6-c1a6-4beb-82de-16eb412bf2f8@opencontainers.org> Reported-by: Jonathan Boulle <jonathanboulle@gmail.com> Signed-off-by: W. Trevor King <wking@tremily.us>
- Loading branch information