#rust-lang

dtolnay: RalfJ: https://play.rust-lang.org/?gist=e7eda34df05562...

nmatsakis: if we make "caller loc unwrap" modify the Abi::Rust part of the *type* then we have to care *so much less* about its interactions

nmatsakis: where can we put stuff like this?

I want to not lose this stuff and idk if there's a tracking issue

eddyb: would it break `let x: fn(Option<i32>) -> i32 = Option::unwrap`?

kennytm: ... yes

ugh

kennytm: I mean we could still have the shim

it'd just be harder to mess up in the compiler

shim all the things

nox: do you want to announce the WG-codegen?

Yep.

cramertj: lol this is already worrying https://play.rust-lang.org/?gist=3498abd639f85c...

cramertj: https://play.rust-lang.org/?gist=179e43e6fb0286...

so this is all safe, now I'm curious what you can do starting from here

cramertj: this OTOH is unsound right? https://play.rust-lang.org/?gist=e1240c2e66d6e6...

cramertj: https://play.rust-lang.org/?gist=05d7ad421e6ba7...

WG-codegen is actually becoming a thing???

wow congrats nox

est31: And I just decreased by 10% the size of the Servo exe today, apparently.

nox: Did you try it ? Did it crash ?

eijebong: ./mach run succeeded. I have yet to run CI.

eijebong: But there is no way in hell I'm letting this go if CI fails, if there is a crash I'll pursue the fix.

nox: \o/

This is insanely important for Gecko after all.

Also I just added some LLVM pass at the end of the LLVM pipeline,

deduplicating all that stuff before most optimisations should also increase build times.

We are going for dinner, see you later.

Yeah, I saw the discussions around it

Bon appétit !

sunken cost fallacy*

est31: I'm confused wrt. your comments on sandboxing

aand a "Rust has unsafe so it is just as bad as C"

i mean... are we running a VM or something?

centril: on modern operating systems you can put processes into sandboxes

oh god I realize most of my messages from today in here

you can say "I want them to have no network access"

were meant for #rustc

est31: OK; so what do we allow for build.rs

centril: it is not sandboxed

est31, centril, nmatsakis: https://github.com/servo/gaol

right ^

btw if we want to just sandbox random stuff like proc macros

Servo has done the hard work :P

centril: but proc macros are being abused quite much for example

I forgot whom I forgot to send the gaol link to

some people do I/O during proc macro eval

this is a funny commit https://github.com/servo/gaol/commit/b309234ff0...

eddyb: but we are not using this gaol for build.rs right?

no

centril: it's just an example of a sandbox thing we can use from Rust effortlessly

I am not suggesting we should

build.rs can send your ssh keys to russia if it wants to

est31: OK; so if it is not sandboxes, then install.rs would not be either?

or draw frolicing ponies on your screen

centril: there is a case for install.rs to be actually sandboxed

it would ensure that it is well behaved

not doing the installation on its own but rather asking cargo to do it

est31: so if we don't sandbox install.rs, then all talk of guarantees is outa the window

no

est31: care to elaborate on 'no' ?

centril: that's a bit fatalistic

its not black vs white

if we say "you should use it this way and that way"