Add `Rvalue::ShallowInitBox` to MIR

[nbdd0121]

Proposal

Introduce a new Rvalue variant, Rvalue::ShallowInitBox. It accepts an operand and a type, and will convert the operand to a shallowly-initialized box. This design is considered the better approach than a Box terminator after the Zulip dicussion.

In today's the MIR, box expr is translated to:

let _1: Box<ty>;
_1 = Box(ty);
(*_1) = expr;

however this does not allow unwind. Unwinding is necessary for rust-lang/rfcs#2116's oom=panic support (tracked in rust-lang/rust#43596).

Under this proposal, the originally Rvalue::NullaryOp(NullOp::Box, ty) is split into two operations, first a call to exchange_malloc(size_of::<ty>(), align_of::<ty>()), then converts it into a shallow-initialized box:

let _1: *mut u8;
let _2: Box<ty>;
bb0: {
	_1 = exchange_malloc(SizeOf(ty), AlignOf(ty)) -> [return: bb1, unwind: bb2];
}
bb1: {
	_2 = ShallowInitBox(move _1, ty);
	(*_2) = expr;
}

with the extra unwinding edge provided by the exchange_malloc call.

The operational behaviour of ShallowInitBox is simply transmuting *mut u8 to Box<T>. The only thing special is that dataflow analysis will treat the content of the box as uninitialized.

An alternative is to remove box_syntax and Box nullary op from MIR at all; however, an attempt in rust-lang/rust#87781 shows very significant regression when doing so (perf run). Here is an godbolt snippet shows a case where box still needs special treatment in the MIR, and thus box_syntax is still needed at least as an implementation detail of the library.

Another alternative is the original proposal, to add a TerminatorKind::Box. However adding a new terminator requires much more changes than the current proposal.

The original proposal

Introduce a new terminator, TerminatorKind::Box to the MIR of form:

_1 = Box(ty) -> [return: bb1, unwind: bb2];

it'll be roughly equivalent to

_1 = Box(ty) // Nullary op Box here
goto -> bb1

in today's MIR, but has an additional unwind edge.

Box needs to be a special terminator rather than a function call, because a function call will deep-initialize the return place while Box will only shallow-initialize its return place.

The overall plan is:

• Introduce TerminatorKind::Box
• Benchmark performance difference from using Box terminator vs Box nullary op
• If performance is reasonable, switch box_syntax to use Box terminator instead of Box nullary op
• Remove Box as a nullary op.

Mentors or Reviewers

If you have a reviewer or mentor in mind for this work, mention then
here. You can put your own name here if you are planning to mentor the
work.

Process

The main points of the Major Change Process are as follows:

• File an issue describing the proposal.
• A compiler team member or contributor who is knowledgeable in the area can second by writing @rustbot second.
  • Finding a "second" suffices for internal changes. If however, you are proposing a new public-facing feature, such as a -C flag, then full team check-off is required.
  • Compiler team members can initiate a check-off via @rfcbot fcp merge on either the MCP or the PR.
• Once an MCP is seconded, the Final Comment Period begins. If no objections are raised after 10 days, the MCP is considered approved.

You can read more about Major Change Proposals on forge.

Comments

This issue is not meant to be used for technical discussion. There is a Zulip stream for that. Use this issue to leave procedural comments, such as volunteering to review, indicating that you second the proposal (or third, etc), or raising a concern that you would like to be addressed.

[rustbot]

This issue is not meant to be used for technical discussion. There is a Zulip stream for that. Use this issue to leave procedural comments, such as volunteering to review, indicating that you second the proposal (or third, etc), or raising a concern that you would like to be addressed.

cc @rust-lang/compiler @rust-lang/compiler-contributors

[jonas-schievink]

Seems like the correct way to move this forward.

@rustbot second

[apiraino]

@rustbot label -final-comment-period +major-change-accepted