Stream: t-compiler/wg-mir-opt

Topic: Optimizing BodyAndCache


Paul Faria (Dec 03 2019 at 18:10, on Zulip):

Hi @oli , now that BodyCache (soon to be BodyAndCache per recommendation from eddyb through https://github.com/rust-lang/rust/pull/66991) is in, should I move forward with figuring out which fn's are good candidates for an #[inline] attribute? Or should I jump back to the wg-parallel-rustc project?

oli (Dec 03 2019 at 18:14, on Zulip):

iirc eddy fixed it themselves. Not sure if the inline stuff gets us anywhere, so... I guess you removed another stumbling stone for parallel rustc and can search for the next one :D

Paul Faria (Dec 03 2019 at 18:23, on Zulip):

I think my change also includes the removal of that body fn we tried getting rid of before. Otherwise it's the same (not sure if his also cleans up the reborrows)

oli (Dec 03 2019 at 18:58, on Zulip):

I'll check it out, but let's wait until their PR is merged

RalfJ (Dec 07 2019 at 10:23, on Zulip):

BodyCache (soon to be BodyAndCache)

And there, Miri breaks again. ;)

RalfJ (Dec 07 2019 at 10:24, on Zulip):

(but we did get Miri into yesterday's nightly so all is fine)

Paul Faria (Dec 09 2019 at 02:23, on Zulip):

I had no idea my changes broke miri. So sorry! What steps would I need to take to ensure it doesn't break in the future? I tend to take longer since I can't work on the project full time, and ensuring my changes cause as little trouble as possible is more important to me than getting something out quickly

oli (Dec 09 2019 at 10:42, on Zulip):

If you want to see whether clippy or miri break, you can run ./x.py test src/tools/miri (or clippy) and see if that still compiles. If it does not, you can fix it until it works again and then open a PR with your changes to the respective repo

RalfJ (Dec 11 2019 at 15:32, on Zulip):

@Paul Faria it's okay, Miri gets broken at least once a month. That is the price we pay for using internal compiler APIs. we also benefit a lot from that.

RalfJ (Dec 11 2019 at 15:32, on Zulip):

We just had a bad streak of Miri being broken in 4 different ways by 4 PRs that landed within 48h or so^^

RalfJ (Dec 11 2019 at 15:33, on Zulip):

testing if clippy/miri break takes a lot of time, and I don't expect all rustc contributors to do that. if people help with the fallout after landing their PR, that is appreciated, but not mandatory. :)

RalfJ (Dec 11 2019 at 15:34, on Zulip):

but it would be good to be mindful that there are (semi-)external consumers of some of these APIs, so we should avoid renaming a type 3 times while iterating towards the best name.

Last update: Sep 18 2020 at 20:00UTC