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.
100 MacroRules: "macro_rules",
105 // Symbols that can be referred to with rustc_span::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,
127 advanced_slice_patterns,
139 allow_internal_unsafe,
140 allow_internal_unstable,
141 allow_internal_unstable_backcompat_hack,
145 arbitrary_enum_discriminant,
146 arbitrary_self_types,
154 associated_type_bounds,
155 associated_type_defaults,
163 augmented_assignments,
164 automatically_derived,
165 avx512_target_feature,
170 bind_by_move_pattern_guards,
174 borrowck_graphviz_format,
175 borrowck_graphviz_postflow,
176 borrowck_graphviz_preflow,
179 braced_empty_structs,
192 cfg_target_has_atomic,
193 cfg_target_thread_local,
201 closure_to_fn_coercion,
203 cmpxchg16b_target_feature,
210 conservative_impl_trait,
212 const_compare_raw_pointers,
221 const_in_array_repeat_expressions,
227 const_raw_ptr_to_usize_cast,
229 const_trait_bound_opt_out,
243 crate_visibility_modifier,
251 custom_inner_attributes,
252 custom_test_frameworks,
261 default_lib_allocator,
262 default_type_parameter_fallback,
264 delay_span_bug_from_inside_query,
279 document_private_items,
280 dotdoteq_in_patterns,
281 dotdot_in_tuple_patterns,
282 double_braced_crate: "{{crate}}",
283 double_braced_impl: "{{impl}}",
284 double_braced_misc: "{{misc}}",
285 double_braced_closure: "{{closure}}",
286 double_braced_constructor: "{{constructor}}",
287 double_braced_constant: "{{constant}}",
288 double_braced_opaque: "{{opaque}}",
290 dropck_parametricity,
305 exclusive_range_pattern,
306 exhaustive_integer_patterns,
312 extern_absolute_paths,
314 extern_crate_item_prelude,
325 field_init_shorthand,
349 generic_associated_types,
355 half_open_range_patterns,
360 hexagon_target_feature,
362 homogeneous_aggregate,
376 if_while_or_patterns,
378 impl_header_lifetime_elision,
380 impl_trait_in_bindings,
388 inclusive_range_syntax,
389 infer_outlives_requirements,
390 infer_static_outlives_requirements,
397 irrefutable_let_patterns,
400 issue_5723_bootstrap,
401 issue_tracker_base_url,
403 item_context: "ItemContext",
422 link_llvm_intrinsics,
433 macro_at_most_once_rep,
436 macro_lifetime_matcher,
437 macro_literal_matcher,
447 match_beginning_vert,
448 match_default_bindings,
469 movbe_target_feature,
508 object_safe_for_dispatch,
510 omit_gdb_pretty_printer_section,
517 optin_builtin_traits,
527 overlapping_marker_traits,
532 panic_implementation,
553 powerpc_target_feature,
554 precise_pointer_size_matching,
559 proc_dash_macro: "proc-macro",
561 proc_macro_attribute,
567 proc_macro_internals,
569 proc_macro_non_items,
570 proc_macro_path_invoc,
575 quad_precision_float,
591 reexport_test_harness_main,
604 re_rebalance_coherence,
622 rustc_allocator_nounwind,
623 rustc_allow_const_fn_ptr,
624 rustc_args_required_const,
628 rustc_const_unstable,
630 rustc_conversion_suggestion,
633 rustc_diagnostic_item,
634 rustc_diagnostic_macros,
637 rustc_dump_env_program_clauses,
638 rustc_dump_program_clauses,
639 rustc_dump_user_substs,
641 rustc_expected_cgu_reuse,
642 rustc_if_this_changed,
643 rustc_inherit_overflow_checks,
645 rustc_layout_scalar_valid_range_end,
646 rustc_layout_scalar_valid_range_start,
647 rustc_macro_transparency,
649 rustc_nonnull_optimization_guaranteed,
650 rustc_object_lifetime_default,
651 rustc_on_unimplemented,
654 rustc_partition_codegened,
655 rustc_partition_reused,
657 rustc_peek_definite_init,
658 rustc_peek_maybe_init,
659 rustc_peek_maybe_uninit,
660 rustc_peek_indirectly_mutable,
662 rustc_proc_macro_decls,
665 rustc_unsafe_specialization_marker,
666 rustc_specialization_trait,
668 rustc_std_internal_symbol,
671 rustc_reservation_impl,
673 rustc_then_this_would_need,
678 rvalue_static_promotion,
701 sse4a_target_feature,
714 stmt_expr_attributes,
716 struct_field_attributes,
726 target_has_atomic_load_store,
732 termination_trait_test,
735 test_accepted_feature,
737 test_removed_feature,
760 type_alias_impl_trait,
765 type_alias_enum_variants,
782 underscore_const_names,
784 underscore_lifetimes,
786 universal_impl_trait,
789 unrestricted_attribute_tokens,
792 unsized_tuple_coercion,
808 visible_private_types,
823 #[derive(Copy, Clone, Eq, HashStable_Generic)]
831 /// Constructs a new identifier from a symbol and a span.
832 pub const fn new(name: Symbol, span: Span) -> Ident {
836 /// Constructs a new identifier with a dummy span.
838 pub const fn with_dummy_span(name: Symbol) -> Ident {
839 Ident::new(name, DUMMY_SP)
843 pub fn invalid() -> Ident {
844 Ident::with_dummy_span(kw::Invalid)
847 /// Maps a string to an identifier with a dummy span.
848 pub fn from_str(string: &str) -> Ident {
849 Ident::with_dummy_span(Symbol::intern(string))
852 /// Maps a string and a span to an identifier.
853 pub fn from_str_and_span(string: &str, span: Span) -> Ident {
854 Ident::new(Symbol::intern(string), span)
857 /// Replaces `lo` and `hi` with those from `span`, but keep hygiene context.
858 pub fn with_span_pos(self, span: Span) -> Ident {
859 Ident::new(self.name, span.with_ctxt(self.span.ctxt()))
862 pub fn without_first_quote(self) -> Ident {
863 Ident::new(Symbol::intern(self.as_str().trim_start_matches('\'')), self.span)
866 /// "Normalize" ident for use in comparisons using "item hygiene".
867 /// Identifiers with same string value become same if they came from the same macro 2.0 macro
868 /// (e.g., `macro` item, but not `macro_rules` item) and stay different if they came from
869 /// different macro 2.0 macros.
870 /// Technically, this operation strips all non-opaque marks from ident's syntactic context.
871 pub fn normalize_to_macros_2_0(self) -> Ident {
872 Ident::new(self.name, self.span.normalize_to_macros_2_0())
875 /// "Normalize" ident for use in comparisons using "local variable hygiene".
876 /// Identifiers with same string value become same if they came from the same non-transparent
877 /// macro (e.g., `macro` or `macro_rules!` items) and stay different if they came from different
878 /// non-transparent macros.
879 /// Technically, this operation strips all transparent marks from ident's syntactic context.
880 pub fn normalize_to_macro_rules(self) -> Ident {
881 Ident::new(self.name, self.span.normalize_to_macro_rules())
884 /// Convert the name to a `SymbolStr`. This is a slowish operation because
885 /// it requires locking the symbol interner.
886 pub fn as_str(self) -> SymbolStr {
891 impl PartialEq for Ident {
892 fn eq(&self, rhs: &Self) -> bool {
893 self.name == rhs.name && self.span.ctxt() == rhs.span.ctxt()
897 impl Hash for Ident {
898 fn hash<H: Hasher>(&self, state: &mut H) {
899 self.name.hash(state);
900 self.span.ctxt().hash(state);
904 impl fmt::Debug for Ident {
905 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
906 fmt::Display::fmt(self, f)?;
907 fmt::Debug::fmt(&self.span.ctxt(), f)
911 /// This implementation is supposed to be used in error messages, so it's expected to be identical
912 /// to printing the original identifier token written in source code (`token_to_string`),
913 /// except that AST identifiers don't keep the rawness flag, so we have to guess it.
914 impl fmt::Display for Ident {
915 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
916 fmt::Display::fmt(&IdentPrinter::new(self.name, self.is_raw_guess(), None), f)
920 impl UseSpecializedEncodable for Ident {
921 fn default_encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
922 s.emit_struct("Ident", 2, |s| {
923 s.emit_struct_field("name", 0, |s| self.name.encode(s))?;
924 s.emit_struct_field("span", 1, |s| self.span.encode(s))
929 impl UseSpecializedDecodable for Ident {
930 fn default_decode<D: Decoder>(d: &mut D) -> Result<Self, D::Error> {
931 d.read_struct("Ident", 2, |d| {
933 name: d.read_struct_field("name", 0, Decodable::decode)?,
934 span: d.read_struct_field("span", 1, Decodable::decode)?,
940 /// This is the most general way to print identifiers.
941 /// AST pretty-printer is used as a fallback for turning AST structures into token streams for
942 /// proc macros. Additionally, proc macros may stringify their input and expect it survive the
943 /// stringification (especially true for proc macro derives written between Rust 1.15 and 1.30).
944 /// So we need to somehow pretty-print `$crate` in a way preserving at least some of its
945 /// hygiene data, most importantly name of the crate it refers to.
946 /// As a result we print `$crate` as `crate` if it refers to the local crate
947 /// and as `::other_crate_name` if it refers to some other crate.
948 /// Note, that this is only done if the ident token is printed from inside of AST pretty-pringing,
949 /// but not otherwise. Pretty-printing is the only way for proc macros to discover token contents,
950 /// so we should not perform this lossy conversion if the top level call to the pretty-printer was
951 /// done for a token stream or a single token.
952 pub struct IdentPrinter {
955 /// Span used for retrieving the crate name to which `$crate` refers to,
956 /// if this field is `None` then the `$crate` conversion doesn't happen.
957 convert_dollar_crate: Option<Span>,
961 /// The most general `IdentPrinter` constructor. Do not use this.
962 pub fn new(symbol: Symbol, is_raw: bool, convert_dollar_crate: Option<Span>) -> IdentPrinter {
963 IdentPrinter { symbol, is_raw, convert_dollar_crate }
966 /// This implementation is supposed to be used when printing identifiers
967 /// as a part of pretty-printing for larger AST pieces.
968 /// Do not use this either.
969 pub fn for_ast_ident(ident: Ident, is_raw: bool) -> IdentPrinter {
970 IdentPrinter::new(ident.name, is_raw, Some(ident.span))
974 impl fmt::Display for IdentPrinter {
975 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
979 if self.symbol == kw::DollarCrate {
980 if let Some(span) = self.convert_dollar_crate {
981 let converted = span.ctxt().dollar_crate_name();
982 if !converted.is_path_segment_keyword() {
985 return fmt::Display::fmt(&converted, f);
989 fmt::Display::fmt(&self.symbol, f)
993 /// An newtype around `Ident` that calls [Ident::normalize_to_macro_rules] on
995 // FIXME(matthewj, petrochenkov) Use this more often, add a similar
996 // `ModernIdent` struct and use that as well.
997 #[derive(Copy, Clone, Eq, PartialEq, Hash)]
998 pub struct MacroRulesNormalizedIdent(Ident);
1000 impl MacroRulesNormalizedIdent {
1001 pub fn new(ident: Ident) -> Self {
1002 Self(ident.normalize_to_macro_rules())
1006 impl fmt::Debug for MacroRulesNormalizedIdent {
1007 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1008 fmt::Debug::fmt(&self.0, f)
1012 impl fmt::Display for MacroRulesNormalizedIdent {
1013 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1014 fmt::Display::fmt(&self.0, f)
1018 /// An interned string.
1020 /// Internally, a `Symbol` is implemented as an index, and all operations
1021 /// (including hashing, equality, and ordering) operate on that index. The use
1022 /// of `rustc_index::newtype_index!` means that `Option<Symbol>` only takes up 4 bytes,
1023 /// because `rustc_index::newtype_index!` reserves the last 256 values for tagging purposes.
1025 /// Note that `Symbol` cannot directly be a `rustc_index::newtype_index!` because it
1026 /// implements `fmt::Debug`, `Encodable`, and `Decodable` in special ways.
1027 #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
1028 pub struct Symbol(SymbolIndex);
1030 rustc_index::newtype_index! {
1031 pub struct SymbolIndex { .. }
1035 const fn new(n: u32) -> Self {
1036 Symbol(SymbolIndex::from_u32(n))
1039 /// Maps a string to its interned representation.
1040 pub fn intern(string: &str) -> Self {
1041 with_interner(|interner| interner.intern(string))
1044 /// Access the symbol's chars. This is a slowish operation because it
1045 /// requires locking the symbol interner.
1046 pub fn with<F: FnOnce(&str) -> R, R>(self, f: F) -> R {
1047 with_interner(|interner| f(interner.get(self)))
1050 /// Convert to a `SymbolStr`. This is a slowish operation because it
1051 /// requires locking the symbol interner.
1052 pub fn as_str(self) -> SymbolStr {
1053 with_interner(|interner| unsafe {
1054 SymbolStr { string: std::mem::transmute::<&str, &str>(interner.get(self)) }
1058 pub fn as_u32(self) -> u32 {
1062 /// This method is supposed to be used in error messages, so it's expected to be
1063 /// identical to printing the original identifier token written in source code
1064 /// (`token_to_string`, `Ident::to_string`), except that symbols don't keep the rawness flag
1065 /// or edition, so we have to guess the rawness using the global edition.
1066 pub fn to_ident_string(self) -> String {
1067 Ident::with_dummy_span(self).to_string()
1071 impl fmt::Debug for Symbol {
1072 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1073 self.with(|str| fmt::Debug::fmt(&str, f))
1077 impl fmt::Display for Symbol {
1078 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1079 self.with(|str| fmt::Display::fmt(&str, f))
1083 impl Encodable for Symbol {
1084 fn encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
1085 self.with(|string| s.emit_str(string))
1089 impl Decodable for Symbol {
1091 fn decode<D: Decoder>(d: &mut D) -> Result<Symbol, D::Error> {
1092 Ok(Symbol::intern(&d.read_str()?))
1096 impl<CTX> HashStable<CTX> for Symbol {
1098 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
1099 self.as_str().hash_stable(hcx, hasher);
1103 impl<CTX> ToStableHashKey<CTX> for Symbol {
1104 type KeyType = SymbolStr;
1107 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {
1112 // The `&'static str`s in this type actually point into the arena.
1114 pub struct Interner {
1115 arena: DroplessArena,
1116 names: FxHashMap<&'static str, Symbol>,
1117 strings: Vec<&'static str>,
1121 fn prefill(init: &[&'static str]) -> Self {
1123 strings: init.into(),
1124 names: init.iter().copied().zip((0..).map(Symbol::new)).collect(),
1125 ..Default::default()
1130 pub fn intern(&mut self, string: &str) -> Symbol {
1131 if let Some(&name) = self.names.get(string) {
1135 let name = Symbol::new(self.strings.len() as u32);
1137 // `from_utf8_unchecked` is safe since we just allocated a `&str` which is known to be
1140 unsafe { str::from_utf8_unchecked(self.arena.alloc_slice(string.as_bytes())) };
1141 // It is safe to extend the arena allocation to `'static` because we only access
1142 // these while the arena is still alive.
1143 let string: &'static str = unsafe { &*(string as *const str) };
1144 self.strings.push(string);
1145 self.names.insert(string, name);
1149 // Get the symbol as a string. `Symbol::as_str()` should be used in
1150 // preference to this function.
1151 pub fn get(&self, symbol: Symbol) -> &str {
1152 self.strings[symbol.0.as_usize()]
1156 // This module has a very short name because it's used a lot.
1157 /// This module contains all the defined keyword `Symbol`s.
1159 /// Given that `kw` is imported, use them like `kw::keyword_name`.
1160 /// For example `kw::Loop` or `kw::Break`.
1166 // This module has a very short name because it's used a lot.
1167 /// This module contains all the defined non-keyword `Symbol`s.
1169 /// Given that `sym` is imported, use them like `sym::symbol_name`.
1170 /// For example `sym::rustfmt` or `sym::u8`.
1171 #[allow(rustc::default_hash_types)]
1174 use std::convert::TryInto;
1178 // Used from a macro in `librustc_feature/accepted.rs`
1179 pub use super::kw::MacroRules as macro_rules;
1181 // Get the symbol for an integer. The first few non-negative integers each
1182 // have a static symbol and therefore are fast.
1183 pub fn integer<N: TryInto<usize> + Copy + ToString>(n: N) -> Symbol {
1184 if let Result::Ok(idx) = n.try_into() {
1185 if let Option::Some(&sym) = digits_array.get(idx) {
1189 Symbol::intern(&n.to_string())
1194 fn is_used_keyword_2018(self) -> bool {
1195 self >= kw::Async && self <= kw::Dyn
1198 fn is_unused_keyword_2018(self) -> bool {
1202 /// Used for sanity checking rustdoc keyword sections.
1203 pub fn is_doc_keyword(self) -> bool {
1207 /// A keyword or reserved identifier that can be used as a path segment.
1208 pub fn is_path_segment_keyword(self) -> bool {
1210 || self == kw::SelfLower
1211 || self == kw::SelfUpper
1212 || self == kw::Crate
1213 || self == kw::PathRoot
1214 || self == kw::DollarCrate
1217 /// Returns `true` if the symbol is `true` or `false`.
1218 pub fn is_bool_lit(self) -> bool {
1219 self == kw::True || self == kw::False
1222 /// This symbol can be a raw identifier.
1223 pub fn can_be_raw(self) -> bool {
1224 self != kw::Invalid && self != kw::Underscore && !self.is_path_segment_keyword()
1229 // Returns `true` for reserved identifiers used internally for elided lifetimes,
1230 // unnamed method parameters, crate root module, error recovery etc.
1231 pub fn is_special(self) -> bool {
1232 self.name <= kw::Underscore
1235 /// Returns `true` if the token is a keyword used in the language.
1236 pub fn is_used_keyword(self) -> bool {
1237 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1238 self.name >= kw::As && self.name <= kw::While
1239 || self.name.is_used_keyword_2018() && self.span.rust_2018()
1242 /// Returns `true` if the token is a keyword reserved for possible future use.
1243 pub fn is_unused_keyword(self) -> bool {
1244 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1245 self.name >= kw::Abstract && self.name <= kw::Yield
1246 || self.name.is_unused_keyword_2018() && self.span.rust_2018()
1249 /// Returns `true` if the token is either a special identifier or a keyword.
1250 pub fn is_reserved(self) -> bool {
1251 self.is_special() || self.is_used_keyword() || self.is_unused_keyword()
1254 /// A keyword or reserved identifier that can be used as a path segment.
1255 pub fn is_path_segment_keyword(self) -> bool {
1256 self.name.is_path_segment_keyword()
1259 /// We see this identifier in a normal identifier position, like variable name or a type.
1260 /// How was it written originally? Did it use the raw form? Let's try to guess.
1261 pub fn is_raw_guess(self) -> bool {
1262 self.name.can_be_raw() && self.is_reserved()
1267 fn with_interner<T, F: FnOnce(&mut Interner) -> T>(f: F) -> T {
1268 GLOBALS.with(|globals| f(&mut *globals.symbol_interner.lock()))
1271 /// An alternative to `Symbol`, useful when the chars within the symbol need to
1272 /// be accessed. It deliberately has limited functionality and should only be
1273 /// used for temporary values.
1275 /// Because the interner outlives any thread which uses this type, we can
1276 /// safely treat `string` which points to interner data, as an immortal string,
1277 /// as long as this type never crosses between threads.
1279 // FIXME: ensure that the interner outlives any thread which uses `SymbolStr`,
1280 // by creating a new thread right after constructing the interner.
1281 #[derive(Clone, Eq, PartialOrd, Ord)]
1282 pub struct SymbolStr {
1283 string: &'static str,
1286 // This impl allows a `SymbolStr` to be directly equated with a `String` or
1288 impl<T: std::ops::Deref<Target = str>> std::cmp::PartialEq<T> for SymbolStr {
1289 fn eq(&self, other: &T) -> bool {
1290 self.string == other.deref()
1294 impl !Send for SymbolStr {}
1295 impl !Sync for SymbolStr {}
1297 /// This impl means that if `ss` is a `SymbolStr`:
1298 /// - `*ss` is a `str`;
1299 /// - `&*ss` is a `&str`;
1300 /// - `&ss as &str` is a `&str`, which means that `&ss` can be passed to a
1301 /// function expecting a `&str`.
1302 impl std::ops::Deref for SymbolStr {
1305 fn deref(&self) -> &str {
1310 impl fmt::Debug for SymbolStr {
1311 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1312 fmt::Debug::fmt(self.string, f)
1316 impl fmt::Display for SymbolStr {
1317 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1318 fmt::Display::fmt(self.string, f)
1322 impl<CTX> HashStable<CTX> for SymbolStr {
1324 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
1325 self.string.hash_stable(hcx, hasher)
1329 impl<CTX> ToStableHashKey<CTX> for SymbolStr {
1330 type KeyType = SymbolStr;
1333 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {