HIR and TyCtxt disagree about the parents of decl macros

[jyn514]

I tried this code:

rust/library/core/src/ptr/mod.rs

Lines 1487 to 1492 in 8cc82ee

	#[unstable(feature = "raw_ref_macros", issue = "73394")]
	#[rustc_macro_transparency = "semitransparent"]
	#[allow_internal_unstable(raw_ref_op)]
	pub macro raw_const($e:expr) {
	&raw const $e
	}

I expected to see this happen: tcx.hir().parent_module(raw_const_hir_id) and tcx.parent(raw_const_def_id).unwrap() are both the module core::ptr.

Instead, this happened: tcx.hir().parent_module(raw_const_hir_id) is core and tcx.parent(raw_const_def_id).unwrap() is core::ptr. I think core::ptr is correct.

Meta

rustc --version: HEAD (8cc82ee)

[jyn514]

This is one of the problems causing #74355.

[RalfJung]

Curious, and good catch!

Who would be a HIR expert we could ask about this?

[jyn514]

I don't know. Maybe @ecstatic-morse?

[danielhenrymantilla]

Hmm, this is still the case when the macro is defined inside a value-namespace-inhabiting scope:

const C: () = {
    pub macro m() {} // <- `def.hir_id()`
};

now yields:

[src/librustdoc/visit_ast.rs:86] self.cx.tcx.parent_module(def.hir_id) = DefId(0:0 ~ core[8787])
[src/librustdoc/visit_ast.rs:87] {
    use rustc_middle::ty::DefIdTree;
    self.cx.tcx.parent(self.cx.tcx.hir().local_def_id(def.hir_id).to_def_id()).unwrap()
} = DefId(0:16466 ~ core[8787]::C)

Should a new issue be opened, or this one be re-opened?

[jyn514]

@danielhenrymantilla which case did #80415 fix? Is the bug fixed for things in modules, but not in other contexts?

[danielhenrymantilla]

It looks like it: it definitely fixes it for macros inside modules, and the whole re-export ordeal works. Basically, in my PR, the only tests to fail were those involving a macro defined within a function or within a constant.

[cjgillot]

Hmm, this is still the case when the macro is defined inside a value-namespace-inhabiting scope:

const C: () = {
    pub macro m() {} // <- `def.hir_id()`
};

now yields:

[src/librustdoc/visit_ast.rs:86] self.cx.tcx.parent_module(def.hir_id) = DefId(0:0 ~ core[8787])
[src/librustdoc/visit_ast.rs:87] {
    use rustc_middle::ty::DefIdTree;
    self.cx.tcx.parent(self.cx.tcx.hir().local_def_id(def.hir_id).to_def_id()).unwrap()
} = DefId(0:16466 ~ core[8787]::C)

Should a new issue be opened, or this one be re-opened?

I don't think this is an issue any more.
The second output core[8787]::C is not a module, so parent_module will not return it.
#80415 allows to compute the correct parent module (ie. not crate root) when the macro is inside a module.

[danielhenrymantilla]

Interesting, so it's intended that .parent() and .parent_module() not always agree. I think we can close this again (and maybe submit a PR clarifying that in the documentation), and I'll adapt my PR accordingly 🙂