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_data_structures::stable_hasher::{HashStable, StableHasher, ToStableHashKey};
8 use rustc_macros::{symbols, HashStable_Generic};
9 use rustc_serialize::{Decodable, Decoder, Encodable, Encoder};
10 use rustc_serialize::{UseSpecializedDecodable, UseSpecializedEncodable};
12 use std::cmp::{Ord, PartialEq, PartialOrd};
14 use std::hash::{Hash, Hasher};
17 use crate::{Span, DUMMY_SP, GLOBALS};
23 // After modifying this list adjust `is_special`, `is_used_keyword`/`is_unused_keyword`,
24 // this should be rarely necessary though if the keywords are kept in alphabetic order.
26 // Special reserved identifiers used internally for elided lifetimes,
27 // unnamed method parameters, crate root module, error recovery etc.
30 DollarCrate: "$crate",
33 // Keywords that are used in stable Rust.
70 // Keywords that are used in unstable Rust or reserved for future use.
84 // Edition-specific keywords that are used in stable Rust.
85 Async: "async", // >= 2018 Edition only
86 Await: "await", // >= 2018 Edition only
87 Dyn: "dyn", // >= 2018 Edition only
89 // Edition-specific keywords that are used in unstable Rust or reserved for future use.
90 Try: "try", // >= 2018 Edition only
92 // Special lifetime names
93 UnderscoreLifetime: "'_",
94 StaticLifetime: "'static",
96 // Weak keywords, have special meaning only in specific contexts.
104 // Symbols that can be referred to with rustc_span::sym::*. The symbol is
105 // the stringified identifier unless otherwise specified (e.g.
106 // `proc_dash_macro` represents "proc-macro").
108 // As well as the symbols listed, there are symbols for the the strings
109 // "0", "1", ..., "9", which are accessible via `sym::integer`.
111 aarch64_target_feature,
115 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,
171 borrowck_graphviz_format,
172 borrowck_graphviz_postflow,
173 borrowck_graphviz_preflow,
176 braced_empty_structs,
188 cfg_target_has_atomic,
189 cfg_target_thread_local,
197 closure_to_fn_coercion,
199 cmpxchg16b_target_feature,
206 conservative_impl_trait,
208 const_compare_raw_pointers,
216 const_in_array_repeat_expressions,
222 const_raw_ptr_to_usize_cast,
224 const_trait_bound_opt_out,
238 crate_visibility_modifier,
246 custom_inner_attributes,
247 custom_test_frameworks,
255 default_lib_allocator,
256 default_type_parameter_fallback,
258 delay_span_bug_from_inside_query,
273 document_private_items,
274 dotdoteq_in_patterns,
275 dotdot_in_tuple_patterns,
276 double_braced_crate: "{{crate}}",
277 double_braced_impl: "{{impl}}",
278 double_braced_misc: "{{misc}}",
279 double_braced_closure: "{{closure}}",
280 double_braced_constructor: "{{constructor}}",
281 double_braced_constant: "{{constant}}",
282 double_braced_opaque: "{{opaque}}",
284 dropck_parametricity,
300 exclusive_range_pattern,
301 exhaustive_integer_patterns,
307 extern_absolute_paths,
309 extern_crate_item_prelude,
320 field_init_shorthand,
344 generic_associated_types,
349 half_open_range_patterns,
354 hexagon_target_feature,
356 homogeneous_aggregate,
370 if_while_or_patterns,
372 impl_header_lifetime_elision,
374 impl_trait_in_bindings,
382 inclusive_range_syntax,
383 infer_outlives_requirements,
384 infer_static_outlives_requirements,
391 irrefutable_let_patterns,
394 issue_5723_bootstrap,
395 issue_tracker_base_url,
397 item_context: "ItemContext",
416 link_llvm_intrinsics,
426 macro_at_most_once_rep,
429 macro_lifetime_matcher,
430 macro_literal_matcher,
441 match_beginning_vert,
442 match_default_bindings,
462 movbe_target_feature,
499 object_safe_for_dispatch,
501 omit_gdb_pretty_printer_section,
508 optin_builtin_traits,
518 overlapping_marker_traits,
523 panic_implementation,
543 poll_with_tls_context,
544 powerpc_target_feature,
545 precise_pointer_size_matching,
550 proc_dash_macro: "proc-macro",
552 proc_macro_attribute,
558 proc_macro_internals,
560 proc_macro_non_items,
561 proc_macro_path_invoc,
566 quad_precision_float,
581 reexport_test_harness_main,
593 re_rebalance_coherence,
611 rustc_allocator_nounwind,
612 rustc_allow_const_fn_ptr,
613 rustc_args_required_const,
617 rustc_const_unstable,
619 rustc_conversion_suggestion,
622 rustc_diagnostic_item,
623 rustc_diagnostic_macros,
626 rustc_dump_env_program_clauses,
627 rustc_dump_program_clauses,
628 rustc_dump_user_substs,
630 rustc_expected_cgu_reuse,
631 rustc_if_this_changed,
632 rustc_inherit_overflow_checks,
634 rustc_layout_scalar_valid_range_end,
635 rustc_layout_scalar_valid_range_start,
636 rustc_macro_transparency,
638 rustc_nonnull_optimization_guaranteed,
639 rustc_object_lifetime_default,
640 rustc_on_unimplemented,
643 rustc_partition_codegened,
644 rustc_partition_reused,
646 rustc_peek_definite_init,
647 rustc_peek_maybe_init,
648 rustc_peek_maybe_uninit,
649 rustc_peek_indirectly_mutable,
651 rustc_proc_macro_decls,
655 rustc_std_internal_symbol,
658 rustc_reservation_impl,
660 rustc_then_this_would_need,
664 rust_eh_unwind_resume,
666 rvalue_static_promotion,
690 sse4a_target_feature,
703 stmt_expr_attributes,
705 struct_field_attributes,
715 target_has_atomic_load_store,
720 termination_trait_test,
723 test_accepted_feature,
725 test_removed_feature,
748 type_alias_impl_trait,
753 type_alias_enum_variants,
770 underscore_const_names,
772 underscore_lifetimes,
774 universal_impl_trait,
777 unrestricted_attribute_tokens,
780 unsized_tuple_coercion,
796 visible_private_types,
811 #[derive(Copy, Clone, Eq, HashStable_Generic)]
819 /// Constructs a new identifier from a symbol and a span.
820 pub const fn new(name: Symbol, span: Span) -> Ident {
824 /// Constructs a new identifier with a dummy span.
826 pub const fn with_dummy_span(name: Symbol) -> Ident {
827 Ident::new(name, DUMMY_SP)
831 pub fn invalid() -> Ident {
832 Ident::with_dummy_span(kw::Invalid)
835 /// Maps a string to an identifier with a dummy span.
836 pub fn from_str(string: &str) -> Ident {
837 Ident::with_dummy_span(Symbol::intern(string))
840 /// Maps a string and a span to an identifier.
841 pub fn from_str_and_span(string: &str, span: Span) -> Ident {
842 Ident::new(Symbol::intern(string), span)
845 /// Replaces `lo` and `hi` with those from `span`, but keep hygiene context.
846 pub fn with_span_pos(self, span: Span) -> Ident {
847 Ident::new(self.name, span.with_ctxt(self.span.ctxt()))
850 pub fn without_first_quote(self) -> Ident {
851 Ident::new(Symbol::intern(self.as_str().trim_start_matches('\'')), self.span)
854 /// "Normalize" ident for use in comparisons using "item hygiene".
855 /// Identifiers with same string value become same if they came from the same "modern" macro
856 /// (e.g., `macro` item, but not `macro_rules` item) and stay different if they came from
857 /// different "modern" macros.
858 /// Technically, this operation strips all non-opaque marks from ident's syntactic context.
859 pub fn modern(self) -> Ident {
860 Ident::new(self.name, self.span.modern())
863 /// "Normalize" ident for use in comparisons using "local variable hygiene".
864 /// Identifiers with same string value become same if they came from the same non-transparent
865 /// macro (e.g., `macro` or `macro_rules!` items) and stay different if they came from different
866 /// non-transparent macros.
867 /// Technically, this operation strips all transparent marks from ident's syntactic context.
868 pub fn modern_and_legacy(self) -> Ident {
869 Ident::new(self.name, self.span.modern_and_legacy())
872 /// Convert the name to a `SymbolStr`. This is a slowish operation because
873 /// it requires locking the symbol interner.
874 pub fn as_str(self) -> SymbolStr {
879 impl PartialEq for Ident {
880 fn eq(&self, rhs: &Self) -> bool {
881 self.name == rhs.name && self.span.ctxt() == rhs.span.ctxt()
885 impl Hash for Ident {
886 fn hash<H: Hasher>(&self, state: &mut H) {
887 self.name.hash(state);
888 self.span.ctxt().hash(state);
892 impl fmt::Debug for Ident {
893 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
894 if self.is_raw_guess() {
897 write!(f, "{}{:?}", self.name, self.span.ctxt())
901 impl fmt::Display for Ident {
902 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
903 if self.is_raw_guess() {
906 fmt::Display::fmt(&self.name, f)
910 impl UseSpecializedEncodable for Ident {
911 fn default_encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
912 s.emit_struct("Ident", 2, |s| {
913 s.emit_struct_field("name", 0, |s| self.name.encode(s))?;
914 s.emit_struct_field("span", 1, |s| self.span.encode(s))
919 impl UseSpecializedDecodable for Ident {
920 fn default_decode<D: Decoder>(d: &mut D) -> Result<Self, D::Error> {
921 d.read_struct("Ident", 2, |d| {
923 name: d.read_struct_field("name", 0, Decodable::decode)?,
924 span: d.read_struct_field("span", 1, Decodable::decode)?,
930 /// An interned string.
932 /// Internally, a `Symbol` is implemented as an index, and all operations
933 /// (including hashing, equality, and ordering) operate on that index. The use
934 /// of `rustc_index::newtype_index!` means that `Option<Symbol>` only takes up 4 bytes,
935 /// because `rustc_index::newtype_index!` reserves the last 256 values for tagging purposes.
937 /// Note that `Symbol` cannot directly be a `rustc_index::newtype_index!` because it
938 /// implements `fmt::Debug`, `Encodable`, and `Decodable` in special ways.
939 #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
940 pub struct Symbol(SymbolIndex);
942 rustc_index::newtype_index! {
943 pub struct SymbolIndex { .. }
947 const fn new(n: u32) -> Self {
948 Symbol(SymbolIndex::from_u32_const(n))
951 /// Maps a string to its interned representation.
952 pub fn intern(string: &str) -> Self {
953 with_interner(|interner| interner.intern(string))
956 /// Access the symbol's chars. This is a slowish operation because it
957 /// requires locking the symbol interner.
958 pub fn with<F: FnOnce(&str) -> R, R>(self, f: F) -> R {
959 with_interner(|interner| f(interner.get(self)))
962 /// Convert to a `SymbolStr`. This is a slowish operation because it
963 /// requires locking the symbol interner.
964 pub fn as_str(self) -> SymbolStr {
965 with_interner(|interner| unsafe {
966 SymbolStr { string: std::mem::transmute::<&str, &str>(interner.get(self)) }
970 pub fn as_u32(self) -> u32 {
975 impl fmt::Debug for Symbol {
976 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
977 self.with(|str| fmt::Debug::fmt(&str, f))
981 impl fmt::Display for Symbol {
982 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
983 self.with(|str| fmt::Display::fmt(&str, f))
987 impl Encodable for Symbol {
988 fn encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
989 self.with(|string| s.emit_str(string))
993 impl Decodable for Symbol {
994 fn decode<D: Decoder>(d: &mut D) -> Result<Symbol, D::Error> {
995 Ok(Symbol::intern(&d.read_str()?))
999 impl<CTX> HashStable<CTX> for Symbol {
1001 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
1002 self.as_str().hash_stable(hcx, hasher);
1006 impl<CTX> ToStableHashKey<CTX> for Symbol {
1007 type KeyType = SymbolStr;
1010 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {
1015 // The `&'static str`s in this type actually point into the arena.
1017 pub struct Interner {
1018 arena: DroplessArena,
1019 names: FxHashMap<&'static str, Symbol>,
1020 strings: Vec<&'static str>,
1024 fn prefill(init: &[&'static str]) -> Self {
1026 strings: init.into(),
1027 names: init.iter().copied().zip((0..).map(Symbol::new)).collect(),
1028 ..Default::default()
1032 pub fn intern(&mut self, string: &str) -> Symbol {
1033 if let Some(&name) = self.names.get(string) {
1037 let name = Symbol::new(self.strings.len() as u32);
1039 // `from_utf8_unchecked` is safe since we just allocated a `&str` which is known to be
1042 unsafe { str::from_utf8_unchecked(self.arena.alloc_slice(string.as_bytes())) };
1043 // It is safe to extend the arena allocation to `'static` because we only access
1044 // these while the arena is still alive.
1045 let string: &'static str = unsafe { &*(string as *const str) };
1046 self.strings.push(string);
1047 self.names.insert(string, name);
1051 // Get the symbol as a string. `Symbol::as_str()` should be used in
1052 // preference to this function.
1053 pub fn get(&self, symbol: Symbol) -> &str {
1054 self.strings[symbol.0.as_usize()]
1058 // This module has a very short name because it's used a lot.
1064 // This module has a very short name because it's used a lot.
1065 #[allow(rustc::default_hash_types)]
1068 use std::convert::TryInto;
1072 // Get the symbol for an integer. The first few non-negative integers each
1073 // have a static symbol and therefore are fast.
1074 pub fn integer<N: TryInto<usize> + Copy + ToString>(n: N) -> Symbol {
1075 if let Result::Ok(idx) = n.try_into() {
1076 if let Option::Some(&sym) = digits_array.get(idx) {
1080 Symbol::intern(&n.to_string())
1085 fn is_used_keyword_2018(self) -> bool {
1086 self >= kw::Async && self <= kw::Dyn
1089 fn is_unused_keyword_2018(self) -> bool {
1093 /// Used for sanity checking rustdoc keyword sections.
1094 pub fn is_doc_keyword(self) -> bool {
1098 /// A keyword or reserved identifier that can be used as a path segment.
1099 pub fn is_path_segment_keyword(self) -> bool {
1101 || self == kw::SelfLower
1102 || self == kw::SelfUpper
1103 || self == kw::Crate
1104 || self == kw::PathRoot
1105 || self == kw::DollarCrate
1108 /// Returns `true` if the symbol is `true` or `false`.
1109 pub fn is_bool_lit(self) -> bool {
1110 self == kw::True || self == kw::False
1113 /// This symbol can be a raw identifier.
1114 pub fn can_be_raw(self) -> bool {
1115 self != kw::Invalid && self != kw::Underscore && !self.is_path_segment_keyword()
1120 // Returns `true` for reserved identifiers used internally for elided lifetimes,
1121 // unnamed method parameters, crate root module, error recovery etc.
1122 pub fn is_special(self) -> bool {
1123 self.name <= kw::Underscore
1126 /// Returns `true` if the token is a keyword used in the language.
1127 pub fn is_used_keyword(self) -> bool {
1128 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1129 self.name >= kw::As && self.name <= kw::While
1130 || self.name.is_used_keyword_2018() && self.span.rust_2018()
1133 /// Returns `true` if the token is a keyword reserved for possible future use.
1134 pub fn is_unused_keyword(self) -> bool {
1135 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1136 self.name >= kw::Abstract && self.name <= kw::Yield
1137 || self.name.is_unused_keyword_2018() && self.span.rust_2018()
1140 /// Returns `true` if the token is either a special identifier or a keyword.
1141 pub fn is_reserved(self) -> bool {
1142 self.is_special() || self.is_used_keyword() || self.is_unused_keyword()
1145 /// A keyword or reserved identifier that can be used as a path segment.
1146 pub fn is_path_segment_keyword(self) -> bool {
1147 self.name.is_path_segment_keyword()
1150 /// We see this identifier in a normal identifier position, like variable name or a type.
1151 /// How was it written originally? Did it use the raw form? Let's try to guess.
1152 pub fn is_raw_guess(self) -> bool {
1153 self.name.can_be_raw() && self.is_reserved()
1158 fn with_interner<T, F: FnOnce(&mut Interner) -> T>(f: F) -> T {
1159 GLOBALS.with(|globals| f(&mut *globals.symbol_interner.lock()))
1162 /// An alternative to `Symbol`, useful when the chars within the symbol need to
1163 /// be accessed. It deliberately has limited functionality and should only be
1164 /// used for temporary values.
1166 /// Because the interner outlives any thread which uses this type, we can
1167 /// safely treat `string` which points to interner data, as an immortal string,
1168 /// as long as this type never crosses between threads.
1170 // FIXME: ensure that the interner outlives any thread which uses `SymbolStr`,
1171 // by creating a new thread right after constructing the interner.
1172 #[derive(Clone, Eq, PartialOrd, Ord)]
1173 pub struct SymbolStr {
1174 string: &'static str,
1177 // This impl allows a `SymbolStr` to be directly equated with a `String` or
1179 impl<T: std::ops::Deref<Target = str>> std::cmp::PartialEq<T> for SymbolStr {
1180 fn eq(&self, other: &T) -> bool {
1181 self.string == other.deref()
1185 impl !Send for SymbolStr {}
1186 impl !Sync for SymbolStr {}
1188 /// This impl means that if `ss` is a `SymbolStr`:
1189 /// - `*ss` is a `str`;
1190 /// - `&*ss` is a `&str`;
1191 /// - `&ss as &str` is a `&str`, which means that `&ss` can be passed to a
1192 /// function expecting a `&str`.
1193 impl std::ops::Deref for SymbolStr {
1196 fn deref(&self) -> &str {
1201 impl fmt::Debug for SymbolStr {
1202 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1203 fmt::Debug::fmt(self.string, f)
1207 impl fmt::Display for SymbolStr {
1208 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1209 fmt::Display::fmt(self.string, f)
1213 impl<CTX> HashStable<CTX> for SymbolStr {
1215 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
1216 self.string.hash_stable(hcx, hasher)
1220 impl<CTX> ToStableHashKey<CTX> for SymbolStr {
1221 type KeyType = SymbolStr;
1224 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {