Closed jingjingwang closed 7 years ago
I expect myself adding some more checks to limit what can be put inside of the do-until-convergence loop, but would like to get the main changes being reviewed first.
For 059673c, I'm not sure if it's the best design. I would like to add a MIN aggregate that takes a list of column indices to compare two tuples in that order, however the current MIN is a binary expression. I added a new aggregate called MULTIMIN instead with less support (e.g. fakedb) than MIN, but maybe extending MIN to an nary expression is another option? Would like to hear your thought.
I've addressed most comments except for a few pending discussions:
Zeroary
MyriaConsumer
for breaking infinite optimizationMIN
vs. MULTIMIN
and should they use named attribute (and what if we don't know the name)We can discuss them in person if necessary.
Goes with https://github.com/uwescience/myria/pull/877.
This PR adds support for a new syntax: a do-until-convergence loop containing a list of special MyriaL statements. The loop terminates when there will be no update to any relation. Examples are those newly added optimizer tests.