Stream: t-compiler/wg-polonius

Topic: sync up


nikomatsakis (Apr 08 2019 at 17:11, on Zulip):

Hey @Albin Stjerna -- sorry, not sure if I communicted very clearly that I would be on vacation last week.

nikomatsakis (Apr 08 2019 at 17:11, on Zulip):

Maybe it'd be good for us to schedule some f2f time to chat and get sync'd again

Albin Stjerna (Apr 08 2019 at 17:42, on Zulip):

@nikomatsakis I don't think you mentioned that (or I didn't hear it anyway), but I presumed as much and made some work up when I got stuck. And yes, that sounds like a good idea!

Albin Stjerna (Apr 08 2019 at 17:43, on Zulip):

(among other things, the "work" I made up is benchmarking compilation of about 500 of the most popular crates in GitHub/crates.io

Albin Stjerna (Apr 08 2019 at 17:43, on Zulip):

Or I did collect 1500 of them, but they are still running

Albin Stjerna (Apr 08 2019 at 17:50, on Zulip):

@nikomatsakis When do you have the time next time, starting (my) tomorrow? I'm too tired tonight to make any sense I think

nikomatsakis (Apr 08 2019 at 17:50, on Zulip):

@Albin Stjerna yes sorry

nikomatsakis (Apr 08 2019 at 17:51, on Zulip):

I'm reasonably available tomorrow -- not sure what times work well for you

nikomatsakis (Apr 08 2019 at 17:52, on Zulip):

Would 3pm UTC-04:00 be ok?

Albin Stjerna (Apr 08 2019 at 17:56, on Zulip):

Would 3pm UTC-04:00 be ok?

That works!

Albin Stjerna (Apr 08 2019 at 18:02, on Zulip):

15:00 is 21:00 for me, which is OK for anything that isn't a lot longer than an hour, but most of your office hours most days work for me, up until 16:00 when it starts getting a bit late

nikomatsakis (Apr 08 2019 at 19:13, on Zulip):

Hey @WG-polonius -- @Albin Stjerna and I were talking about trying to sync up tomorrow at 15:00. Not sure if any of y'all could make it.

nikomatsakis (Apr 08 2019 at 19:13, on Zulip):

I'll create a calendar invite, anyway

nikomatsakis (Apr 08 2019 at 19:14, on Zulip):

Done

nikomatsakis (Apr 09 2019 at 18:30, on Zulip):

@WG-polonius -- would it be possible to start the meeting now?

Albin Stjerna (Apr 09 2019 at 18:32, on Zulip):

@WG-polonius -- would it be possible to start the meeting now?

Yes! Just give me a few minutes to get ready

nikomatsakis (Apr 09 2019 at 18:36, on Zulip):

Zoom link

nikomatsakis (Apr 09 2019 at 18:40, on Zulip):

Paper document

Albin Stjerna (Apr 09 2019 at 20:04, on Zulip):

Ok, the output now looks like this:

"Start(B2[1])" has live variables: ["V1"]
"Start(B2[0])" has live variables: ["V1"]
"Mid(B2[0])" has live variables: ["V1"]
"Mid(B1[1])" has live variables: ["V1"]
"Mid(B2[1])" has live variables: ["V1"]
"Start(B1[1])" has live variables: ["V1"]

I just added a method:

pub(crate) fn untern_vec(&self, data: &Vec<TargetType>) -> Vec<&str> {
        data.into_iter().map(|d| self.untern(*d)).collect()
    }
Albin Stjerna (Apr 09 2019 at 20:06, on Zulip):

(I now think that loop in the test is fishy in that it only goes through the liveness annotations, but I'll have a look at it tomorrow when it's not in the middle of the night)

nikomatsakis (Apr 09 2019 at 20:15, on Zulip):

@Albin Stjerna hm, that .. might be correct? I sort of forget the test :)

Albin Stjerna (Apr 10 2019 at 08:17, on Zulip):

@nikomatsakis I think it's wrong; given this:

        block B0 {
            invalidates(L0); // generate a point
            goto B1;
        }

        block B1 {
            var_defined(V1); // V1 dies
            invalidates(L0);
            goto B2;
        }

        block B2 {
            invalidates(L0);
            var_used(V1);
        }

The expected outcome would be that V1 is dead at B1 mid-point and up, right? If I recall correctly what you said yesterday. In the output, it is live at Mid(B1[1]) and dead before that

nikomatsakis (Apr 11 2019 at 14:34, on Zulip):

@Albin Stjerna it should go dead at Mid(B1[0]) -- and I think it does

nikomatsakis (Apr 11 2019 at 14:34, on Zulip):

i.e., I see it live at Start(B1[1]) -- but not B1[0]

Albin Stjerna (Apr 11 2019 at 14:35, on Zulip):

Yes, and those are the only places it's live at

Albin Stjerna (Apr 11 2019 at 14:35, on Zulip):

So it sounds like it's about right

nikomatsakis (Apr 11 2019 at 14:46, on Zulip):

those look like the right results -- but @Albin Stjerna, if we change the definition back from extend_anti to filter_anti, does the result change?

Albin Stjerna (Apr 11 2019 at 14:49, on Zulip):

Yes, it changes:

"Start(B2[1])" has live variables: ["V1"]
"Start(B2[0])" has live variables: ["V1"]
"Mid(B2[0])" has live variables: ["V1"]
"Mid(B1[1])" has live variables: ["V1"]
"Mid(B2[1])" has live variables: ["V1"]
"Start(B1[1])" has live variables: ["V1"]
"Mid(B1[0])" has live variables: ["V1"]
Albin Stjerna (Apr 11 2019 at 15:16, on Zulip):

I have pushed the new version using extend_anti, with a test that now verifies the edge case (ft. intelligible debug output)

nikomatsakis (Apr 11 2019 at 20:43, on Zulip):

@Albin Stjerna great

Albin Stjerna (Apr 12 2019 at 08:42, on Zulip):

@nikomatsakis Regarding the plan: I would like to just submit the paperwork on Tuesday or Wednesday next week, ideally, as I realised I have a train trip booked as a student in two weeks and risk fines if I'm not registered by then. I checked with the person who does the registering, and he can fast-track me so that everything goes through the same day though, as long as you sign the form I emailed and I have a plan to submit.

Albin Stjerna (Apr 12 2019 at 08:44, on Zulip):

I'm suggesting we treat the plan as a timeout; when the time is up, I submit whatever we have, and we can still revise it if we aren't done by then

nikomatsakis (Apr 12 2019 at 17:42, on Zulip):

@Albin Stjerna ok sorry will do that today

nikomatsakis (Apr 12 2019 at 17:43, on Zulip):

@Albin Stjerna we should create a new doodle, I think, to reschedule the meeting from Thursday -- and maybe we can use the next one to try and make a more detailed plan

Albin Stjerna (Apr 12 2019 at 17:46, on Zulip):

@nikomatsakis Sounds good, but I will have to use the preliminary one I have for my project application, but I think that's probably fine

Last update: Nov 15 2019 at 20:50UTC