Stream: t-compiler/wg-polonius

Topic: tomorrow's focus?


nikomatsakis (Nov 19 2019 at 17:10, on Zulip):

Not sure what we should do tomorrow, curious to hear thoughts -- seems like both @lqd and @Albin Stjerna are working on the coding side of things, I could try to explore the equality stuff some tomorrow? update docs? not sure.

lqd (Nov 19 2019 at 17:10, on Zulip):

either the equality stuff or higher ranked subtyping with chalk ?

nikomatsakis (Nov 19 2019 at 17:11, on Zulip):

I think that the rules hackmd is fairly complete at this point, modulo some naming inconsistencies

nikomatsakis (Nov 19 2019 at 17:11, on Zulip):

(er and modulo equality)

lqd (Nov 19 2019 at 17:12, on Zulip):

omg how could I forget I had gathered the rules for the equality work :face_palm:

lqd (Nov 19 2019 at 17:12, on Zulip):

https://github.com/lqd/datapond/blob/master/tests/flow_sensitive_equality_rules.rs :face_palm:

Albin Stjerna (Nov 19 2019 at 17:12, on Zulip):

I'm going to have less time tomorrow, but I'll try to steal a few hours

Albin Stjerna (Nov 19 2019 at 17:13, on Zulip):

I’ll try to at least implement Niko’s changes to my rules for initialisation

lqd (Nov 19 2019 at 17:14, on Zulip):

nice

Albin Stjerna (Nov 19 2019 at 17:14, on Zulip):

I’m suggesting my next focus is measureme

lqd (Nov 19 2019 at 17:15, on Zulip):

that would be worthwhile as well

Albin Stjerna (Nov 19 2019 at 17:15, on Zulip):

For Polonius, not fact generation

Albin Stjerna (Nov 19 2019 at 17:15, on Zulip):

I’m imagining a larger fact generation restructuring later, and I think profiling fits nicely into that

nikomatsakis (Nov 19 2019 at 18:50, on Zulip):

One question @lqd that might be worth talking over tomorrow is whether we can get a version of the naive rules that performs "reasonably" -- e.g., by limiting ourselves to only those loans that are invalidated anywhere or something -- I know you had various bits of work in that direction? I'm just thinking that all these different PRs would be easier to land if they only had to contend with the naive rules

lqd (Nov 19 2019 at 19:03, on Zulip):

true

lqd (Nov 19 2019 at 19:04, on Zulip):

this should be super easy however iiuc

lqd (Nov 19 2019 at 19:05, on Zulip):

except the results are a bit unclear now

lqd (Nov 19 2019 at 19:06, on Zulip):

it would also need adjustments to account for placeholder loans

nikomatsakis (Nov 19 2019 at 19:07, on Zulip):

except the results are a bit unclear now

what results are unclear?

lqd (Nov 19 2019 at 19:08, on Zulip):

and with the placeholder loans live everywhere I’m not sure how the filtering would impact things

lqd (Nov 19 2019 at 19:08, on Zulip):

the performance results over a bigger set of benchmarks

lqd (Nov 19 2019 at 19:09, on Zulip):

it was very effective clap-like datasets where a lot of work was effectively done to not find any errors

lqd (Nov 19 2019 at 19:09, on Zulip):

it’s probably not as extreme everywhere

lqd (Nov 19 2019 at 19:14, on Zulip):

I mean, it can’t hurt (if we are careful) :)

nikomatsakis (Nov 19 2019 at 19:22, on Zulip):

the performance results over a bigger set of benchmarks

OK, I see. I tend to think we should land it regardless, but the real benchmark I suppose (right now) is the "time to run a x.py test run" ?

nikomatsakis (Nov 19 2019 at 19:22, on Zulip):

that is, I still think we want to get to the place where we are testing with x.py test --compare-mode=polonius pretty regularly

nikomatsakis (Nov 19 2019 at 19:22, on Zulip):

but I don't really care that much about overall perf just now so long as a test run can complete in a reasonable amount of time

lqd (Nov 19 2019 at 19:23, on Zulip):

ah but the tests did pass with the prototype filtering (except the OOMs in fact generation)

lqd (Nov 19 2019 at 19:24, on Zulip):

and yes agreed, let’s land this soon

lqd (Nov 19 2019 at 19:25, on Zulip):

it’d be great to be able to run the compare mode on CI, I feel we’re pretty close to being able to do that

lqd (Nov 19 2019 at 19:26, on Zulip):

(the time it takes to run tests did seem acceptable)

lqd (Nov 19 2019 at 19:28, on Zulip):

all this rambling to say, yes, it could be an interesting topic on which to spend a bit of time tomorrow :)

nikomatsakis (Nov 20 2019 at 14:02, on Zulip):

Hey y'all :)

nikomatsakis (Nov 20 2019 at 14:02, on Zulip):

I may not be able to stick around quite as long today

nikomatsakis (Nov 20 2019 at 14:02, on Zulip):

I have a last minute errand I have to do before upcoming trip next week and this morning is literally the only time I can do it, it seems

lqd (Nov 20 2019 at 14:05, on Zulip):

iceland :snowman: :)

Last update: Dec 12 2019 at 00:55UTC