b8c051c89f
It occurred to me today that I implemented the sequencer incorrectly and hadn't yet noticed because I don't have any tests that are complex/interleaved/perhaps-non-deterministic to find the problem. The problem is that the sequencer's current implementation only keeps track of the last LPN for any Tango stream. The fix is to do what the paper actually says: the sequencer keeps a *list* of the last $K$ LPNs for each stream. Derp. Yes, that's really necessary to avoid a pretty simple race condition with 2 actors simultaneously updating a single Tango stream. 1st commit: fix the implementation and the smoke test. The broken-everything-else will be repaired in later commits. |
||
|---|---|---|
| .. | ||
| tango-prototype | ||