Is there any way to keep Box as a primitive type in MIR?

[zhuyt]

I want to do some analysis in MIR. I want to know is there any command(-C or -Z) to generate MIR where Box occurs as a primitive type?

For example,

let mut boxn = Box::new(Node { x : 100}); // Node is a struct with only one filed.
boxn.x = 100;

Then, the generated MIR looks like below.

_3 = std::boxed::Box::<Node>::new(Node {{ x: 14_i32 }})
_24 = (((_3.0: std::ptr::Unique<Node>).0: std::ptr::NonNull<Node>).0: *const Node);
((*_24).0: i32) = 100_i32;

As we can see, boxn.x is unwind as a Place(_24) expression in MIR. I wonder is there any way to remove the first assignment. The generated MIR looks like

((*_3).0 : i32) = 100_i32

In addition, we can write

let mut boxn = Box::new(Node { x : 100}); // Node is a struct with only one filed.
boxn.deref_mut().x = 100;

Thus, we can get

_3 = std::boxed::Box::<Node>::new(Node {{ x: 14_i32 }})
_24 = <std::boxed::Box<Node> as std::ops::DerefMut>::deref_mut(move _25)
((*_24).0: i32) = 100_i32;

Whatever, can we keep Box as a primitive type in MIR?

[beepster]

You'll want to get MIR from before the passes that do this transformation. -Zdump-mir=Underefer.before should give it to you (unless I wrote the pass name wrong)

If you're trying to write a program that analyzes MIR I'd recommend using rustc as a library instead of analyzing the text output. You'd be able to get help with that on internals.rust-lang.org or the zulip

[scottmcm]

@zhuyt

Whatever, can we keep Box as a primitive type in MIR?

What's your actual goal? The direction in general (see also things like RFC: No (opsem) Magic Boxes by chorman0773 · Pull Request #3712 · rust-lang/rfcs · GitHub ) is making Box less special, and thus especially in runtime MIR it would be entirely normal and expected for types like Box to be SRoA'd away.