rust/crates/ra_hir/src/ids.rs

339 lines
11 KiB
Rust
Raw Normal View History

use ra_db::{LocationIntener, FileId};
2019-01-23 17:49:11 +01:00
use ra_syntax::{TreeArc, SyntaxNode, SourceFile, AstNode, ast};
2019-01-04 14:15:50 +01:00
use ra_arena::{Arena, RawId, impl_arena_id};
2019-01-01 21:21:16 +01:00
2019-01-08 13:57:45 +01:00
use crate::{
2019-01-23 17:49:11 +01:00
HirDatabase, Def, Function, Struct, Enum, EnumVariant, ImplBlock, Crate,
2019-01-11 19:02:12 +01:00
Module, Trait, Type, Static, Const,
2019-01-08 13:57:45 +01:00
};
2019-01-06 13:16:21 +01:00
2019-01-24 10:41:08 +01:00
#[derive(Debug, Default)]
pub struct HirInterner {
defs: LocationIntener<DefLoc, DefId>,
macros: LocationIntener<MacroCallLoc, MacroCallId>,
}
impl HirInterner {
pub fn len(&self) -> usize {
self.defs.len() + self.macros.len()
}
}
/// hir makes heavy use of ids: integer (u32) handlers to various things. You
2019-01-01 20:47:10 +01:00
/// can think of id as a pointer (but without a lifetime) or a file descriptor
/// (but for hir objects).
///
/// This module defines a bunch of ids we are using. The most important ones are
/// probably `HirFileId` and `DefId`.
/// Input to the analyzer is a set of files, where each file is indentified by
2019-01-01 20:47:10 +01:00
/// `FileId` and contains source code. However, another source of source code in
/// Rust are macros: each macro can be thought of as producing a "temporary
/// file". To assign an id to such a file, we use the id of the macro call that
2019-01-01 20:47:10 +01:00
/// produced the file. So, a `HirFileId` is either a `FileId` (source code
/// written by user), or a `MacroCallId` (source code produced by macro).
///
/// What is a `MacroCallId`? Simplifying, it's a `HirFileId` of a file containin
/// the call plus the offset of the macro call in the file. Note that this is a
/// recursive definition! However, the size_of of `HirFileId` is finite
2019-01-01 20:47:10 +01:00
/// (because everything bottoms out at the real `FileId`) and small
/// (`MacroCallId` uses the location interner).
2019-01-01 20:47:10 +01:00
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
2019-01-01 21:21:16 +01:00
pub struct HirFileId(HirFileIdRepr);
impl HirFileId {
2019-01-02 14:00:01 +01:00
/// For macro-expansion files, returns the file original source file the
/// expansionoriginated from.
pub fn original_file(self, db: &impl HirDatabase) -> FileId {
2019-01-01 21:21:16 +01:00
match self.0 {
HirFileIdRepr::File(file_id) => file_id,
HirFileIdRepr::Macro(macro_call_id) => {
let loc = macro_call_id.loc(db);
2019-01-02 14:00:35 +01:00
loc.source_item_id.file_id.original_file(db)
2019-01-01 21:21:16 +01:00
}
}
}
pub(crate) fn as_original_file(self) -> FileId {
match self.0 {
HirFileIdRepr::File(file_id) => file_id,
HirFileIdRepr::Macro(_r) => panic!("macro generated file: {:?}", self),
}
}
2019-01-02 14:00:01 +01:00
2019-01-03 19:28:35 +01:00
pub(crate) fn as_macro_call_id(self) -> Option<MacroCallId> {
match self.0 {
HirFileIdRepr::Macro(it) => Some(it),
_ => None,
}
}
2019-01-08 09:28:42 +01:00
pub(crate) fn hir_source_file(
db: &impl HirDatabase,
file_id: HirFileId,
) -> TreeArc<SourceFile> {
2019-01-01 21:21:16 +01:00
match file_id.0 {
HirFileIdRepr::File(file_id) => db.source_file(file_id),
HirFileIdRepr::Macro(m) => {
if let Some(exp) = db.expand_macro_invocation(m) {
return exp.file();
}
// returning an empty string looks fishy...
2019-01-08 09:28:42 +01:00
SourceFile::parse("")
2019-01-01 21:21:16 +01:00
}
}
}
}
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
enum HirFileIdRepr {
2019-01-01 20:47:10 +01:00
File(FileId),
Macro(MacroCallId),
}
2019-01-01 21:21:16 +01:00
impl From<FileId> for HirFileId {
fn from(file_id: FileId) -> HirFileId {
HirFileId(HirFileIdRepr::File(file_id))
}
}
impl From<MacroCallId> for HirFileId {
fn from(macro_call_id: MacroCallId) -> HirFileId {
HirFileId(HirFileIdRepr::Macro(macro_call_id))
2019-01-01 20:47:10 +01:00
}
}
2019-01-01 22:30:00 +01:00
2019-01-01 22:37:36 +01:00
/// `MacroCallId` identifies a particular macro invocation, like
/// `println!("Hello, {}", world)`.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
2019-01-08 13:53:32 +01:00
pub struct MacroCallId(RawId);
impl_arena_id!(MacroCallId);
2019-01-01 22:37:36 +01:00
#[derive(Debug, Clone, PartialEq, Eq, Hash)]
2019-01-01 22:37:36 +01:00
pub struct MacroCallLoc {
pub(crate) module: Module,
2019-01-01 22:37:36 +01:00
pub(crate) source_item_id: SourceItemId,
}
impl MacroCallId {
2019-01-24 10:41:08 +01:00
pub(crate) fn loc(self, db: &impl AsRef<HirInterner>) -> MacroCallLoc {
db.as_ref().macros.id2loc(self)
2019-01-01 22:37:36 +01:00
}
}
impl MacroCallLoc {
#[allow(unused)]
2019-01-24 10:41:08 +01:00
pub(crate) fn id(&self, db: &impl AsRef<HirInterner>) -> MacroCallId {
db.as_ref().macros.loc2id(&self)
2019-01-01 22:37:36 +01:00
}
}
2019-01-01 22:30:00 +01:00
/// Def's are a core concept of hir. A `Def` is an Item (function, module, etc)
/// in a specific module.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
2019-01-08 13:53:32 +01:00
pub struct DefId(RawId);
impl_arena_id!(DefId);
2019-01-01 22:30:00 +01:00
#[derive(Clone, Debug, PartialEq, Eq, Hash)]
pub struct DefLoc {
pub(crate) kind: DefKind,
pub(crate) module: Module,
2019-01-01 22:30:00 +01:00
pub(crate) source_item_id: SourceItemId,
}
#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
pub(crate) enum DefKind {
Function,
Struct,
Enum,
EnumVariant,
2019-01-11 19:02:12 +01:00
Const,
Static,
Trait,
Type,
2019-01-01 22:30:00 +01:00
Item,
/// The constructor of a struct. E.g. if we have `struct Foo(usize)`, the
/// name `Foo` needs to resolve to different types depending on whether we
/// are in the types or values namespace: As a type, `Foo` of course refers
/// to the struct `Foo`; as a value, `Foo` is a callable type with signature
/// `(usize) -> Foo`. The cleanest approach to handle this seems to be to
/// have different defs in the two namespaces.
///
/// rustc does the same; note that it even creates a struct constructor if
/// the struct isn't a tuple struct (see `CtorKind::Fictive` in rustc).
2019-01-01 22:30:00 +01:00
StructCtor,
}
impl DefId {
2019-01-24 10:41:08 +01:00
pub(crate) fn loc(self, db: &impl AsRef<HirInterner>) -> DefLoc {
db.as_ref().defs.id2loc(self)
2019-01-01 22:30:00 +01:00
}
pub fn resolve(self, db: &impl HirDatabase) -> Def {
2019-01-01 22:30:00 +01:00
let loc = self.loc(db);
match loc.kind {
2019-01-01 22:30:00 +01:00
DefKind::Function => {
let function = Function::new(self);
Def::Function(function)
}
DefKind::Struct => {
let struct_def = Struct::new(self);
Def::Struct(struct_def)
}
DefKind::Enum => Def::Enum(Enum::new(self)),
DefKind::EnumVariant => Def::EnumVariant(EnumVariant::new(self)),
2019-01-11 19:02:12 +01:00
DefKind::Const => {
let def = Const::new(self);
Def::Const(def)
}
DefKind::Static => {
let def = Static::new(self);
Def::Static(def)
}
DefKind::Trait => {
let def = Trait::new(self);
Def::Trait(def)
}
DefKind::Type => {
let def = Type::new(self);
Def::Type(def)
}
2019-01-01 22:30:00 +01:00
DefKind::StructCtor => Def::Item,
DefKind::Item => Def::Item,
}
2019-01-01 22:30:00 +01:00
}
pub(crate) fn source(self, db: &impl HirDatabase) -> (HirFileId, TreeArc<SyntaxNode>) {
let loc = self.loc(db);
let syntax = db.file_item(loc.source_item_id);
(loc.source_item_id.file_id, syntax)
}
2019-01-01 22:30:00 +01:00
/// For a module, returns that module; for any other def, returns the containing module.
2019-01-15 17:18:52 +01:00
pub fn module(self, db: &impl HirDatabase) -> Module {
self.loc(db).module
2019-01-01 22:30:00 +01:00
}
/// Returns the containing crate.
2019-01-15 17:18:52 +01:00
pub fn krate(&self, db: &impl HirDatabase) -> Option<Crate> {
self.module(db).krate(db)
}
/// Returns the containing impl block, if this is an impl item.
2019-01-15 17:18:52 +01:00
pub fn impl_block(self, db: &impl HirDatabase) -> Option<ImplBlock> {
let module_impls = db.impls_in_module(self.loc(db).module);
2019-01-15 17:18:52 +01:00
ImplBlock::containing(module_impls, self)
}
2019-01-01 22:30:00 +01:00
}
impl DefLoc {
2019-01-24 10:41:08 +01:00
pub(crate) fn id(&self, db: &impl AsRef<HirInterner>) -> DefId {
db.as_ref().defs.loc2id(&self)
2019-01-01 22:30:00 +01:00
}
}
/// Identifier of item within a specific file. This is stable over reparses, so
/// it's OK to use it as a salsa key/value.
2019-01-04 14:15:50 +01:00
#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
pub struct SourceFileItemId(RawId);
impl_arena_id!(SourceFileItemId);
2019-01-01 22:30:00 +01:00
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
pub struct SourceItemId {
pub(crate) file_id: HirFileId,
/// None for the whole file.
pub(crate) item_id: Option<SourceFileItemId>,
}
/// Maps items' `SyntaxNode`s to `SourceFileItemId`s and back.
2019-01-01 22:30:00 +01:00
#[derive(Debug, PartialEq, Eq)]
pub struct SourceFileItems {
file_id: HirFileId,
arena: Arena<SourceFileItemId, TreeArc<SyntaxNode>>,
2019-01-01 22:30:00 +01:00
}
impl SourceFileItems {
2019-01-08 09:28:42 +01:00
pub(crate) fn new(file_id: HirFileId, source_file: &SourceFile) -> SourceFileItems {
2019-01-01 22:30:00 +01:00
let mut res = SourceFileItems {
file_id,
arena: Arena::default(),
};
res.init(source_file);
res
}
2019-01-08 09:28:42 +01:00
fn init(&mut self, source_file: &SourceFile) {
// By walking the tree in bread-first order we make sure that parents
// get lower ids then children. That is, addding a new child does not
// change parent's id. This means that, say, adding a new function to a
// trait does not chage ids of top-level items, which helps caching.
bfs(source_file.syntax(), |it| {
if let Some(enum_variant) = ast::EnumVariant::cast(it) {
self.alloc(enum_variant.syntax().to_owned());
} else if let Some(module_item) = ast::ModuleItem::cast(it) {
2019-01-08 09:28:42 +01:00
self.alloc(module_item.syntax().to_owned());
2019-01-01 22:30:00 +01:00
} else if let Some(macro_call) = ast::MacroCall::cast(it) {
2019-01-08 09:28:42 +01:00
self.alloc(macro_call.syntax().to_owned());
2019-01-01 22:30:00 +01:00
}
})
2019-01-01 22:30:00 +01:00
}
fn alloc(&mut self, item: TreeArc<SyntaxNode>) -> SourceFileItemId {
2019-01-01 22:30:00 +01:00
self.arena.alloc(item)
}
2019-01-08 09:28:42 +01:00
pub(crate) fn id_of(&self, file_id: HirFileId, item: &SyntaxNode) -> SourceFileItemId {
2019-01-01 22:30:00 +01:00
assert_eq!(
self.file_id, file_id,
"SourceFileItems: wrong file, expected {:?}, got {:?}",
self.file_id, file_id
);
self.id_of_unchecked(item)
}
2019-01-08 09:28:42 +01:00
pub(crate) fn id_of_unchecked(&self, item: &SyntaxNode) -> SourceFileItemId {
if let Some((id, _)) = self.arena.iter().find(|(_id, i)| *i == item) {
2019-01-01 22:30:00 +01:00
return id;
}
// This should not happen. Let's try to give a sensible diagnostics.
if let Some((id, i)) = self.arena.iter().find(|(_id, i)| i.range() == item.range()) {
// FIXME(#288): whyyy are we getting here?
log::error!(
"unequal syntax nodes with the same range:\n{:?}\n{:?}",
item,
i
);
return id;
}
panic!(
"Can't find {:?} in SourceFileItems:\n{:?}",
item,
self.arena.iter().map(|(_id, i)| i).collect::<Vec<_>>(),
);
}
pub fn id_of_source_file(&self) -> SourceFileItemId {
let (id, _syntax) = self.arena.iter().next().unwrap();
id
}
}
impl std::ops::Index<SourceFileItemId> for SourceFileItems {
type Output = SyntaxNode;
fn index(&self, idx: SourceFileItemId) -> &SyntaxNode {
&self.arena[idx]
}
}
/// Walks the subtree in bfs order, calling `f` for each node.
fn bfs(node: &SyntaxNode, mut f: impl FnMut(&SyntaxNode)) {
let mut curr_layer = vec![node];
let mut next_layer = vec![];
while !curr_layer.is_empty() {
curr_layer.drain(..).for_each(|node| {
next_layer.extend(node.children());
f(node);
});
std::mem::swap(&mut curr_layer, &mut next_layer);
}
}