Preserve explicit register names in inline asm diagnostics

[qaijuang]

This PR adds diagnostic-only metadata directly to each register-bearing HIR operand, and replaces #117912 AST lookup with the new HIR metadata.

r? @workingjubilee

Fixes #159409

[rustbot]

workingjubilee is currently at their maximum review capacity.
They may take a while to respond.

[rustbot]

clippy is developed in its own repository. If possible, consider making this change to rust-lang/rust-clippy instead.

cc @rust-lang/clippy

[workingjubilee]

cc @Amanieu

View changes since this review

[workingjubilee]

Is there a reason it shouldn't, instead of this scheme of re-adding it in metadata?

[qaijuang]

I don't see or find any reason why it shouldn't

[workingjubilee]

May I see how the PR looks like in that form instead, then?

[qaijuang]

That would be duplicating the InlineAsmRegOrRegClass type and impl in HiR, is it acceptable?

[qaijuang]

The exact HiR impl

/// Stores explicit register name from source
/// for diagnostics only.
#[derive(Debug, Clone, Copy, StableHash)]
pub enum InlineAsmRegOrRegClass {
    Reg { reg: asm::InlineAsmReg, source_name: Option<Symbol> },
    RegClass(asm::InlineAsmRegClass),
}

impl InlineAsmRegOrRegClass {
    // For `rustc_mir_build` and `clippy_utils`
    pub fn as_target(self) -> asm::InlineAsmRegOrRegClass {
        match self {
            Self::Reg { reg, .. } => asm::InlineAsmRegOrRegClass::Reg(reg),
            Self::RegClass(reg_class) => asm::InlineAsmRegOrRegClass::RegClass(reg_class),
        }
    }

    // For `rustc_ast_lowering`
    pub fn reg_class(self) -> asm::InlineAsmRegClass {
        self.as_target().reg_class()
    }

    // For `rustc_ast_lowering`
    pub fn source_name(self) -> Option<Symbol> {
        match self {
            Self::Reg { source_name, .. } => source_name,
            Self::RegClass(_) => None,
        }
    }
}

// For `rustc_hir_pretty`
impl fmt::Display for InlineAsmRegOrRegClass {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        self.as_target().fmt(f)
    }
}

[Amanieu]

Either variants seems fine to me.

[qaijuang]

@workingjubilee let me know the direction we are taking.

[qaijuang]

@rustbot reroll (reviewer seems busy)

[hanna-kruppe]

@rustbot reroll