1 use crate::{HashStableContext, Symbol};
2 use rustc_data_structures::fingerprint::Fingerprint;
3 use rustc_data_structures::stable_hasher::{HashStable, StableHasher, ToStableHashKey};
4 use rustc_data_structures::AtomicRef;
5 use rustc_index::vec::Idx;
6 use rustc_macros::HashStable_Generic;
7 use rustc_serialize::{Decodable, Decoder, Encodable, Encoder};
8 use std::borrow::Borrow;
10 use std::hash::{Hash, Hasher};
12 rustc_index::newtype_index! {
14 #[debug_format = "crate{}"]
15 pub struct CrateNum {}
18 /// Item definitions in the currently-compiled crate would have the `CrateNum`
19 /// `LOCAL_CRATE` in their `DefId`.
20 pub const LOCAL_CRATE: CrateNum = CrateNum::from_u32(0);
24 pub fn new(x: usize) -> CrateNum {
25 CrateNum::from_usize(x)
29 pub fn as_def_id(self) -> DefId {
30 DefId { krate: self, index: CRATE_DEF_INDEX }
34 impl fmt::Display for CrateNum {
35 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
36 fmt::Display::fmt(&self.as_u32(), f)
40 /// As a local identifier, a `CrateNum` is only meaningful within its context, e.g. within a tcx.
41 /// Therefore, make sure to include the context when encode a `CrateNum`.
42 impl<E: Encoder> Encodable<E> for CrateNum {
43 default fn encode(&self, s: &mut E) {
44 s.emit_u32(self.as_u32());
48 impl<D: Decoder> Decodable<D> for CrateNum {
49 default fn decode(d: &mut D) -> CrateNum {
50 CrateNum::from_u32(d.read_u32())
54 /// A `DefPathHash` is a fixed-size representation of a `DefPath` that is
55 /// stable across crate and compilation session boundaries. It consists of two
56 /// separate 64-bit hashes. The first uniquely identifies the crate this
57 /// `DefPathHash` originates from (see [StableCrateId]), and the second
58 /// uniquely identifies the corresponding `DefPath` within that crate. Together
59 /// they form a unique identifier within an entire crate graph.
61 /// There is a very small chance of hash collisions, which would mean that two
62 /// different `DefPath`s map to the same `DefPathHash`. Proceeding compilation
63 /// with such a hash collision would very probably lead to an ICE, and in the
64 /// worst case lead to a silent mis-compilation. The compiler therefore actively
65 /// and exhaustively checks for such hash collisions and aborts compilation if
68 /// `DefPathHash` uses 64-bit hashes for both the crate-id part and the
69 /// crate-internal part, even though it is likely that there are many more
70 /// `LocalDefId`s in a single crate than there are individual crates in a crate
71 /// graph. Since we use the same number of bits in both cases, the collision
72 /// probability for the crate-local part will be quite a bit higher (though
73 /// still very small).
75 /// This imbalance is not by accident: A hash collision in the
76 /// crate-local part of a `DefPathHash` will be detected and reported while
77 /// compiling the crate in question. Such a collision does not depend on
78 /// outside factors and can be easily fixed by the crate maintainer (e.g. by
79 /// renaming the item in question or by bumping the crate version in a harmless
82 /// A collision between crate-id hashes on the other hand is harder to fix
83 /// because it depends on the set of crates in the entire crate graph of a
84 /// compilation session. Again, using the same crate with a different version
85 /// number would fix the issue with a high probability -- but that might be
86 /// easier said then done if the crates in questions are dependencies of
87 /// third-party crates.
89 /// That being said, given a high quality hash function, the collision
90 /// probabilities in question are very small. For example, for a big crate like
91 /// `rustc_middle` (with ~50000 `LocalDefId`s as of the time of writing) there
92 /// is a probability of roughly 1 in 14,750,000,000 of a crate-internal
93 /// collision occurring. For a big crate graph with 1000 crates in it, there is
94 /// a probability of 1 in 36,890,000,000,000 of a `StableCrateId` collision.
95 #[derive(Copy, Clone, Hash, PartialEq, Eq, PartialOrd, Ord, Debug)]
96 #[derive(HashStable_Generic, Encodable, Decodable)]
97 pub struct DefPathHash(pub Fingerprint);
100 /// Returns the [StableCrateId] identifying the crate this [DefPathHash]
103 pub fn stable_crate_id(&self) -> StableCrateId {
104 StableCrateId(self.0.as_value().0)
107 /// Returns the crate-local part of the [DefPathHash].
111 pub fn local_hash(&self) -> u64 {
115 /// Builds a new [DefPathHash] with the given [StableCrateId] and
116 /// `local_hash`, where `local_hash` must be unique within its crate.
117 pub fn new(stable_crate_id: StableCrateId, local_hash: u64) -> DefPathHash {
118 DefPathHash(Fingerprint::new(stable_crate_id.0, local_hash))
122 impl Borrow<Fingerprint> for DefPathHash {
124 fn borrow(&self) -> &Fingerprint {
129 /// A [`StableCrateId`] is a 64-bit hash of a crate name, together with all
130 /// `-Cmetadata` arguments, and some other data. It is to [`CrateNum`] what [`DefPathHash`] is to
131 /// [`DefId`]. It is stable across compilation sessions.
133 /// Since the ID is a hash value, there is a small chance that two crates
134 /// end up with the same [`StableCrateId`]. The compiler will check for such
135 /// collisions when loading crates and abort compilation in order to avoid
138 /// For more information on the possibility of hash collisions in rustc,
139 /// see the discussion in [`DefId`].
140 #[derive(Copy, Clone, Hash, PartialEq, Eq, PartialOrd, Ord, Debug)]
141 #[derive(HashStable_Generic, Encodable, Decodable)]
142 pub struct StableCrateId(pub(crate) u64);
145 pub fn to_u64(self) -> u64 {
149 /// Computes the stable ID for a crate with the given name and
150 /// `-Cmetadata` arguments.
151 pub fn new(crate_name: Symbol, is_exe: bool, mut metadata: Vec<String>) -> StableCrateId {
152 let mut hasher = StableHasher::new();
153 // We must hash the string text of the crate name, not the id, as the id is not stable
155 crate_name.as_str().hash(&mut hasher);
157 // We don't want the stable crate ID to depend on the order of
158 // -C metadata arguments, so sort them:
160 // Every distinct -C metadata value is only incorporated once:
163 hasher.write(b"metadata");
165 // Also incorporate the length of a metadata string, so that we generate
166 // different values for `-Cmetadata=ab -Cmetadata=c` and
167 // `-Cmetadata=a -Cmetadata=bc`
168 hasher.write_usize(s.len());
169 hasher.write(s.as_bytes());
172 // Also incorporate crate type, so that we don't get symbol conflicts when
173 // linking against a library of the same name, if this is an executable.
174 hasher.write(if is_exe { b"exe" } else { b"lib" });
176 // Also incorporate the rustc version. Otherwise, with -Zsymbol-mangling-version=v0
177 // and no -Cmetadata, symbols from the same crate compiled with different versions of
178 // rustc are named the same.
180 // RUSTC_FORCE_RUSTC_VERSION is used to inject rustc version information
182 if let Some(val) = std::env::var_os("RUSTC_FORCE_RUSTC_VERSION") {
183 hasher.write(val.to_string_lossy().into_owned().as_bytes())
185 hasher.write(option_env!("CFG_VERSION").unwrap_or("unknown version").as_bytes());
188 StableCrateId(hasher.finish())
192 rustc_index::newtype_index! {
193 /// A DefIndex is an index into the hir-map for a crate, identifying a
194 /// particular definition. It should really be considered an interned
195 /// shorthand for a particular DefPath.
196 #[custom_encodable] // (only encodable in metadata)
197 #[debug_format = "DefIndex({})"]
198 pub struct DefIndex {
199 /// The crate root is always assigned index 0 by the AST Map code,
200 /// thanks to `NodeCollector::new`.
201 const CRATE_DEF_INDEX = 0;
205 impl<E: Encoder> Encodable<E> for DefIndex {
206 default fn encode(&self, _: &mut E) {
207 panic!("cannot encode `DefIndex` with `{}`", std::any::type_name::<E>());
211 impl<D: Decoder> Decodable<D> for DefIndex {
212 default fn decode(_: &mut D) -> DefIndex {
213 panic!("cannot decode `DefIndex` with `{}`", std::any::type_name::<D>());
217 /// A `DefId` identifies a particular *definition*, by combining a crate
218 /// index and a def index.
220 /// You can create a `DefId` from a `LocalDefId` using `local_def_id.to_def_id()`.
221 #[derive(Clone, PartialEq, Eq, Copy)]
222 // Don't derive order on 64-bit big-endian, so we can be consistent regardless of field order.
223 #[cfg_attr(not(all(target_pointer_width = "64", target_endian = "big")), derive(PartialOrd, Ord))]
224 // On below-64 bit systems we can simply use the derived `Hash` impl
225 #[cfg_attr(not(target_pointer_width = "64"), derive(Hash))]
227 #[rustc_pass_by_value]
228 // We guarantee field order. Note that the order is essential here, see below why.
230 // cfg-ing the order of fields so that the `DefIndex` which is high entropy always ends up in
231 // the lower bits no matter the endianness. This allows the compiler to turn that `Hash` impl
232 // into a direct call to 'u64::hash(_)`.
233 #[cfg(not(all(target_pointer_width = "64", target_endian = "big")))]
236 #[cfg(all(target_pointer_width = "64", target_endian = "big"))]
240 // On 64-bit systems, we can hash the whole `DefId` as one `u64` instead of two `u32`s. This
241 // improves performance without impairing `FxHash` quality. So the below code gets compiled to a
242 // noop on little endian systems because the memory layout of `DefId` is as follows:
245 // +-1--------------31-+-32-------------63-+
247 // +-------------------+-------------------+
250 // The order here has direct impact on `FxHash` quality because we have far more `DefIndex` per
251 // crate than we have `Crate`s within one compilation. Or in other words, this arrangement puts
252 // more entropy in the low bits than the high bits. The reason this matters is that `FxHash`, which
253 // is used throughout rustc, has problems distributing the entropy from the high bits, so reversing
254 // the order would lead to a large number of collisions and thus far worse performance.
256 // On 64-bit big-endian systems, this compiles to a 64-bit rotation by 32 bits, which is still
257 // faster than another `FxHash` round.
258 #[cfg(target_pointer_width = "64")]
259 impl Hash for DefId {
260 fn hash<H: Hasher>(&self, h: &mut H) {
261 (((self.krate.as_u32() as u64) << 32) | (self.index.as_u32() as u64)).hash(h)
265 // Implement the same comparison as derived with the other field order.
266 #[cfg(all(target_pointer_width = "64", target_endian = "big"))]
269 fn cmp(&self, other: &DefId) -> std::cmp::Ordering {
270 Ord::cmp(&(self.index, self.krate), &(other.index, other.krate))
273 #[cfg(all(target_pointer_width = "64", target_endian = "big"))]
274 impl PartialOrd for DefId {
276 fn partial_cmp(&self, other: &DefId) -> Option<std::cmp::Ordering> {
277 Some(self.cmp(other))
282 /// Makes a local `DefId` from the given `DefIndex`.
284 pub fn local(index: DefIndex) -> DefId {
285 DefId { krate: LOCAL_CRATE, index }
288 /// Returns whether the item is defined in the crate currently being compiled.
290 pub fn is_local(self) -> bool {
291 self.krate == LOCAL_CRATE
295 pub fn as_local(self) -> Option<LocalDefId> {
296 if self.is_local() { Some(LocalDefId { local_def_index: self.index }) } else { None }
301 pub fn expect_local(self) -> LocalDefId {
302 // NOTE: `match` below is required to apply `#[track_caller]`,
303 // i.e. don't use closures.
304 match self.as_local() {
305 Some(local_def_id) => local_def_id,
306 None => panic!("DefId::expect_local: `{self:?}` isn't local"),
311 pub fn is_crate_root(self) -> bool {
312 self.index == CRATE_DEF_INDEX
316 pub fn as_crate_root(self) -> Option<CrateNum> {
317 if self.is_crate_root() { Some(self.krate) } else { None }
321 pub fn is_top_level_module(self) -> bool {
322 self.is_local() && self.is_crate_root()
326 impl From<LocalDefId> for DefId {
327 fn from(local: LocalDefId) -> DefId {
332 impl<E: Encoder> Encodable<E> for DefId {
333 default fn encode(&self, s: &mut E) {
334 self.krate.encode(s);
335 self.index.encode(s);
339 impl<D: Decoder> Decodable<D> for DefId {
340 default fn decode(d: &mut D) -> DefId {
341 DefId { krate: Decodable::decode(d), index: Decodable::decode(d) }
345 pub fn default_def_id_debug(def_id: DefId, f: &mut fmt::Formatter<'_>) -> fmt::Result {
346 f.debug_struct("DefId").field("krate", &def_id.krate).field("index", &def_id.index).finish()
349 pub static DEF_ID_DEBUG: AtomicRef<fn(DefId, &mut fmt::Formatter<'_>) -> fmt::Result> =
350 AtomicRef::new(&(default_def_id_debug as fn(_, &mut fmt::Formatter<'_>) -> _));
352 impl fmt::Debug for DefId {
353 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
354 (*DEF_ID_DEBUG)(*self, f)
358 rustc_data_structures::define_id_collections!(DefIdMap, DefIdSet, DefIdMapEntry, DefId);
360 /// A `LocalDefId` is equivalent to a `DefId` with `krate == LOCAL_CRATE`. Since
361 /// we encode this information in the type, we can ensure at compile time that
362 /// no `DefId`s from upstream crates get thrown into the mix. There are quite a
363 /// few cases where we know that only `DefId`s from the local crate are expected;
364 /// a `DefId` from a different crate would signify a bug somewhere. This
365 /// is when `LocalDefId` comes in handy.
366 #[derive(Clone, Copy, PartialEq, Eq, Hash)]
367 pub struct LocalDefId {
368 pub local_def_index: DefIndex,
371 // To ensure correctness of incremental compilation,
372 // `LocalDefId` must not implement `Ord` or `PartialOrd`.
373 // See https://github.com/rust-lang/rust/issues/90317.
374 impl !Ord for LocalDefId {}
375 impl !PartialOrd for LocalDefId {}
377 pub const CRATE_DEF_ID: LocalDefId = LocalDefId { local_def_index: CRATE_DEF_INDEX };
379 impl Idx for LocalDefId {
381 fn new(idx: usize) -> Self {
382 LocalDefId { local_def_index: Idx::new(idx) }
385 fn index(self) -> usize {
386 self.local_def_index.index()
392 pub fn to_def_id(self) -> DefId {
393 DefId { krate: LOCAL_CRATE, index: self.local_def_index }
397 pub fn is_top_level_module(self) -> bool {
402 impl fmt::Debug for LocalDefId {
403 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
404 self.to_def_id().fmt(f)
408 impl<E: Encoder> Encodable<E> for LocalDefId {
409 fn encode(&self, s: &mut E) {
410 self.to_def_id().encode(s);
414 impl<D: Decoder> Decodable<D> for LocalDefId {
415 fn decode(d: &mut D) -> LocalDefId {
416 DefId::decode(d).expect_local()
420 rustc_data_structures::define_id_collections!(
427 impl<CTX: HashStableContext> HashStable<CTX> for DefId {
429 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
430 self.to_stable_hash_key(hcx).hash_stable(hcx, hasher);
434 impl<CTX: HashStableContext> HashStable<CTX> for LocalDefId {
436 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
437 self.to_stable_hash_key(hcx).hash_stable(hcx, hasher);
441 impl<CTX: HashStableContext> HashStable<CTX> for CrateNum {
443 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
444 self.to_stable_hash_key(hcx).hash_stable(hcx, hasher);
448 impl<CTX: HashStableContext> ToStableHashKey<CTX> for DefId {
449 type KeyType = DefPathHash;
452 fn to_stable_hash_key(&self, hcx: &CTX) -> DefPathHash {
453 hcx.def_path_hash(*self)
457 impl<CTX: HashStableContext> ToStableHashKey<CTX> for LocalDefId {
458 type KeyType = DefPathHash;
461 fn to_stable_hash_key(&self, hcx: &CTX) -> DefPathHash {
462 hcx.def_path_hash(self.to_def_id())
466 impl<CTX: HashStableContext> ToStableHashKey<CTX> for CrateNum {
467 type KeyType = DefPathHash;
470 fn to_stable_hash_key(&self, hcx: &CTX) -> DefPathHash {
471 self.as_def_id().to_stable_hash_key(hcx)