Stream: t-compiler/wg-polonius

Topic: meeting 2019.10.22


lqd (Oct 22 2019 at 16:23, on Zulip):

I think Albin is at SPLASH today, so I'm not sure there will be many people tonight (in 2.5 hours).

lqd (Oct 22 2019 at 16:23, on Zulip):

I guess I can give my update now: I've worked on placeholder loans (and there is a dedicated thread with a couple thoughts/concerns) and I've since then added it to DatafrogOpt. All in all it was easy and I should have done it sooner. And it's at least compatible with pruning/focusing on invalidated loans (or potential loan errors from LocationInsensitive, but as mentioned in the thread, these errors kinda defeat the usefulness of this variant right now)

lqd (Oct 22 2019 at 16:24, on Zulip):

No PR yet, because of the existing PRs in-flight and I'm unsure on how best to report these different kinds of errors, eg contexts from #134 changes how the variants return errors, so feedback on the general direction could affect what I've done for illegal subset errors. Albin is somewhat in the same boat for their initialization PR, with move errors (and I'm not at all aware of the plans about those, or when they are expected to be computed, and the impact of initialization on performance).

lqd (Oct 22 2019 at 16:25, on Zulip):

if someone is here tonight we can chat, but otherwise apart from this work on placeholder loans (and some random things not particularly worth discussing like parser work for eventually moving off of lalrpop), things/questions/etc are looking very much like last week

nikomatsakis (Oct 22 2019 at 19:04, on Zulip):

hey all

nikomatsakis (Oct 22 2019 at 19:04, on Zulip):

I was having issues with the zulip app

lqd (Oct 22 2019 at 19:04, on Zulip):

:wave:

nikomatsakis (Oct 22 2019 at 19:04, on Zulip):

I was hoping to spend the last 30 minutes or so catching up on polonius threads but...

nikomatsakis (Oct 22 2019 at 19:04, on Zulip):

I guess the website is working though, so maybe i'll do that now

nikomatsakis (Oct 22 2019 at 19:04, on Zulip):

my talk went well I think

lqd (Oct 22 2019 at 19:04, on Zulip):

ah nice I was about to ask

lqd (Oct 22 2019 at 19:06, on Zulip):

did people have questions ?

nikomatsakis (Oct 22 2019 at 19:06, on Zulip):

yes, lots

nikomatsakis (Oct 22 2019 at 19:06, on Zulip):

I stopped in the middle to let people ask questions and they did not disappoint

nikomatsakis (Oct 22 2019 at 19:07, on Zulip):

which I think was good because for people to really get it, you kind of had to say it 3 or 4 times in a row

lqd (Oct 22 2019 at 19:07, on Zulip):

oh :)

nikomatsakis (Oct 22 2019 at 19:07, on Zulip):

and having people ask questions gave me a chance to do that

nikomatsakis (Oct 22 2019 at 19:07, on Zulip):

without sounds repetitive :)

lqd (Oct 22 2019 at 19:07, on Zulip):

that's lucky and useful :)

nikomatsakis (Oct 22 2019 at 19:07, on Zulip):

there should be a video at some point

nikomatsakis (Oct 22 2019 at 19:07, on Zulip):

though I believe carol10cents was promising bad things for the next person to ask for a specific date

nikomatsakis (Oct 22 2019 at 19:08, on Zulip):

ok so shall I go catch up on threads?

lqd (Oct 22 2019 at 19:08, on Zulip):

sure, if you have some time

lqd (Oct 22 2019 at 19:09, on Zulip):

though I believe carol10cents was promising bad things for the next person to ask for a specific date

= the exact reason why I didn't ask for when it would be available ^^
tho shep mentioned you had a shoutout to the WG at the end :)

nikomatsakis (Oct 22 2019 at 19:09, on Zulip):

heh yes I did

nikomatsakis (Oct 22 2019 at 19:10, on Zulip):

https://nikomatsakis.github.io/rust-belt-rust-2019/#97

nikomatsakis (Oct 22 2019 at 19:10, on Zulip):

@lqd I was thinking --

nikomatsakis (Oct 22 2019 at 19:11, on Zulip):

I'm still having this feeling like "gah there are so many threads of exploration", feeling a bit overwhelmed by it all :)

nikomatsakis (Oct 22 2019 at 19:11, on Zulip):

probably because I just haven't had many spare cycles to really check in but

nikomatsakis (Oct 22 2019 at 19:11, on Zulip):

I think it'd be good for us to start landing some of those great work you've been doing

nikomatsakis (Oct 22 2019 at 19:11, on Zulip):

what do you think about the "datafrog compiler"?

nikomatsakis (Oct 22 2019 at 19:11, on Zulip):

(to start)

nikomatsakis (Oct 22 2019 at 19:11, on Zulip):

it looked pretty awesome

lqd (Oct 22 2019 at 19:12, on Zulip):

there are a lot of threads :)

lqd (Oct 22 2019 at 19:12, on Zulip):

land it in datafrog ?

nikomatsakis (Oct 22 2019 at 19:12, on Zulip):

well

nikomatsakis (Oct 22 2019 at 19:12, on Zulip):

maybe

nikomatsakis (Oct 22 2019 at 19:12, on Zulip):

yeah probably

nikomatsakis (Oct 22 2019 at 19:12, on Zulip):

not same crate I would say

nikomatsakis (Oct 22 2019 at 19:12, on Zulip):

but I think same repo is probably logical

lqd (Oct 22 2019 at 19:13, on Zulip):

I had published it temporarily in https://github.com/lqd/datapond for Vytautas to play around

nikomatsakis (Oct 22 2019 at 19:13, on Zulip):

one thing I am thinking about is

nikomatsakis (Oct 22 2019 at 19:13, on Zulip):

the role of the more optimized variants

nikomatsakis (Oct 22 2019 at 19:13, on Zulip):

I would kind of like to "pare back" the repo dramatically

lqd (Oct 22 2019 at 19:13, on Zulip):

and dogfood it a bit before upstreaming but yeah we can do that whenever

nikomatsakis (Oct 22 2019 at 19:13, on Zulip):

it seemed like you were able to make naive "competitive" by applying some filtering

nikomatsakis (Oct 22 2019 at 19:14, on Zulip):

I think right now I'm feeling like I'm not that interested in perf, it seems like (as we've said) we should be focusing on figuring out the full computation first

lqd (Oct 22 2019 at 19:14, on Zulip):

I'd like to check this on a bigger set of benchmarks for sure tho

nikomatsakis (Oct 22 2019 at 19:14, on Zulip):

and I will say that your compiler seems like it offers a big win -- if the only problem is we lose leapjoin, well, that's just perf we can recover later

nikomatsakis (Oct 22 2019 at 19:14, on Zulip):

I'd like to check this on a bigger set of benchmarks for sure tho

the main thing would be if running the x.py test suite takes like 200 years

nikomatsakis (Oct 22 2019 at 19:14, on Zulip):

that seems like a problem

lqd (Oct 22 2019 at 19:14, on Zulip):

it's not that severe, but the OOMs are a blocker

lqd (Oct 22 2019 at 19:15, on Zulip):

(and probably the Hybrid variant helps in keeping a lot of tests from going through the costlier variant)

lqd (Oct 22 2019 at 19:16, on Zulip):

also initialization and liveness (the latter can be similarly pruned if my understanding is correct) are dominating right now

nikomatsakis (Oct 22 2019 at 19:16, on Zulip):

yeah so keeping the hybrid setup might be ok too

nikomatsakis (Oct 22 2019 at 19:17, on Zulip):

I guess it's not that important to remove datafrog-opt

lqd (Oct 22 2019 at 19:17, on Zulip):

but yeah it would be interesting to have some holistic design, wrt to all the different parts of computations, the errors they each produce, etc

nikomatsakis (Oct 22 2019 at 19:18, on Zulip):

yes, I'm mostly feeling an urge to get the "big picture" cleared up

nikomatsakis (Oct 22 2019 at 19:18, on Zulip):

anyway I guess I'll go study threads a bit

nikomatsakis (Oct 22 2019 at 19:18, on Zulip):

but I watned to put that out there

nikomatsakis (Oct 22 2019 at 19:18, on Zulip):

second thought is that I want to think about doing some kind of "polonius work week"

lqd (Oct 22 2019 at 19:18, on Zulip):

there is some premise of that in one of the PRs

nikomatsakis (Oct 22 2019 at 19:18, on Zulip):

not necessarily a week, but basically finding some period of time where we can try to do a sprint of sustained attention

lqd (Oct 22 2019 at 19:19, on Zulip):

that would be super nice

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

I feel like we could make a lot of progress that way

lqd (Oct 22 2019 at 19:19, on Zulip):

yeah, same

lqd (Oct 22 2019 at 19:19, on Zulip):

having a lot of "disjointed" work can be counterproductive

Albin Stjerna (Oct 22 2019 at 19:20, on Zulip):

Agreed!

nikomatsakis (Oct 22 2019 at 19:20, on Zulip):

I feel like I could plan to take one of the upcoming weeks, or a lot o fit

nikomatsakis (Oct 22 2019 at 19:20, on Zulip):

/me has earned a "vacation" dang it :P

Albin Stjerna (Oct 22 2019 at 19:20, on Zulip):

YES

nikomatsakis (Oct 22 2019 at 19:20, on Zulip):

a more crazy idea would be to fly out to France or Europe somewhere :P

lqd (Oct 22 2019 at 19:20, on Zulip):

:D

nikomatsakis (Oct 22 2019 at 19:20, on Zulip):

but I'd have to bring my family with or they would kill me ;)

lqd (Oct 22 2019 at 19:21, on Zulip):

the more the merrier

nikomatsakis (Oct 22 2019 at 19:21, on Zulip):

anyway, presuming I did not fly to Europe, but rather we just kind of planned to do a sprint in a few weeks or something -- seems like y'all would be into that? I wasn't sure how it would work out with jobs etc :)

lqd (Oct 22 2019 at 19:21, on Zulip):

yeah something can be worked out :)

lqd (Oct 22 2019 at 19:22, on Zulip):

(there were a lot of different things to do tbf, I'm not saying it's wrong for us to have been doing it the way we have, esp as I'm the one bumbling around in many different directions :)

nikomatsakis (Oct 22 2019 at 19:23, on Zulip):

no, not wrong

nikomatsakis (Oct 22 2019 at 19:23, on Zulip):

it just feels like it'd be good to pull things together again

lqd (Oct 22 2019 at 19:23, on Zulip):

right exactly

Albin Stjerna (Oct 22 2019 at 19:24, on Zulip):

I have to go, but I’ll read up on what you are doing

lqd (Oct 22 2019 at 19:24, on Zulip):

and have a clear vision for the rest of the things needing to be done

lqd (Oct 22 2019 at 19:24, on Zulip):

@Albin Stjerna enjoying SPLASH hopefully :)

nikomatsakis (Oct 22 2019 at 19:25, on Zulip):

ok let me look at calendar, but I'm excited about this idea

Albin Stjerna (Oct 22 2019 at 19:25, on Zulip):

Right now I’m probably going to bed because I have to be student volunteering at 7:30 tomorrow, at the reception when we will have the big onslaught of visitors to get their badges but yes

lqd (Oct 22 2019 at 19:25, on Zulip):

there's rustfest soon (9-10 nov IIRC)

nikomatsakis (Oct 22 2019 at 19:30, on Zulip):

Hmm, I wasn't planning on attending rustfest, but I suppose I could conceivably try to crash last minute

nikomatsakis (Oct 22 2019 at 19:30, on Zulip):

but I think it'd really be a lot more realistic to just try to work remotely as usual ;)

nikomatsakis (Oct 22 2019 at 19:30, on Zulip):

are y'all going?

lqd (Oct 22 2019 at 19:31, on Zulip):

I am

lqd (Oct 22 2019 at 19:31, on Zulip):

I don't think Albin is

lqd (Oct 22 2019 at 19:32, on Zulip):

(oli, vytautas, pietro & tmandry ofc, etc)

lqd (Oct 22 2019 at 19:32, on Zulip):

I wouldn't be super available around those dates

lqd (Oct 22 2019 at 19:33, on Zulip):

so working remotely is no problem at all

Albin Stjerna (Oct 22 2019 at 19:51, on Zulip):

I'm not going :(

Albin Stjerna (Oct 22 2019 at 19:51, on Zulip):

Ooooo I wonder if I have vacation days I can take out, I'll have to look into that

Albin Stjerna (Oct 22 2019 at 19:52, on Zulip):

...oh wait, that's a weekend

lqd (Oct 22 2019 at 19:54, on Zulip):

tickets might be sold out :/

Albin Stjerna (Oct 22 2019 at 19:55, on Zulip):

I didn't mean I would go, but I could certainly be available

Albin Stjerna (Oct 22 2019 at 19:56, on Zulip):

(I'm waiting for my flatmate to come home so I can sleep; we have exactly one key to the AirBnB)

lqd (Oct 22 2019 at 20:56, on Zulip):

https://nikomatsakis.github.io/rust-belt-rust-2019/#97

omg <3
also loving the one for @Matthew Jasper :)

lqd (Oct 22 2019 at 20:57, on Zulip):

thanks @nikomatsakis :)

Albin Stjerna (Oct 23 2019 at 11:14, on Zulip):

Yeah thanks

Last update: Nov 15 2019 at 20:25UTC