Rustup

[oli-obk]

No description provided.

[RalfJung]

Why does this not use the usual memory write functions, but special-case locals?

(I realize you did not change this, but I am still curious and this might be a good time to add a comment.^^)

[oli-obk]

The memory write functions would force an allocation... at least for uninit. not sure why init is implemented the same way, it does need the allocation for bigger things.

[RalfJung]

Okay. Would be nice to have a comment along those lines.

[RalfJung]

Same question as for init

[RalfJung]

Uh, strange assertion failures in cast.rs when running for a foreign architecture...

[RalfJung]

This is the failing assertion:

https://github.com/rust-lang/rust/blob/a8763b5370ed044b6e703981217f074e8ddf1e7c/src/librustc_mir/interpret/cast.rs#L158

[oli-obk]

try running the code with RUST_LOG=rustc_mir::interpret

[RalfJung]

I first have to compile latest rustc with debug assertions to get there...

[RalfJung]

I am afraid I have no idea how to debug this. My self-compiled rust does not have a 32bit libstd, building one with xargo does not work (see #202), and using https://github.com/kennytm/rustup-toolchain-install-master/ does not help either as it seems we do not have a 32bit alt build.

[RalfJung]

Finally, got a debug setup. For the record:
I changed Xargo.toml to

[dependencies]
core = {stage = 1}
compiler_builtins = {stage = 2}

then ran

XARGO_RUST_SRC=~/src/rust/rustc.2/src RUSTFLAGS='-Zalways-encode-mir -Zmir-emit-validate=1' xargo build --target i686-unknown-linux-gnu

and patched the code of a failing testcase to not require std:

#![feature(panic_implementation)]
#![feature(lang_items)]
#![no_std]
use core::panic::PanicInfo;

#[panic_implementation]
fn my_panic(pi: &PanicInfo) -> ! {
    loop {}
}

#[lang = "eh_personality"] extern fn eh_personality() {}

fn main() {
    assert_eq!(core::char::from_u32('x' as u32), Some('x'));
}

[RalfJung]

Yeah that looks wrong

Casting Bits { size: 4, bits: 1 }: u64 to isize

we should never have a u64 with size 4...

Seems the value is coming from

_4 = const core::intrinsics::discriminant_value(move _5)

[RalfJung]

What, why does that one commit work now...?

Oh. AppVeyor does not actually test the given commit. It tests the PR at the time the queue proceeds far enough. WTF?!?