fix: resolve doc path from parent module if outer comments exist on module

[Hmikihiro]

fix: #19506
#19471

I set placement information about inner or outer in attribute.
When I check link on module, I search from parent module if attribute of doc item have a outer comment.

[Veykril]

This grows attributes by 8 bytes which will be noticable, we'll need to avoid that somehow if we can. Note quite sure how yet. Probably by using one of the AttrId bits (we already do something there for cfg_attr but that is not really used so we can replace that with this likely). Will write oiut better instructions later

[Hmikihiro]

rust-analyzer/crates/hir-expand/src/attrs.rs

Lines 185 to 189 in 588948f

	impl AttrId {
	const CFG_ATTR_BITS: usize = 7;
	const AST_INDEX_MASK: usize = 0x00FF_FFFF;
	const AST_INDEX_BITS: usize = Self::AST_INDEX_MASK.count_ones() as usize;
	const CFG_ATTR_SET_BITS: u32 = 1 << 31;

Right, that means AttrId currently uses bits like this:

• Bit 31 → is_cfg_attr
• Bits 24–30 → cfg_attr_index (7 bits)
• Bits 0–23 → ast_index (24 bits)

So, we could either remove or reduce the size of cfg_attr_index, and reuse the freed bit for placement information.

• Bit 31 → inner_or_outer
• Bits 24–30 → (unused)or (is_cfg_attr & cfg_attr_index(6bit))
• Bits 0–23 → ast_index

[Hmikihiro]

rust-analyzer/crates/hir-expand/src/attrs.rs

Lines 108 to 110 in 588948f

	let mut it = it.clone();
	it.id.id = (it.id.ast_index() as u32 + last_ast_index)
	| ((it.id.cfg_attr_index().unwrap_or(0) as u32)

And cfg_attr was accessed through cfg_attr_index() but this function was only used to copy it when creating a new [Attr]. So would we remove cfg_attr?

[Veykril]

IF the cfg attr stuff isnt actually used then lets remove it yes. Then just use the most significant bit for encoding the "inner-ness"

[Hmikihiro]

I see. The inner-ness uses the most significant bit, and bits 24–30 are currently unused.

[Veykril]

We don't need the mask anymore, we just need to make sure to strip the inner attribute bit when using the index as an actual index

[Hmikihiro]

I see. We don't need to clear 24-30bit. AST_INDEX uses 0-30bit.

78cce17

[Veykril]

This should be a parameter of resolve_doc_path_on as it depends on the attribute the link was usually fetched from. A module for example can very likely have both inner and outer attributes mixed

[Hmikihiro]

fix commit 778322e

I see. In old commits, if both inner and outer attributes mixed, those attributes was resolved as outer attributes.

Instead to resolve inner-ness in resolve_doc_path_on, this resolves inner-ness each attributes.

rust-analyzer/crates/ide-db/src/documentation.rs

Line 58 in 778322e

pub fn map(&self, range: TextRange) -> Option<(InFile<TextRange>, AttrId)> {

rust-analyzer/crates/ide/src/doc_links.rs

Lines 333 to 342 in 778322e

	let (in_expansion_range, link, ns, is_inner) =
	extract_definitions_from_docs(&docs).into_iter().find_map(|(range, link, ns)| {
	let (mapped, idx) = doc_mapping.map(range)?;
	(mapped.value.contains(abs_in_expansion_offset)).then_some((mapped.value, link, ns, idx.is_inner_attr()))
	})?;
	// get the relative range to the doc/attribute in the expansion
	let in_expansion_relative_range = in_expansion_range - descended_prefix_len - token_start;
	// Apply relative range to the original input comment
	let absolute_range = in_expansion_relative_range + original_start + prefix_len;
	let def = resolve_doc_path_for_def(sema.db, def, &link, ns, is_inner)?;

[Veykril]

One small thing, otherwise this looks good to me!

[Veykril]

Suggested change

	Self {
	id: if is_inner {
	id | Self::INNER_ATTR_SET_BIT
	} else {
	id & !Self::INNER_ATTR_SET_BIT
	} as u32,
	assert!(id <= !Self::INNER_ATTR_SET_BIT);
	let id = as u32;
	Self {
	id: if is_inner {
	id | Self::INNER_ATTR_SET_BIT
	} else {
	id
	}

[Hmikihiro]

fix commit 7c7d440

rust-analyzer/crates/hir-expand/src/attrs.rs

Lines 180 to 184 in 7c7d440

	pub fn new(id: usize, is_inner: bool) -> Self {
	assert!(id <= !Self::INNER_ATTR_SET_BIT as usize);
	let id = id as u32;
	Self { id: if is_inner { id | Self::INNER_ATTR_SET_BIT } else { id } }
	}

[Veykril]

Thanks!