In between meetings, I spent most of today continuing yesterday’s work on the citation use case. Further query rewrites and testing exposed an interesting bug in the optimizer due to a mismatch between logical algebra representation and the actual system implementation behavior — the optimizer assumed the system could perform an operation which the system could not. Fixed the bug, which was thankfully short-lived!

After the last few days’ worth of heckling, what I believe is the largest computational step in the algorithm is now finished. At the start, it took about 4 hours to find shared citations 19 levels deep, at which point Myria’s built-in iteration limit (there to protect against infinite loops by bad queries that I write) was triggered. I raised the iteration limit and with all the improvements to Raco and to Myria, it now takes 4 hours to fully walk the graph.

The longest train of citations is 69 levels deep! Tomorrow, I hope to find the actual shared citations for pairs of papers.