1 //! An "interner" is a data structure that associates values with usize tags and
2 //! allows bidirectional lookup; i.e., given a value, one can easily find the
3 //! type, and vice versa.
5 use arena::DroplessArena;
6 use rustc_data_structures::fx::FxHashMap;
7 use rustc_index::vec::Idx;
8 use rustc_macros::{symbols, HashStable_Generic};
9 use rustc_serialize::{Decodable, Decoder, Encodable, Encoder};
10 use rustc_serialize::{UseSpecializedDecodable, UseSpecializedEncodable};
11 use rustc_data_structures::stable_hasher::{HashStable, ToStableHashKey, StableHasher};
13 use std::cmp::{PartialEq, PartialOrd, Ord};
15 use std::hash::{Hash, Hasher};
18 use crate::{Span, DUMMY_SP, GLOBALS};
24 // After modifying this list adjust `is_special`, `is_used_keyword`/`is_unused_keyword`,
25 // this should be rarely necessary though if the keywords are kept in alphabetic order.
27 // Special reserved identifiers used internally for elided lifetimes,
28 // unnamed method parameters, crate root module, error recovery etc.
31 DollarCrate: "$crate",
34 // Keywords that are used in stable Rust.
71 // Keywords that are used in unstable Rust or reserved for future use.
85 // Edition-specific keywords that are used in stable Rust.
86 Async: "async", // >= 2018 Edition only
87 Await: "await", // >= 2018 Edition only
88 Dyn: "dyn", // >= 2018 Edition only
90 // Edition-specific keywords that are used in unstable Rust or reserved for future use.
91 Try: "try", // >= 2018 Edition only
93 // Special lifetime names
94 UnderscoreLifetime: "'_",
95 StaticLifetime: "'static",
97 // Weak keywords, have special meaning only in specific contexts.
105 // Symbols that can be referred to with syntax_pos::sym::*. The symbol is
106 // the stringified identifier unless otherwise specified (e.g.
107 // `proc_dash_macro` represents "proc-macro").
109 // As well as the symbols listed, there are symbols for the the strings
110 // "0", "1", ..., "9", which are accessible via `sym::integer`.
112 aarch64_target_feature,
116 abi_msp430_interrupt,
125 advanced_slice_patterns,
137 allow_internal_unsafe,
138 allow_internal_unstable,
139 allow_internal_unstable_backcompat_hack,
143 arbitrary_enum_discriminant,
144 arbitrary_self_types,
151 associated_type_bounds,
152 associated_type_defaults,
160 augmented_assignments,
161 automatically_derived,
162 avx512_target_feature,
167 bind_by_move_pattern_guards,
170 borrowck_graphviz_postflow,
171 borrowck_graphviz_preflow,
174 braced_empty_structs,
186 cfg_target_has_atomic,
187 cfg_target_thread_local,
195 closure_to_fn_coercion,
197 cmpxchg16b_target_feature,
204 conservative_impl_trait,
206 const_compare_raw_pointers,
214 const_in_array_repeat_expressions,
220 const_raw_ptr_to_usize_cast,
234 crate_visibility_modifier,
242 custom_inner_attributes,
243 custom_test_frameworks,
251 default_lib_allocator,
252 default_type_parameter_fallback,
254 delay_span_bug_from_inside_query,
269 document_private_items,
270 dotdoteq_in_patterns,
271 dotdot_in_tuple_patterns,
272 double_braced_crate: "{{crate}}",
273 double_braced_impl: "{{impl}}",
274 double_braced_misc: "{{misc}}",
275 double_braced_closure: "{{closure}}",
276 double_braced_constructor: "{{constructor}}",
277 double_braced_constant: "{{constant}}",
278 double_braced_opaque: "{{opaque}}",
280 dropck_parametricity,
296 exclusive_range_pattern,
297 exhaustive_integer_patterns,
303 extern_absolute_paths,
305 extern_crate_item_prelude,
316 field_init_shorthand,
339 generic_associated_types,
348 hexagon_target_feature,
350 homogeneous_aggregate,
364 if_while_or_patterns,
366 impl_header_lifetime_elision,
368 impl_trait_in_bindings,
376 inclusive_range_syntax,
377 infer_outlives_requirements,
378 infer_static_outlives_requirements,
385 irrefutable_let_patterns,
388 issue_5723_bootstrap,
389 issue_tracker_base_url,
391 item_context: "ItemContext",
410 link_llvm_intrinsics,
420 macro_at_most_once_rep,
423 macro_lifetime_matcher,
424 macro_literal_matcher,
435 match_beginning_vert,
436 match_default_bindings,
454 movbe_target_feature,
490 object_safe_for_dispatch,
492 omit_gdb_pretty_printer_section,
499 optin_builtin_traits,
509 overlapping_marker_traits,
514 panic_implementation,
534 poll_with_tls_context,
535 powerpc_target_feature,
536 precise_pointer_size_matching,
541 proc_dash_macro: "proc-macro",
543 proc_macro_attribute,
549 proc_macro_internals,
551 proc_macro_non_items,
552 proc_macro_path_invoc,
557 quad_precision_float,
572 reexport_test_harness_main,
584 re_rebalance_coherence,
602 rustc_allocator_nounwind,
603 rustc_allow_const_fn_ptr,
604 rustc_args_required_const,
608 rustc_const_unstable,
610 rustc_conversion_suggestion,
613 rustc_diagnostic_item,
614 rustc_diagnostic_macros,
617 rustc_dump_env_program_clauses,
618 rustc_dump_program_clauses,
619 rustc_dump_user_substs,
621 rustc_expected_cgu_reuse,
622 rustc_if_this_changed,
623 rustc_inherit_overflow_checks,
625 rustc_layout_scalar_valid_range_end,
626 rustc_layout_scalar_valid_range_start,
627 rustc_macro_transparency,
629 rustc_nonnull_optimization_guaranteed,
630 rustc_object_lifetime_default,
631 rustc_on_unimplemented,
634 rustc_partition_codegened,
635 rustc_partition_reused,
637 rustc_peek_definite_init,
638 rustc_peek_maybe_init,
639 rustc_peek_maybe_uninit,
640 rustc_peek_indirectly_mutable,
642 rustc_proc_macro_decls,
646 rustc_std_internal_symbol,
649 rustc_reservation_impl,
651 rustc_then_this_would_need,
655 rust_eh_unwind_resume,
657 rvalue_static_promotion,
681 sse4a_target_feature,
694 stmt_expr_attributes,
696 struct_field_attributes,
706 target_has_atomic_load_store,
711 termination_trait_test,
714 test_accepted_feature,
716 test_removed_feature,
737 type_alias_impl_trait,
742 type_alias_enum_variants,
754 underscore_const_names,
756 underscore_lifetimes,
758 universal_impl_trait,
761 unrestricted_attribute_tokens,
764 unsized_tuple_coercion,
779 visible_private_types,
794 #[derive(Copy, Clone, Eq, HashStable_Generic)]
802 /// Constructs a new identifier from a symbol and a span.
803 pub const fn new(name: Symbol, span: Span) -> Ident {
807 /// Constructs a new identifier with a dummy span.
809 pub const fn with_dummy_span(name: Symbol) -> Ident {
810 Ident::new(name, DUMMY_SP)
814 pub fn invalid() -> Ident {
815 Ident::with_dummy_span(kw::Invalid)
818 /// Maps a string to an identifier with a dummy span.
819 pub fn from_str(string: &str) -> Ident {
820 Ident::with_dummy_span(Symbol::intern(string))
823 /// Maps a string and a span to an identifier.
824 pub fn from_str_and_span(string: &str, span: Span) -> Ident {
825 Ident::new(Symbol::intern(string), span)
828 /// Replaces `lo` and `hi` with those from `span`, but keep hygiene context.
829 pub fn with_span_pos(self, span: Span) -> Ident {
830 Ident::new(self.name, span.with_ctxt(self.span.ctxt()))
833 pub fn without_first_quote(self) -> Ident {
834 Ident::new(Symbol::intern(self.as_str().trim_start_matches('\'')), self.span)
837 /// "Normalize" ident for use in comparisons using "item hygiene".
838 /// Identifiers with same string value become same if they came from the same "modern" macro
839 /// (e.g., `macro` item, but not `macro_rules` item) and stay different if they came from
840 /// different "modern" macros.
841 /// Technically, this operation strips all non-opaque marks from ident's syntactic context.
842 pub fn modern(self) -> Ident {
843 Ident::new(self.name, self.span.modern())
846 /// "Normalize" ident for use in comparisons using "local variable hygiene".
847 /// Identifiers with same string value become same if they came from the same non-transparent
848 /// macro (e.g., `macro` or `macro_rules!` items) and stay different if they came from different
849 /// non-transparent macros.
850 /// Technically, this operation strips all transparent marks from ident's syntactic context.
851 pub fn modern_and_legacy(self) -> Ident {
852 Ident::new(self.name, self.span.modern_and_legacy())
855 /// Convert the name to a `SymbolStr`. This is a slowish operation because
856 /// it requires locking the symbol interner.
857 pub fn as_str(self) -> SymbolStr {
862 impl PartialEq for Ident {
863 fn eq(&self, rhs: &Self) -> bool {
864 self.name == rhs.name && self.span.ctxt() == rhs.span.ctxt()
868 impl Hash for Ident {
869 fn hash<H: Hasher>(&self, state: &mut H) {
870 self.name.hash(state);
871 self.span.ctxt().hash(state);
875 impl fmt::Debug for Ident {
876 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
877 if self.is_raw_guess() {
880 write!(f, "{}{:?}", self.name, self.span.ctxt())
884 impl fmt::Display for Ident {
885 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
886 if self.is_raw_guess() {
889 fmt::Display::fmt(&self.name, f)
893 impl UseSpecializedEncodable for Ident {
894 fn default_encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
895 s.emit_struct("Ident", 2, |s| {
896 s.emit_struct_field("name", 0, |s| {
899 s.emit_struct_field("span", 1, |s| {
906 impl UseSpecializedDecodable for Ident {
907 fn default_decode<D: Decoder>(d: &mut D) -> Result<Self, D::Error> {
908 d.read_struct("Ident", 2, |d| {
910 name: d.read_struct_field("name", 0, Decodable::decode)?,
911 span: d.read_struct_field("span", 1, Decodable::decode)?,
917 /// An interned string.
919 /// Internally, a `Symbol` is implemented as an index, and all operations
920 /// (including hashing, equality, and ordering) operate on that index. The use
921 /// of `rustc_index::newtype_index!` means that `Option<Symbol>` only takes up 4 bytes,
922 /// because `rustc_index::newtype_index!` reserves the last 256 values for tagging purposes.
924 /// Note that `Symbol` cannot directly be a `rustc_index::newtype_index!` because it
925 /// implements `fmt::Debug`, `Encodable`, and `Decodable` in special ways.
926 #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
927 pub struct Symbol(SymbolIndex);
929 rustc_index::newtype_index! {
930 pub struct SymbolIndex { .. }
934 const fn new(n: u32) -> Self {
935 Symbol(SymbolIndex::from_u32_const(n))
938 /// Maps a string to its interned representation.
939 pub fn intern(string: &str) -> Self {
940 with_interner(|interner| interner.intern(string))
943 /// Access the symbol's chars. This is a slowish operation because it
944 /// requires locking the symbol interner.
945 pub fn with<F: FnOnce(&str) -> R, R>(self, f: F) -> R {
946 with_interner(|interner| {
947 f(interner.get(self))
951 /// Convert to a `SymbolStr`. This is a slowish operation because it
952 /// requires locking the symbol interner.
953 pub fn as_str(self) -> SymbolStr {
954 with_interner(|interner| unsafe {
956 string: std::mem::transmute::<&str, &str>(interner.get(self))
961 pub fn as_u32(self) -> u32 {
966 impl fmt::Debug for Symbol {
967 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
968 self.with(|str| fmt::Debug::fmt(&str, f))
972 impl fmt::Display for Symbol {
973 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
974 self.with(|str| fmt::Display::fmt(&str, f))
978 impl Encodable for Symbol {
979 fn encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
980 self.with(|string| s.emit_str(string))
984 impl Decodable for Symbol {
985 fn decode<D: Decoder>(d: &mut D) -> Result<Symbol, D::Error> {
986 Ok(Symbol::intern(&d.read_str()?))
990 impl<CTX> HashStable<CTX> for Symbol {
992 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
993 self.as_str().hash_stable(hcx, hasher);
997 impl<CTX> ToStableHashKey<CTX> for Symbol {
998 type KeyType = SymbolStr;
1001 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {
1006 // The `&'static str`s in this type actually point into the arena.
1008 pub struct Interner {
1009 arena: DroplessArena,
1010 names: FxHashMap<&'static str, Symbol>,
1011 strings: Vec<&'static str>,
1015 fn prefill(init: &[&'static str]) -> Self {
1017 strings: init.into(),
1018 names: init.iter().copied().zip((0..).map(Symbol::new)).collect(),
1019 ..Default::default()
1023 pub fn intern(&mut self, string: &str) -> Symbol {
1024 if let Some(&name) = self.names.get(string) {
1028 let name = Symbol::new(self.strings.len() as u32);
1030 // `from_utf8_unchecked` is safe since we just allocated a `&str` which is known to be
1032 let string: &str = unsafe {
1033 str::from_utf8_unchecked(self.arena.alloc_slice(string.as_bytes()))
1035 // It is safe to extend the arena allocation to `'static` because we only access
1036 // these while the arena is still alive.
1037 let string: &'static str = unsafe {
1038 &*(string as *const str)
1040 self.strings.push(string);
1041 self.names.insert(string, name);
1045 // Get the symbol as a string. `Symbol::as_str()` should be used in
1046 // preference to this function.
1047 pub fn get(&self, symbol: Symbol) -> &str {
1048 self.strings[symbol.0.as_usize()]
1052 // This module has a very short name because it's used a lot.
1058 // This module has a very short name because it's used a lot.
1060 use std::convert::TryInto;
1065 // Get the symbol for an integer. The first few non-negative integers each
1066 // have a static symbol and therefore are fast.
1067 pub fn integer<N: TryInto<usize> + Copy + ToString>(n: N) -> Symbol {
1068 if let Result::Ok(idx) = n.try_into() {
1069 if let Option::Some(&sym) = digits_array.get(idx) {
1073 Symbol::intern(&n.to_string())
1078 fn is_used_keyword_2018(self) -> bool {
1079 self >= kw::Async && self <= kw::Dyn
1082 fn is_unused_keyword_2018(self) -> bool {
1086 /// Used for sanity checking rustdoc keyword sections.
1087 pub fn is_doc_keyword(self) -> bool {
1091 /// A keyword or reserved identifier that can be used as a path segment.
1092 pub fn is_path_segment_keyword(self) -> bool {
1093 self == kw::Super ||
1094 self == kw::SelfLower ||
1095 self == kw::SelfUpper ||
1096 self == kw::Crate ||
1097 self == kw::PathRoot ||
1098 self == kw::DollarCrate
1101 /// Returns `true` if the symbol is `true` or `false`.
1102 pub fn is_bool_lit(self) -> bool {
1103 self == kw::True || self == kw::False
1106 /// This symbol can be a raw identifier.
1107 pub fn can_be_raw(self) -> bool {
1108 self != kw::Invalid && self != kw::Underscore && !self.is_path_segment_keyword()
1113 // Returns `true` for reserved identifiers used internally for elided lifetimes,
1114 // unnamed method parameters, crate root module, error recovery etc.
1115 pub fn is_special(self) -> bool {
1116 self.name <= kw::Underscore
1119 /// Returns `true` if the token is a keyword used in the language.
1120 pub fn is_used_keyword(self) -> bool {
1121 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1122 self.name >= kw::As && self.name <= kw::While ||
1123 self.name.is_used_keyword_2018() && self.span.rust_2018()
1126 /// Returns `true` if the token is a keyword reserved for possible future use.
1127 pub fn is_unused_keyword(self) -> bool {
1128 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1129 self.name >= kw::Abstract && self.name <= kw::Yield ||
1130 self.name.is_unused_keyword_2018() && self.span.rust_2018()
1133 /// Returns `true` if the token is either a special identifier or a keyword.
1134 pub fn is_reserved(self) -> bool {
1135 self.is_special() || self.is_used_keyword() || self.is_unused_keyword()
1138 /// A keyword or reserved identifier that can be used as a path segment.
1139 pub fn is_path_segment_keyword(self) -> bool {
1140 self.name.is_path_segment_keyword()
1143 /// We see this identifier in a normal identifier position, like variable name or a type.
1144 /// How was it written originally? Did it use the raw form? Let's try to guess.
1145 pub fn is_raw_guess(self) -> bool {
1146 self.name.can_be_raw() && self.is_reserved()
1151 fn with_interner<T, F: FnOnce(&mut Interner) -> T>(f: F) -> T {
1152 GLOBALS.with(|globals| f(&mut *globals.symbol_interner.lock()))
1155 /// An alternative to `Symbol`, useful when the chars within the symbol need to
1156 /// be accessed. It deliberately has limited functionality and should only be
1157 /// used for temporary values.
1159 /// Because the interner outlives any thread which uses this type, we can
1160 /// safely treat `string` which points to interner data, as an immortal string,
1161 /// as long as this type never crosses between threads.
1163 // FIXME: ensure that the interner outlives any thread which uses `SymbolStr`,
1164 // by creating a new thread right after constructing the interner.
1165 #[derive(Clone, Eq, PartialOrd, Ord)]
1166 pub struct SymbolStr {
1167 string: &'static str,
1170 // This impl allows a `SymbolStr` to be directly equated with a `String` or
1172 impl<T: std::ops::Deref<Target = str>> std::cmp::PartialEq<T> for SymbolStr {
1173 fn eq(&self, other: &T) -> bool {
1174 self.string == other.deref()
1178 impl !Send for SymbolStr {}
1179 impl !Sync for SymbolStr {}
1181 /// This impl means that if `ss` is a `SymbolStr`:
1182 /// - `*ss` is a `str`;
1183 /// - `&*ss` is a `&str`;
1184 /// - `&ss as &str` is a `&str`, which means that `&ss` can be passed to a
1185 /// function expecting a `&str`.
1186 impl std::ops::Deref for SymbolStr {
1189 fn deref(&self) -> &str { self.string }
1192 impl fmt::Debug for SymbolStr {
1193 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1194 fmt::Debug::fmt(self.string, f)
1198 impl fmt::Display for SymbolStr {
1199 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1200 fmt::Display::fmt(self.string, f)
1204 impl<CTX> HashStable<CTX> for SymbolStr {
1206 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
1207 self.string.hash_stable(hcx, hasher)
1211 impl<CTX> ToStableHashKey<CTX> for SymbolStr {
1212 type KeyType = SymbolStr;
1215 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {