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,
219 const_raw_ptr_to_usize_cast,
233 crate_visibility_modifier,
241 custom_inner_attributes,
242 custom_test_frameworks,
250 default_lib_allocator,
251 default_type_parameter_fallback,
253 delay_span_bug_from_inside_query,
268 document_private_items,
269 dotdoteq_in_patterns,
270 dotdot_in_tuple_patterns,
271 double_braced_crate: "{{crate}}",
272 double_braced_impl: "{{impl}}",
273 double_braced_misc: "{{misc}}",
274 double_braced_closure: "{{closure}}",
275 double_braced_constructor: "{{constructor}}",
276 double_braced_constant: "{{constant}}",
277 double_braced_opaque: "{{opaque}}",
279 dropck_parametricity,
295 exclusive_range_pattern,
296 exhaustive_integer_patterns,
302 extern_absolute_paths,
304 extern_crate_item_prelude,
315 field_init_shorthand,
338 generic_associated_types,
347 hexagon_target_feature,
349 homogeneous_aggregate,
363 if_while_or_patterns,
365 impl_header_lifetime_elision,
367 impl_trait_in_bindings,
375 inclusive_range_syntax,
376 infer_outlives_requirements,
377 infer_static_outlives_requirements,
384 irrefutable_let_patterns,
387 issue_5723_bootstrap,
388 issue_tracker_base_url,
390 item_context: "ItemContext",
409 link_llvm_intrinsics,
419 macro_at_most_once_rep,
422 macro_lifetime_matcher,
423 macro_literal_matcher,
434 match_beginning_vert,
435 match_default_bindings,
453 movbe_target_feature,
489 object_safe_for_dispatch,
491 omit_gdb_pretty_printer_section,
498 optin_builtin_traits,
508 overlapping_marker_traits,
513 panic_implementation,
533 poll_with_tls_context,
534 powerpc_target_feature,
535 precise_pointer_size_matching,
540 proc_dash_macro: "proc-macro",
542 proc_macro_attribute,
548 proc_macro_internals,
550 proc_macro_non_items,
551 proc_macro_path_invoc,
556 quad_precision_float,
571 reexport_test_harness_main,
583 re_rebalance_coherence,
601 rustc_allocator_nounwind,
602 rustc_allow_const_fn_ptr,
603 rustc_args_required_const,
607 rustc_const_unstable,
608 rustc_conversion_suggestion,
611 rustc_diagnostic_item,
612 rustc_diagnostic_macros,
615 rustc_dump_env_program_clauses,
616 rustc_dump_program_clauses,
617 rustc_dump_user_substs,
619 rustc_expected_cgu_reuse,
620 rustc_if_this_changed,
621 rustc_inherit_overflow_checks,
623 rustc_layout_scalar_valid_range_end,
624 rustc_layout_scalar_valid_range_start,
625 rustc_macro_transparency,
627 rustc_nonnull_optimization_guaranteed,
628 rustc_object_lifetime_default,
629 rustc_on_unimplemented,
632 rustc_partition_codegened,
633 rustc_partition_reused,
635 rustc_peek_definite_init,
636 rustc_peek_maybe_init,
637 rustc_peek_maybe_uninit,
638 rustc_peek_indirectly_mutable,
640 rustc_proc_macro_decls,
644 rustc_std_internal_symbol,
647 rustc_reservation_impl,
649 rustc_then_this_would_need,
653 rust_eh_unwind_resume,
655 rvalue_static_promotion,
679 sse4a_target_feature,
692 stmt_expr_attributes,
694 struct_field_attributes,
704 target_has_atomic_load_store,
709 termination_trait_test,
712 test_accepted_feature,
714 test_removed_feature,
735 type_alias_impl_trait,
740 type_alias_enum_variants,
752 underscore_const_names,
754 underscore_lifetimes,
756 universal_impl_trait,
759 unrestricted_attribute_tokens,
762 unsized_tuple_coercion,
777 visible_private_types,
792 #[derive(Copy, Clone, Eq, HashStable_Generic)]
800 /// Constructs a new identifier from a symbol and a span.
801 pub const fn new(name: Symbol, span: Span) -> Ident {
805 /// Constructs a new identifier with a dummy span.
807 pub const fn with_dummy_span(name: Symbol) -> Ident {
808 Ident::new(name, DUMMY_SP)
812 pub fn invalid() -> Ident {
813 Ident::with_dummy_span(kw::Invalid)
816 /// Maps a string to an identifier with a dummy span.
817 pub fn from_str(string: &str) -> Ident {
818 Ident::with_dummy_span(Symbol::intern(string))
821 /// Maps a string and a span to an identifier.
822 pub fn from_str_and_span(string: &str, span: Span) -> Ident {
823 Ident::new(Symbol::intern(string), span)
826 /// Replaces `lo` and `hi` with those from `span`, but keep hygiene context.
827 pub fn with_span_pos(self, span: Span) -> Ident {
828 Ident::new(self.name, span.with_ctxt(self.span.ctxt()))
831 pub fn without_first_quote(self) -> Ident {
832 Ident::new(Symbol::intern(self.as_str().trim_start_matches('\'')), self.span)
835 /// "Normalize" ident for use in comparisons using "item hygiene".
836 /// Identifiers with same string value become same if they came from the same "modern" macro
837 /// (e.g., `macro` item, but not `macro_rules` item) and stay different if they came from
838 /// different "modern" macros.
839 /// Technically, this operation strips all non-opaque marks from ident's syntactic context.
840 pub fn modern(self) -> Ident {
841 Ident::new(self.name, self.span.modern())
844 /// "Normalize" ident for use in comparisons using "local variable hygiene".
845 /// Identifiers with same string value become same if they came from the same non-transparent
846 /// macro (e.g., `macro` or `macro_rules!` items) and stay different if they came from different
847 /// non-transparent macros.
848 /// Technically, this operation strips all transparent marks from ident's syntactic context.
849 pub fn modern_and_legacy(self) -> Ident {
850 Ident::new(self.name, self.span.modern_and_legacy())
853 /// Convert the name to a `SymbolStr`. This is a slowish operation because
854 /// it requires locking the symbol interner.
855 pub fn as_str(self) -> SymbolStr {
860 impl PartialEq for Ident {
861 fn eq(&self, rhs: &Self) -> bool {
862 self.name == rhs.name && self.span.ctxt() == rhs.span.ctxt()
866 impl Hash for Ident {
867 fn hash<H: Hasher>(&self, state: &mut H) {
868 self.name.hash(state);
869 self.span.ctxt().hash(state);
873 impl fmt::Debug for Ident {
874 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
875 if self.is_raw_guess() {
878 write!(f, "{}{:?}", self.name, self.span.ctxt())
882 impl fmt::Display for Ident {
883 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
884 if self.is_raw_guess() {
887 fmt::Display::fmt(&self.name, f)
891 impl UseSpecializedEncodable for Ident {
892 fn default_encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
893 s.emit_struct("Ident", 2, |s| {
894 s.emit_struct_field("name", 0, |s| {
897 s.emit_struct_field("span", 1, |s| {
904 impl UseSpecializedDecodable for Ident {
905 fn default_decode<D: Decoder>(d: &mut D) -> Result<Self, D::Error> {
906 d.read_struct("Ident", 2, |d| {
908 name: d.read_struct_field("name", 0, Decodable::decode)?,
909 span: d.read_struct_field("span", 1, Decodable::decode)?,
915 /// An interned string.
917 /// Internally, a `Symbol` is implemented as an index, and all operations
918 /// (including hashing, equality, and ordering) operate on that index. The use
919 /// of `rustc_index::newtype_index!` means that `Option<Symbol>` only takes up 4 bytes,
920 /// because `rustc_index::newtype_index!` reserves the last 256 values for tagging purposes.
922 /// Note that `Symbol` cannot directly be a `rustc_index::newtype_index!` because it
923 /// implements `fmt::Debug`, `Encodable`, and `Decodable` in special ways.
924 #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
925 pub struct Symbol(SymbolIndex);
927 rustc_index::newtype_index! {
928 pub struct SymbolIndex { .. }
932 const fn new(n: u32) -> Self {
933 Symbol(SymbolIndex::from_u32_const(n))
936 /// Maps a string to its interned representation.
937 pub fn intern(string: &str) -> Self {
938 with_interner(|interner| interner.intern(string))
941 /// Access the symbol's chars. This is a slowish operation because it
942 /// requires locking the symbol interner.
943 pub fn with<F: FnOnce(&str) -> R, R>(self, f: F) -> R {
944 with_interner(|interner| {
945 f(interner.get(self))
949 /// Convert to a `SymbolStr`. This is a slowish operation because it
950 /// requires locking the symbol interner.
951 pub fn as_str(self) -> SymbolStr {
952 with_interner(|interner| unsafe {
954 string: std::mem::transmute::<&str, &str>(interner.get(self))
959 pub fn as_u32(self) -> u32 {
964 impl fmt::Debug for Symbol {
965 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
966 self.with(|str| fmt::Debug::fmt(&str, f))
970 impl fmt::Display for Symbol {
971 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
972 self.with(|str| fmt::Display::fmt(&str, f))
976 impl Encodable for Symbol {
977 fn encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
978 self.with(|string| s.emit_str(string))
982 impl Decodable for Symbol {
983 fn decode<D: Decoder>(d: &mut D) -> Result<Symbol, D::Error> {
984 Ok(Symbol::intern(&d.read_str()?))
988 impl<CTX> HashStable<CTX> for Symbol {
990 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
991 self.as_str().hash_stable(hcx, hasher);
995 impl<CTX> ToStableHashKey<CTX> for Symbol {
996 type KeyType = SymbolStr;
999 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {
1004 // The `&'static str`s in this type actually point into the arena.
1006 pub struct Interner {
1007 arena: DroplessArena,
1008 names: FxHashMap<&'static str, Symbol>,
1009 strings: Vec<&'static str>,
1013 fn prefill(init: &[&'static str]) -> Self {
1015 strings: init.into(),
1016 names: init.iter().copied().zip((0..).map(Symbol::new)).collect(),
1017 ..Default::default()
1021 pub fn intern(&mut self, string: &str) -> Symbol {
1022 if let Some(&name) = self.names.get(string) {
1026 let name = Symbol::new(self.strings.len() as u32);
1028 // `from_utf8_unchecked` is safe since we just allocated a `&str` which is known to be
1030 let string: &str = unsafe {
1031 str::from_utf8_unchecked(self.arena.alloc_slice(string.as_bytes()))
1033 // It is safe to extend the arena allocation to `'static` because we only access
1034 // these while the arena is still alive.
1035 let string: &'static str = unsafe {
1036 &*(string as *const str)
1038 self.strings.push(string);
1039 self.names.insert(string, name);
1043 // Get the symbol as a string. `Symbol::as_str()` should be used in
1044 // preference to this function.
1045 pub fn get(&self, symbol: Symbol) -> &str {
1046 self.strings[symbol.0.as_usize()]
1050 // This module has a very short name because it's used a lot.
1056 // This module has a very short name because it's used a lot.
1058 use std::convert::TryInto;
1063 // Get the symbol for an integer. The first few non-negative integers each
1064 // have a static symbol and therefore are fast.
1065 pub fn integer<N: TryInto<usize> + Copy + ToString>(n: N) -> Symbol {
1066 if let Result::Ok(idx) = n.try_into() {
1067 if let Option::Some(&sym) = digits_array.get(idx) {
1071 Symbol::intern(&n.to_string())
1076 fn is_used_keyword_2018(self) -> bool {
1077 self >= kw::Async && self <= kw::Dyn
1080 fn is_unused_keyword_2018(self) -> bool {
1084 /// Used for sanity checking rustdoc keyword sections.
1085 pub fn is_doc_keyword(self) -> bool {
1089 /// A keyword or reserved identifier that can be used as a path segment.
1090 pub fn is_path_segment_keyword(self) -> bool {
1091 self == kw::Super ||
1092 self == kw::SelfLower ||
1093 self == kw::SelfUpper ||
1094 self == kw::Crate ||
1095 self == kw::PathRoot ||
1096 self == kw::DollarCrate
1099 /// Returns `true` if the symbol is `true` or `false`.
1100 pub fn is_bool_lit(self) -> bool {
1101 self == kw::True || self == kw::False
1104 /// This symbol can be a raw identifier.
1105 pub fn can_be_raw(self) -> bool {
1106 self != kw::Invalid && self != kw::Underscore && !self.is_path_segment_keyword()
1111 // Returns `true` for reserved identifiers used internally for elided lifetimes,
1112 // unnamed method parameters, crate root module, error recovery etc.
1113 pub fn is_special(self) -> bool {
1114 self.name <= kw::Underscore
1117 /// Returns `true` if the token is a keyword used in the language.
1118 pub fn is_used_keyword(self) -> bool {
1119 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1120 self.name >= kw::As && self.name <= kw::While ||
1121 self.name.is_used_keyword_2018() && self.span.rust_2018()
1124 /// Returns `true` if the token is a keyword reserved for possible future use.
1125 pub fn is_unused_keyword(self) -> bool {
1126 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1127 self.name >= kw::Abstract && self.name <= kw::Yield ||
1128 self.name.is_unused_keyword_2018() && self.span.rust_2018()
1131 /// Returns `true` if the token is either a special identifier or a keyword.
1132 pub fn is_reserved(self) -> bool {
1133 self.is_special() || self.is_used_keyword() || self.is_unused_keyword()
1136 /// A keyword or reserved identifier that can be used as a path segment.
1137 pub fn is_path_segment_keyword(self) -> bool {
1138 self.name.is_path_segment_keyword()
1141 /// We see this identifier in a normal identifier position, like variable name or a type.
1142 /// How was it written originally? Did it use the raw form? Let's try to guess.
1143 pub fn is_raw_guess(self) -> bool {
1144 self.name.can_be_raw() && self.is_reserved()
1149 fn with_interner<T, F: FnOnce(&mut Interner) -> T>(f: F) -> T {
1150 GLOBALS.with(|globals| f(&mut *globals.symbol_interner.lock()))
1153 /// An alternative to `Symbol`, useful when the chars within the symbol need to
1154 /// be accessed. It deliberately has limited functionality and should only be
1155 /// used for temporary values.
1157 /// Because the interner outlives any thread which uses this type, we can
1158 /// safely treat `string` which points to interner data, as an immortal string,
1159 /// as long as this type never crosses between threads.
1161 // FIXME: ensure that the interner outlives any thread which uses `SymbolStr`,
1162 // by creating a new thread right after constructing the interner.
1163 #[derive(Clone, Eq, PartialOrd, Ord)]
1164 pub struct SymbolStr {
1165 string: &'static str,
1168 // This impl allows a `SymbolStr` to be directly equated with a `String` or
1170 impl<T: std::ops::Deref<Target = str>> std::cmp::PartialEq<T> for SymbolStr {
1171 fn eq(&self, other: &T) -> bool {
1172 self.string == other.deref()
1176 impl !Send for SymbolStr {}
1177 impl !Sync for SymbolStr {}
1179 /// This impl means that if `ss` is a `SymbolStr`:
1180 /// - `*ss` is a `str`;
1181 /// - `&*ss` is a `&str`;
1182 /// - `&ss as &str` is a `&str`, which means that `&ss` can be passed to a
1183 /// function expecting a `&str`.
1184 impl std::ops::Deref for SymbolStr {
1187 fn deref(&self) -> &str { self.string }
1190 impl fmt::Debug for SymbolStr {
1191 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1192 fmt::Debug::fmt(self.string, f)
1196 impl fmt::Display for SymbolStr {
1197 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1198 fmt::Display::fmt(self.string, f)
1202 impl<CTX> HashStable<CTX> for SymbolStr {
1204 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
1205 self.string.hash_stable(hcx, hasher)
1209 impl<CTX> ToStableHashKey<CTX> for SymbolStr {
1210 type KeyType = SymbolStr;
1213 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {