@Chris Vittal btw I rebased the
relation branch and pushed some WIP commits starting to add something that uses the edge relation to compute subset relation. I confess I've not yet even tried it. :P
and we'll need to improve the tests to compare for equality...
also, it does a bunch of extra work now to compute equal results to timely (e.g., working out the transitive closure), but I figure we can modify how the timing works to just time the "inner loop"
also I want it known that I am using the word "bespoke" ironically =)
I'M NOT THAT HIP NORMALLY
...it is sort of a fun word to say though...
Alright. I'll take a look. Maybe try to finish it up, add some tests.
ok, are you actively poking at it? (I am not just now, but might)
Not right now
ok so @Chris Vittal I pushed a commit that actually runs the new relation
it panics :)
I will look maybe a bit into why, though I was thinking maybe of writing a "naive relation" so I can test the harness itself better...
oh, the panic turns out to be a bug in the harness definitely
ok, fixed it, and the test passes now
(however, running on clap still panics)
@qmx where'd you push those stats changes?
@nikomatsakis I've forgot to push, will do in a few
@nikomatsakis I'm preparing the PR now, do you want it against another branch than master?
master would do
ps we're in the wrong thread :)
that's the annoying part about threads I guess...
I think you can change the topic when editing a message
@Chris Vittal ok pushed the driver + tests etc
er, wait, did my rebase screw something up...
oh, I see
ok, now pushed
got it, what are the current todos?
well, first off it panics when run against the clap test
jfyi, I did some experimentation with a matrix version too. Neither one works when deployed at scale =)
also, I pushed some things to the branch, hopefully won't disturb ya'
@nikomatsakis Most of what I was able to do this weekend was find a test that reveals a bug in the edge removal code. It's in your repo now. I'll dig into why it fails in the morning.
ok -- I made great progress on the timely dataflow-based implementation, but I haven't made any more on this
@Chris Vittal to be perfectly honest, with this PR doing so well, I'm thinking we it's better to stick with differential-dataflow
though I definitely have no objection to us trying to do better via custom code
Okay. I'll spend some of today trying to fix it and see where it is with respect to your new PR.
But I won't take more than today and maybe tomorrow, but considering I have a 10 line test case, it shouldn't be too hard to debug and fix. (I hope)
we need a :famous last words: emoji
Zulip does support custom emojis =)
Okay, I hope that the kinks have been ironed out, but the graph relationship is slow. Slower than even the naive version. shrugs
@nikomatsakis what now? I could do some of the UI stuff, like command line arguments. But I think further poking on this particular representation may not be a good use of time.