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 rustc_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,
128 advanced_slice_patterns,
140 allow_internal_unsafe,
141 allow_internal_unstable,
142 allow_internal_unstable_backcompat_hack,
146 arbitrary_enum_discriminant,
147 arbitrary_self_types,
156 associated_type_bounds,
157 associated_type_defaults,
167 augmented_assignments,
168 automatically_derived,
169 avx512_target_feature,
174 bind_by_move_pattern_guards,
178 borrowck_graphviz_format,
179 borrowck_graphviz_postflow,
180 borrowck_graphviz_preflow,
183 braced_empty_structs,
196 cfg_target_has_atomic,
197 cfg_target_thread_local,
206 closure_to_fn_coercion,
208 cmpxchg16b_target_feature,
215 conservative_impl_trait,
217 const_compare_raw_pointers,
226 const_in_array_repeat_expressions,
231 const_precise_live_drops,
233 const_raw_ptr_to_usize_cast,
235 const_trait_bound_opt_out,
250 crate_visibility_modifier,
258 custom_inner_attributes,
259 custom_test_frameworks,
268 default_lib_allocator,
269 default_type_parameter_fallback,
271 delay_span_bug_from_inside_query,
286 document_private_items,
287 dotdoteq_in_patterns,
288 dotdot_in_tuple_patterns,
289 double_braced_crate: "{{crate}}",
290 double_braced_impl: "{{impl}}",
291 double_braced_misc: "{{misc}}",
292 double_braced_closure: "{{closure}}",
293 double_braced_constructor: "{{constructor}}",
294 double_braced_constant: "{{constant}}",
295 double_braced_opaque: "{{opaque}}",
297 dropck_parametricity,
312 exclusive_range_pattern,
313 exhaustive_integer_patterns,
319 extern_absolute_paths,
321 extern_crate_item_prelude,
334 field_init_shorthand,
359 generic_associated_types,
365 half_open_range_patterns,
370 hexagon_target_feature,
372 homogeneous_aggregate,
386 if_while_or_patterns,
390 impl_header_lifetime_elision,
392 impl_trait_in_bindings,
400 inclusive_range_syntax,
401 infer_outlives_requirements,
402 infer_static_outlives_requirements,
409 irrefutable_let_patterns,
412 issue_5723_bootstrap,
413 issue_tracker_base_url,
415 item_context: "ItemContext",
436 link_llvm_intrinsics,
447 macro_at_most_once_rep,
450 macro_lifetime_matcher,
451 macro_literal_matcher,
461 match_beginning_vert,
462 match_default_bindings,
483 movbe_target_feature,
525 object_safe_for_dispatch,
528 omit_gdb_pretty_printer_section,
535 optin_builtin_traits,
547 overlapping_marker_traits,
552 panic_implementation,
573 powerpc_target_feature,
574 precise_pointer_size_matching,
580 proc_dash_macro: "proc-macro",
582 proc_macro_attribute,
588 proc_macro_internals,
590 proc_macro_non_items,
591 proc_macro_path_invoc,
600 quad_precision_float,
617 reexport_test_harness_main,
630 re_rebalance_coherence,
635 riscv_target_feature,
649 rustc_allocator_nounwind,
650 rustc_allow_const_fn_ptr,
651 rustc_args_required_const,
655 rustc_const_unstable,
657 rustc_conversion_suggestion,
660 rustc_diagnostic_item,
661 rustc_diagnostic_macros,
664 rustc_dump_env_program_clauses,
665 rustc_dump_program_clauses,
666 rustc_dump_user_substs,
668 rustc_expected_cgu_reuse,
669 rustc_if_this_changed,
670 rustc_inherit_overflow_checks,
672 rustc_layout_scalar_valid_range_end,
673 rustc_layout_scalar_valid_range_start,
674 rustc_macro_transparency,
676 rustc_nonnull_optimization_guaranteed,
677 rustc_object_lifetime_default,
678 rustc_on_unimplemented,
681 rustc_partition_codegened,
682 rustc_partition_reused,
684 rustc_peek_definite_init,
686 rustc_peek_maybe_init,
687 rustc_peek_maybe_uninit,
688 rustc_peek_indirectly_mutable,
690 rustc_proc_macro_decls,
693 rustc_unsafe_specialization_marker,
694 rustc_specialization_trait,
696 rustc_std_internal_symbol,
699 rustc_reservation_impl,
701 rustc_then_this_would_need,
706 rvalue_static_promotion,
729 sse4a_target_feature,
742 stmt_expr_attributes,
744 struct_field_attributes,
756 target_has_atomic_load_store,
762 termination_trait_test,
765 test_accepted_feature,
767 test_removed_feature,
790 type_alias_impl_trait,
795 type_alias_enum_variants,
812 underscore_const_names,
814 underscore_lifetimes,
816 universal_impl_trait,
820 unrestricted_attribute_tokens,
821 unsafe_block_in_unsafe_fn,
824 unsized_tuple_coercion,
842 visible_private_types,
857 #[derive(Copy, Clone, Eq, HashStable_Generic)]
865 /// Constructs a new identifier from a symbol and a span.
866 pub const fn new(name: Symbol, span: Span) -> Ident {
870 /// Constructs a new identifier with a dummy span.
872 pub const fn with_dummy_span(name: Symbol) -> Ident {
873 Ident::new(name, DUMMY_SP)
877 pub fn invalid() -> Ident {
878 Ident::with_dummy_span(kw::Invalid)
881 /// Maps a string to an identifier with a dummy span.
882 pub fn from_str(string: &str) -> Ident {
883 Ident::with_dummy_span(Symbol::intern(string))
886 /// Maps a string and a span to an identifier.
887 pub fn from_str_and_span(string: &str, span: Span) -> Ident {
888 Ident::new(Symbol::intern(string), span)
891 /// Replaces `lo` and `hi` with those from `span`, but keep hygiene context.
892 pub fn with_span_pos(self, span: Span) -> Ident {
893 Ident::new(self.name, span.with_ctxt(self.span.ctxt()))
896 pub fn without_first_quote(self) -> Ident {
897 Ident::new(Symbol::intern(self.as_str().trim_start_matches('\'')), self.span)
900 /// "Normalize" ident for use in comparisons using "item hygiene".
901 /// Identifiers with same string value become same if they came from the same macro 2.0 macro
902 /// (e.g., `macro` item, but not `macro_rules` item) and stay different if they came from
903 /// different macro 2.0 macros.
904 /// Technically, this operation strips all non-opaque marks from ident's syntactic context.
905 pub fn normalize_to_macros_2_0(self) -> Ident {
906 Ident::new(self.name, self.span.normalize_to_macros_2_0())
909 /// "Normalize" ident for use in comparisons using "local variable hygiene".
910 /// Identifiers with same string value become same if they came from the same non-transparent
911 /// macro (e.g., `macro` or `macro_rules!` items) and stay different if they came from different
912 /// non-transparent macros.
913 /// Technically, this operation strips all transparent marks from ident's syntactic context.
914 pub fn normalize_to_macro_rules(self) -> Ident {
915 Ident::new(self.name, self.span.normalize_to_macro_rules())
918 /// Convert the name to a `SymbolStr`. This is a slowish operation because
919 /// it requires locking the symbol interner.
920 pub fn as_str(self) -> SymbolStr {
925 impl PartialEq for Ident {
926 fn eq(&self, rhs: &Self) -> bool {
927 self.name == rhs.name && self.span.ctxt() == rhs.span.ctxt()
931 impl Hash for Ident {
932 fn hash<H: Hasher>(&self, state: &mut H) {
933 self.name.hash(state);
934 self.span.ctxt().hash(state);
938 impl fmt::Debug for Ident {
939 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
940 fmt::Display::fmt(self, f)?;
941 fmt::Debug::fmt(&self.span.ctxt(), f)
945 /// This implementation is supposed to be used in error messages, so it's expected to be identical
946 /// to printing the original identifier token written in source code (`token_to_string`),
947 /// except that AST identifiers don't keep the rawness flag, so we have to guess it.
948 impl fmt::Display for Ident {
949 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
950 fmt::Display::fmt(&IdentPrinter::new(self.name, self.is_raw_guess(), None), f)
954 impl UseSpecializedEncodable for Ident {
955 fn default_encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
956 s.emit_struct("Ident", 2, |s| {
957 s.emit_struct_field("name", 0, |s| self.name.encode(s))?;
958 s.emit_struct_field("span", 1, |s| self.span.encode(s))
963 impl UseSpecializedDecodable for Ident {
964 fn default_decode<D: Decoder>(d: &mut D) -> Result<Self, D::Error> {
965 d.read_struct("Ident", 2, |d| {
967 name: d.read_struct_field("name", 0, Decodable::decode)?,
968 span: d.read_struct_field("span", 1, Decodable::decode)?,
974 /// This is the most general way to print identifiers.
975 /// AST pretty-printer is used as a fallback for turning AST structures into token streams for
976 /// proc macros. Additionally, proc macros may stringify their input and expect it survive the
977 /// stringification (especially true for proc macro derives written between Rust 1.15 and 1.30).
978 /// So we need to somehow pretty-print `$crate` in a way preserving at least some of its
979 /// hygiene data, most importantly name of the crate it refers to.
980 /// As a result we print `$crate` as `crate` if it refers to the local crate
981 /// and as `::other_crate_name` if it refers to some other crate.
982 /// Note, that this is only done if the ident token is printed from inside of AST pretty-pringing,
983 /// but not otherwise. Pretty-printing is the only way for proc macros to discover token contents,
984 /// so we should not perform this lossy conversion if the top level call to the pretty-printer was
985 /// done for a token stream or a single token.
986 pub struct IdentPrinter {
989 /// Span used for retrieving the crate name to which `$crate` refers to,
990 /// if this field is `None` then the `$crate` conversion doesn't happen.
991 convert_dollar_crate: Option<Span>,
995 /// The most general `IdentPrinter` constructor. Do not use this.
996 pub fn new(symbol: Symbol, is_raw: bool, convert_dollar_crate: Option<Span>) -> IdentPrinter {
997 IdentPrinter { symbol, is_raw, convert_dollar_crate }
1000 /// This implementation is supposed to be used when printing identifiers
1001 /// as a part of pretty-printing for larger AST pieces.
1002 /// Do not use this either.
1003 pub fn for_ast_ident(ident: Ident, is_raw: bool) -> IdentPrinter {
1004 IdentPrinter::new(ident.name, is_raw, Some(ident.span))
1008 impl fmt::Display for IdentPrinter {
1009 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1013 if self.symbol == kw::DollarCrate {
1014 if let Some(span) = self.convert_dollar_crate {
1015 let converted = span.ctxt().dollar_crate_name();
1016 if !converted.is_path_segment_keyword() {
1019 return fmt::Display::fmt(&converted, f);
1023 fmt::Display::fmt(&self.symbol, f)
1027 /// An newtype around `Ident` that calls [Ident::normalize_to_macro_rules] on
1029 // FIXME(matthewj, petrochenkov) Use this more often, add a similar
1030 // `ModernIdent` struct and use that as well.
1031 #[derive(Copy, Clone, Eq, PartialEq, Hash)]
1032 pub struct MacroRulesNormalizedIdent(Ident);
1034 impl MacroRulesNormalizedIdent {
1035 pub fn new(ident: Ident) -> Self {
1036 Self(ident.normalize_to_macro_rules())
1040 impl fmt::Debug for MacroRulesNormalizedIdent {
1041 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1042 fmt::Debug::fmt(&self.0, f)
1046 impl fmt::Display for MacroRulesNormalizedIdent {
1047 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1048 fmt::Display::fmt(&self.0, f)
1052 /// An interned string.
1054 /// Internally, a `Symbol` is implemented as an index, and all operations
1055 /// (including hashing, equality, and ordering) operate on that index. The use
1056 /// of `rustc_index::newtype_index!` means that `Option<Symbol>` only takes up 4 bytes,
1057 /// because `rustc_index::newtype_index!` reserves the last 256 values for tagging purposes.
1059 /// Note that `Symbol` cannot directly be a `rustc_index::newtype_index!` because it
1060 /// implements `fmt::Debug`, `Encodable`, and `Decodable` in special ways.
1061 #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
1062 pub struct Symbol(SymbolIndex);
1064 rustc_index::newtype_index! {
1065 pub struct SymbolIndex { .. }
1069 const fn new(n: u32) -> Self {
1070 Symbol(SymbolIndex::from_u32(n))
1073 /// Maps a string to its interned representation.
1074 pub fn intern(string: &str) -> Self {
1075 with_interner(|interner| interner.intern(string))
1078 /// Access the symbol's chars. This is a slowish operation because it
1079 /// requires locking the symbol interner.
1080 pub fn with<F: FnOnce(&str) -> R, R>(self, f: F) -> R {
1081 with_interner(|interner| f(interner.get(self)))
1084 /// Convert to a `SymbolStr`. This is a slowish operation because it
1085 /// requires locking the symbol interner.
1086 pub fn as_str(self) -> SymbolStr {
1087 with_interner(|interner| unsafe {
1088 SymbolStr { string: std::mem::transmute::<&str, &str>(interner.get(self)) }
1092 pub fn as_u32(self) -> u32 {
1096 /// This method is supposed to be used in error messages, so it's expected to be
1097 /// identical to printing the original identifier token written in source code
1098 /// (`token_to_string`, `Ident::to_string`), except that symbols don't keep the rawness flag
1099 /// or edition, so we have to guess the rawness using the global edition.
1100 pub fn to_ident_string(self) -> String {
1101 Ident::with_dummy_span(self).to_string()
1105 impl fmt::Debug for Symbol {
1106 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1107 self.with(|str| fmt::Debug::fmt(&str, f))
1111 impl fmt::Display for Symbol {
1112 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1113 self.with(|str| fmt::Display::fmt(&str, f))
1117 impl Encodable for Symbol {
1118 fn encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
1119 self.with(|string| s.emit_str(string))
1123 impl Decodable for Symbol {
1125 fn decode<D: Decoder>(d: &mut D) -> Result<Symbol, D::Error> {
1126 Ok(Symbol::intern(&d.read_str()?))
1130 impl<CTX> HashStable<CTX> for Symbol {
1132 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
1133 self.as_str().hash_stable(hcx, hasher);
1137 impl<CTX> ToStableHashKey<CTX> for Symbol {
1138 type KeyType = SymbolStr;
1141 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {
1146 // The `&'static str`s in this type actually point into the arena.
1148 pub struct Interner {
1149 arena: DroplessArena,
1150 names: FxHashMap<&'static str, Symbol>,
1151 strings: Vec<&'static str>,
1155 fn prefill(init: &[&'static str]) -> Self {
1157 strings: init.into(),
1158 names: init.iter().copied().zip((0..).map(Symbol::new)).collect(),
1159 ..Default::default()
1164 pub fn intern(&mut self, string: &str) -> Symbol {
1165 if let Some(&name) = self.names.get(string) {
1169 let name = Symbol::new(self.strings.len() as u32);
1171 // `from_utf8_unchecked` is safe since we just allocated a `&str` which is known to be
1174 unsafe { str::from_utf8_unchecked(self.arena.alloc_slice(string.as_bytes())) };
1175 // It is safe to extend the arena allocation to `'static` because we only access
1176 // these while the arena is still alive.
1177 let string: &'static str = unsafe { &*(string as *const str) };
1178 self.strings.push(string);
1179 self.names.insert(string, name);
1183 // Get the symbol as a string. `Symbol::as_str()` should be used in
1184 // preference to this function.
1185 pub fn get(&self, symbol: Symbol) -> &str {
1186 self.strings[symbol.0.as_usize()]
1190 // This module has a very short name because it's used a lot.
1191 /// This module contains all the defined keyword `Symbol`s.
1193 /// Given that `kw` is imported, use them like `kw::keyword_name`.
1194 /// For example `kw::Loop` or `kw::Break`.
1200 // This module has a very short name because it's used a lot.
1201 /// This module contains all the defined non-keyword `Symbol`s.
1203 /// Given that `sym` is imported, use them like `sym::symbol_name`.
1204 /// For example `sym::rustfmt` or `sym::u8`.
1205 #[allow(rustc::default_hash_types)]
1208 use std::convert::TryInto;
1212 // Used from a macro in `librustc_feature/accepted.rs`
1213 pub use super::kw::MacroRules as macro_rules;
1215 // Get the symbol for an integer. The first few non-negative integers each
1216 // have a static symbol and therefore are fast.
1217 pub fn integer<N: TryInto<usize> + Copy + ToString>(n: N) -> Symbol {
1218 if let Result::Ok(idx) = n.try_into() {
1219 if let Option::Some(&sym_) = digits_array.get(idx) {
1223 Symbol::intern(&n.to_string())
1228 fn is_used_keyword_2018(self) -> bool {
1229 self >= kw::Async && self <= kw::Dyn
1232 fn is_unused_keyword_2018(self) -> bool {
1236 /// Used for sanity checking rustdoc keyword sections.
1237 pub fn is_doc_keyword(self) -> bool {
1241 /// A keyword or reserved identifier that can be used as a path segment.
1242 pub fn is_path_segment_keyword(self) -> bool {
1244 || self == kw::SelfLower
1245 || self == kw::SelfUpper
1246 || self == kw::Crate
1247 || self == kw::PathRoot
1248 || self == kw::DollarCrate
1251 /// Returns `true` if the symbol is `true` or `false`.
1252 pub fn is_bool_lit(self) -> bool {
1253 self == kw::True || self == kw::False
1256 /// This symbol can be a raw identifier.
1257 pub fn can_be_raw(self) -> bool {
1258 self != kw::Invalid && self != kw::Underscore && !self.is_path_segment_keyword()
1263 // Returns `true` for reserved identifiers used internally for elided lifetimes,
1264 // unnamed method parameters, crate root module, error recovery etc.
1265 pub fn is_special(self) -> bool {
1266 self.name <= kw::Underscore
1269 /// Returns `true` if the token is a keyword used in the language.
1270 pub fn is_used_keyword(self) -> bool {
1271 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1272 self.name >= kw::As && self.name <= kw::While
1273 || self.name.is_used_keyword_2018() && self.span.rust_2018()
1276 /// Returns `true` if the token is a keyword reserved for possible future use.
1277 pub fn is_unused_keyword(self) -> bool {
1278 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1279 self.name >= kw::Abstract && self.name <= kw::Yield
1280 || self.name.is_unused_keyword_2018() && self.span.rust_2018()
1283 /// Returns `true` if the token is either a special identifier or a keyword.
1284 pub fn is_reserved(self) -> bool {
1285 self.is_special() || self.is_used_keyword() || self.is_unused_keyword()
1288 /// A keyword or reserved identifier that can be used as a path segment.
1289 pub fn is_path_segment_keyword(self) -> bool {
1290 self.name.is_path_segment_keyword()
1293 /// We see this identifier in a normal identifier position, like variable name or a type.
1294 /// How was it written originally? Did it use the raw form? Let's try to guess.
1295 pub fn is_raw_guess(self) -> bool {
1296 self.name.can_be_raw() && self.is_reserved()
1301 fn with_interner<T, F: FnOnce(&mut Interner) -> T>(f: F) -> T {
1302 GLOBALS.with(|globals| f(&mut *globals.symbol_interner.lock()))
1305 /// An alternative to `Symbol`, useful when the chars within the symbol need to
1306 /// be accessed. It deliberately has limited functionality and should only be
1307 /// used for temporary values.
1309 /// Because the interner outlives any thread which uses this type, we can
1310 /// safely treat `string` which points to interner data, as an immortal string,
1311 /// as long as this type never crosses between threads.
1313 // FIXME: ensure that the interner outlives any thread which uses `SymbolStr`,
1314 // by creating a new thread right after constructing the interner.
1315 #[derive(Clone, Eq, PartialOrd, Ord)]
1316 pub struct SymbolStr {
1317 string: &'static str,
1320 // This impl allows a `SymbolStr` to be directly equated with a `String` or
1322 impl<T: std::ops::Deref<Target = str>> std::cmp::PartialEq<T> for SymbolStr {
1323 fn eq(&self, other: &T) -> bool {
1324 self.string == other.deref()
1328 impl !Send for SymbolStr {}
1329 impl !Sync for SymbolStr {}
1331 /// This impl means that if `ss` is a `SymbolStr`:
1332 /// - `*ss` is a `str`;
1333 /// - `&*ss` is a `&str`;
1334 /// - `&ss as &str` is a `&str`, which means that `&ss` can be passed to a
1335 /// function expecting a `&str`.
1336 impl std::ops::Deref for SymbolStr {
1339 fn deref(&self) -> &str {
1344 impl fmt::Debug for SymbolStr {
1345 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1346 fmt::Debug::fmt(self.string, f)
1350 impl fmt::Display for SymbolStr {
1351 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1352 fmt::Display::fmt(self.string, f)
1356 impl<CTX> HashStable<CTX> for SymbolStr {
1358 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
1359 self.string.hash_stable(hcx, hasher)
1363 impl<CTX> ToStableHashKey<CTX> for SymbolStr {
1364 type KeyType = SymbolStr;
1367 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {