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,
126 advanced_slice_patterns,
138 allow_internal_unsafe,
139 allow_internal_unstable,
140 allow_internal_unstable_backcompat_hack,
144 arbitrary_enum_discriminant,
145 arbitrary_self_types,
152 associated_type_bounds,
153 associated_type_defaults,
161 augmented_assignments,
162 automatically_derived,
163 avx512_target_feature,
168 bind_by_move_pattern_guards,
172 borrowck_graphviz_format,
173 borrowck_graphviz_postflow,
174 borrowck_graphviz_preflow,
177 braced_empty_structs,
189 cfg_target_has_atomic,
190 cfg_target_thread_local,
198 closure_to_fn_coercion,
200 cmpxchg16b_target_feature,
207 conservative_impl_trait,
209 const_compare_raw_pointers,
217 const_in_array_repeat_expressions,
223 const_raw_ptr_to_usize_cast,
225 const_trait_bound_opt_out,
239 crate_visibility_modifier,
247 custom_inner_attributes,
248 custom_test_frameworks,
256 default_lib_allocator,
257 default_type_parameter_fallback,
259 delay_span_bug_from_inside_query,
274 document_private_items,
275 dotdoteq_in_patterns,
276 dotdot_in_tuple_patterns,
277 double_braced_crate: "{{crate}}",
278 double_braced_impl: "{{impl}}",
279 double_braced_misc: "{{misc}}",
280 double_braced_closure: "{{closure}}",
281 double_braced_constructor: "{{constructor}}",
282 double_braced_constant: "{{constant}}",
283 double_braced_opaque: "{{opaque}}",
285 dropck_parametricity,
301 exclusive_range_pattern,
302 exhaustive_integer_patterns,
308 extern_absolute_paths,
310 extern_crate_item_prelude,
321 field_init_shorthand,
345 generic_associated_types,
350 half_open_range_patterns,
355 hexagon_target_feature,
357 homogeneous_aggregate,
371 if_while_or_patterns,
373 impl_header_lifetime_elision,
375 impl_trait_in_bindings,
383 inclusive_range_syntax,
384 infer_outlives_requirements,
385 infer_static_outlives_requirements,
392 irrefutable_let_patterns,
395 issue_5723_bootstrap,
396 issue_tracker_base_url,
398 item_context: "ItemContext",
417 link_llvm_intrinsics,
427 macro_at_most_once_rep,
430 macro_lifetime_matcher,
431 macro_literal_matcher,
441 match_beginning_vert,
442 match_default_bindings,
462 movbe_target_feature,
500 object_safe_for_dispatch,
502 omit_gdb_pretty_printer_section,
509 optin_builtin_traits,
519 overlapping_marker_traits,
524 panic_implementation,
544 poll_with_tls_context,
545 powerpc_target_feature,
546 precise_pointer_size_matching,
551 proc_dash_macro: "proc-macro",
553 proc_macro_attribute,
559 proc_macro_internals,
561 proc_macro_non_items,
562 proc_macro_path_invoc,
567 quad_precision_float,
582 reexport_test_harness_main,
595 re_rebalance_coherence,
613 rustc_allocator_nounwind,
614 rustc_allow_const_fn_ptr,
615 rustc_args_required_const,
619 rustc_const_unstable,
621 rustc_conversion_suggestion,
624 rustc_diagnostic_item,
625 rustc_diagnostic_macros,
628 rustc_dump_env_program_clauses,
629 rustc_dump_program_clauses,
630 rustc_dump_user_substs,
632 rustc_expected_cgu_reuse,
633 rustc_if_this_changed,
634 rustc_inherit_overflow_checks,
636 rustc_layout_scalar_valid_range_end,
637 rustc_layout_scalar_valid_range_start,
638 rustc_macro_transparency,
640 rustc_nonnull_optimization_guaranteed,
641 rustc_object_lifetime_default,
642 rustc_on_unimplemented,
645 rustc_partition_codegened,
646 rustc_partition_reused,
648 rustc_peek_definite_init,
649 rustc_peek_maybe_init,
650 rustc_peek_maybe_uninit,
651 rustc_peek_indirectly_mutable,
653 rustc_proc_macro_decls,
657 rustc_std_internal_symbol,
660 rustc_reservation_impl,
662 rustc_then_this_would_need,
666 rust_eh_unwind_resume,
668 rvalue_static_promotion,
692 sse4a_target_feature,
705 stmt_expr_attributes,
707 struct_field_attributes,
717 target_has_atomic_load_store,
722 termination_trait_test,
725 test_accepted_feature,
727 test_removed_feature,
750 type_alias_impl_trait,
755 type_alias_enum_variants,
772 underscore_const_names,
774 underscore_lifetimes,
776 universal_impl_trait,
779 unrestricted_attribute_tokens,
782 unsized_tuple_coercion,
798 visible_private_types,
813 #[derive(Copy, Clone, Eq, HashStable_Generic)]
821 /// Constructs a new identifier from a symbol and a span.
822 pub const fn new(name: Symbol, span: Span) -> Ident {
826 /// Constructs a new identifier with a dummy span.
828 pub const fn with_dummy_span(name: Symbol) -> Ident {
829 Ident::new(name, DUMMY_SP)
833 pub fn invalid() -> Ident {
834 Ident::with_dummy_span(kw::Invalid)
837 /// Maps a string to an identifier with a dummy span.
838 pub fn from_str(string: &str) -> Ident {
839 Ident::with_dummy_span(Symbol::intern(string))
842 /// Maps a string and a span to an identifier.
843 pub fn from_str_and_span(string: &str, span: Span) -> Ident {
844 Ident::new(Symbol::intern(string), span)
847 /// Replaces `lo` and `hi` with those from `span`, but keep hygiene context.
848 pub fn with_span_pos(self, span: Span) -> Ident {
849 Ident::new(self.name, span.with_ctxt(self.span.ctxt()))
852 pub fn without_first_quote(self) -> Ident {
853 Ident::new(Symbol::intern(self.as_str().trim_start_matches('\'')), self.span)
856 /// "Normalize" ident for use in comparisons using "item hygiene".
857 /// Identifiers with same string value become same if they came from the same "modern" macro
858 /// (e.g., `macro` item, but not `macro_rules` item) and stay different if they came from
859 /// different "modern" macros.
860 /// Technically, this operation strips all non-opaque marks from ident's syntactic context.
861 pub fn modern(self) -> Ident {
862 Ident::new(self.name, self.span.modern())
865 /// "Normalize" ident for use in comparisons using "local variable hygiene".
866 /// Identifiers with same string value become same if they came from the same non-transparent
867 /// macro (e.g., `macro` or `macro_rules!` items) and stay different if they came from different
868 /// non-transparent macros.
869 /// Technically, this operation strips all transparent marks from ident's syntactic context.
870 pub fn modern_and_legacy(self) -> Ident {
871 Ident::new(self.name, self.span.modern_and_legacy())
874 /// Convert the name to a `SymbolStr`. This is a slowish operation because
875 /// it requires locking the symbol interner.
876 pub fn as_str(self) -> SymbolStr {
881 impl PartialEq for Ident {
882 fn eq(&self, rhs: &Self) -> bool {
883 self.name == rhs.name && self.span.ctxt() == rhs.span.ctxt()
887 impl Hash for Ident {
888 fn hash<H: Hasher>(&self, state: &mut H) {
889 self.name.hash(state);
890 self.span.ctxt().hash(state);
894 impl fmt::Debug for Ident {
895 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
896 fmt::Display::fmt(self, f)?;
897 fmt::Debug::fmt(&self.span.ctxt(), f)
901 /// This implementation is supposed to be used in error messages, so it's expected to be identical
902 /// to printing the original identifier token written in source code (`token_to_string`),
903 /// except that AST identifiers don't keep the rawness flag, so we have to guess it.
904 impl fmt::Display for Ident {
905 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
906 fmt::Display::fmt(&IdentPrinter::new(self.name, self.is_raw_guess(), None), 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 /// This is the most general way to print identifiers.
931 /// AST pretty-printer is used as a fallback for turning AST structures into token streams for
932 /// proc macros. Additionally, proc macros may stringify their input and expect it survive the
933 /// stringification (especially true for proc macro derives written between Rust 1.15 and 1.30).
934 /// So we need to somehow pretty-print `$crate` in a way preserving at least some of its
935 /// hygiene data, most importantly name of the crate it refers to.
936 /// As a result we print `$crate` as `crate` if it refers to the local crate
937 /// and as `::other_crate_name` if it refers to some other crate.
938 /// Note, that this is only done if the ident token is printed from inside of AST pretty-pringing,
939 /// but not otherwise. Pretty-printing is the only way for proc macros to discover token contents,
940 /// so we should not perform this lossy conversion if the top level call to the pretty-printer was
941 /// done for a token stream or a single token.
942 pub struct IdentPrinter {
945 /// Span used for retrieving the crate name to which `$crate` refers to,
946 /// if this field is `None` then the `$crate` conversion doesn't happen.
947 convert_dollar_crate: Option<Span>,
951 /// The most general `IdentPrinter` constructor. Do not use this.
952 pub fn new(symbol: Symbol, is_raw: bool, convert_dollar_crate: Option<Span>) -> IdentPrinter {
953 IdentPrinter { symbol, is_raw, convert_dollar_crate }
956 /// This implementation is supposed to be used when printing identifiers
957 /// as a part of pretty-printing for larger AST pieces.
958 /// Do not use this either.
959 pub fn for_ast_ident(ident: Ident, is_raw: bool) -> IdentPrinter {
960 IdentPrinter::new(ident.name, is_raw, Some(ident.span))
964 impl fmt::Display for IdentPrinter {
965 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
969 if self.symbol == kw::DollarCrate {
970 if let Some(span) = self.convert_dollar_crate {
971 let converted = span.ctxt().dollar_crate_name();
972 if !converted.is_path_segment_keyword() {
975 return fmt::Display::fmt(&converted, f);
979 fmt::Display::fmt(&self.symbol, f)
983 /// An interned string.
985 /// Internally, a `Symbol` is implemented as an index, and all operations
986 /// (including hashing, equality, and ordering) operate on that index. The use
987 /// of `rustc_index::newtype_index!` means that `Option<Symbol>` only takes up 4 bytes,
988 /// because `rustc_index::newtype_index!` reserves the last 256 values for tagging purposes.
990 /// Note that `Symbol` cannot directly be a `rustc_index::newtype_index!` because it
991 /// implements `fmt::Debug`, `Encodable`, and `Decodable` in special ways.
992 #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
993 pub struct Symbol(SymbolIndex);
995 rustc_index::newtype_index! {
996 pub struct SymbolIndex { .. }
1000 const fn new(n: u32) -> Self {
1001 Symbol(SymbolIndex::from_u32_const(n))
1004 /// Maps a string to its interned representation.
1005 pub fn intern(string: &str) -> Self {
1006 with_interner(|interner| interner.intern(string))
1009 /// Access the symbol's chars. This is a slowish operation because it
1010 /// requires locking the symbol interner.
1011 pub fn with<F: FnOnce(&str) -> R, R>(self, f: F) -> R {
1012 with_interner(|interner| f(interner.get(self)))
1015 /// Convert to a `SymbolStr`. This is a slowish operation because it
1016 /// requires locking the symbol interner.
1017 pub fn as_str(self) -> SymbolStr {
1018 with_interner(|interner| unsafe {
1019 SymbolStr { string: std::mem::transmute::<&str, &str>(interner.get(self)) }
1023 pub fn as_u32(self) -> u32 {
1027 /// This method is supposed to be used in error messages, so it's expected to be
1028 /// identical to printing the original identifier token written in source code
1029 /// (`token_to_string`, `Ident::to_string`), except that symbols don't keep the rawness flag
1030 /// or edition, so we have to guess the rawness using the global edition.
1031 pub fn to_ident_string(self) -> String {
1032 Ident::with_dummy_span(self).to_string()
1036 impl fmt::Debug for Symbol {
1037 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1038 self.with(|str| fmt::Debug::fmt(&str, f))
1042 impl fmt::Display for Symbol {
1043 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1044 self.with(|str| fmt::Display::fmt(&str, f))
1048 impl Encodable for Symbol {
1049 fn encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
1050 self.with(|string| s.emit_str(string))
1054 impl Decodable for Symbol {
1056 fn decode<D: Decoder>(d: &mut D) -> Result<Symbol, D::Error> {
1057 Ok(Symbol::intern(&d.read_str()?))
1061 impl<CTX> HashStable<CTX> for Symbol {
1063 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
1064 self.as_str().hash_stable(hcx, hasher);
1068 impl<CTX> ToStableHashKey<CTX> for Symbol {
1069 type KeyType = SymbolStr;
1072 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {
1077 // The `&'static str`s in this type actually point into the arena.
1079 pub struct Interner {
1080 arena: DroplessArena,
1081 names: FxHashMap<&'static str, Symbol>,
1082 strings: Vec<&'static str>,
1086 fn prefill(init: &[&'static str]) -> Self {
1088 strings: init.into(),
1089 names: init.iter().copied().zip((0..).map(Symbol::new)).collect(),
1090 ..Default::default()
1095 pub fn intern(&mut self, string: &str) -> Symbol {
1096 if let Some(&name) = self.names.get(string) {
1100 let name = Symbol::new(self.strings.len() as u32);
1102 // `from_utf8_unchecked` is safe since we just allocated a `&str` which is known to be
1105 unsafe { str::from_utf8_unchecked(self.arena.alloc_slice(string.as_bytes())) };
1106 // It is safe to extend the arena allocation to `'static` because we only access
1107 // these while the arena is still alive.
1108 let string: &'static str = unsafe { &*(string as *const str) };
1109 self.strings.push(string);
1110 self.names.insert(string, name);
1114 // Get the symbol as a string. `Symbol::as_str()` should be used in
1115 // preference to this function.
1116 pub fn get(&self, symbol: Symbol) -> &str {
1117 self.strings[symbol.0.as_usize()]
1121 // This module has a very short name because it's used a lot.
1127 // This module has a very short name because it's used a lot.
1128 #[allow(rustc::default_hash_types)]
1131 use std::convert::TryInto;
1135 // Used from a macro in `librustc_feature/accepted.rs`
1136 pub use super::kw::MacroRules as macro_rules;
1138 // Get the symbol for an integer. The first few non-negative integers each
1139 // have a static symbol and therefore are fast.
1140 pub fn integer<N: TryInto<usize> + Copy + ToString>(n: N) -> Symbol {
1141 if let Result::Ok(idx) = n.try_into() {
1142 if let Option::Some(&sym) = digits_array.get(idx) {
1146 Symbol::intern(&n.to_string())
1151 fn is_used_keyword_2018(self) -> bool {
1152 self >= kw::Async && self <= kw::Dyn
1155 fn is_unused_keyword_2018(self) -> bool {
1159 /// Used for sanity checking rustdoc keyword sections.
1160 pub fn is_doc_keyword(self) -> bool {
1164 /// A keyword or reserved identifier that can be used as a path segment.
1165 pub fn is_path_segment_keyword(self) -> bool {
1167 || self == kw::SelfLower
1168 || self == kw::SelfUpper
1169 || self == kw::Crate
1170 || self == kw::PathRoot
1171 || self == kw::DollarCrate
1174 /// Returns `true` if the symbol is `true` or `false`.
1175 pub fn is_bool_lit(self) -> bool {
1176 self == kw::True || self == kw::False
1179 /// This symbol can be a raw identifier.
1180 pub fn can_be_raw(self) -> bool {
1181 self != kw::Invalid && self != kw::Underscore && !self.is_path_segment_keyword()
1186 // Returns `true` for reserved identifiers used internally for elided lifetimes,
1187 // unnamed method parameters, crate root module, error recovery etc.
1188 pub fn is_special(self) -> bool {
1189 self.name <= kw::Underscore
1192 /// Returns `true` if the token is a keyword used in the language.
1193 pub fn is_used_keyword(self) -> bool {
1194 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1195 self.name >= kw::As && self.name <= kw::While
1196 || self.name.is_used_keyword_2018() && self.span.rust_2018()
1199 /// Returns `true` if the token is a keyword reserved for possible future use.
1200 pub fn is_unused_keyword(self) -> bool {
1201 // Note: `span.edition()` is relatively expensive, don't call it unless necessary.
1202 self.name >= kw::Abstract && self.name <= kw::Yield
1203 || self.name.is_unused_keyword_2018() && self.span.rust_2018()
1206 /// Returns `true` if the token is either a special identifier or a keyword.
1207 pub fn is_reserved(self) -> bool {
1208 self.is_special() || self.is_used_keyword() || self.is_unused_keyword()
1211 /// A keyword or reserved identifier that can be used as a path segment.
1212 pub fn is_path_segment_keyword(self) -> bool {
1213 self.name.is_path_segment_keyword()
1216 /// We see this identifier in a normal identifier position, like variable name or a type.
1217 /// How was it written originally? Did it use the raw form? Let's try to guess.
1218 pub fn is_raw_guess(self) -> bool {
1219 self.name.can_be_raw() && self.is_reserved()
1224 fn with_interner<T, F: FnOnce(&mut Interner) -> T>(f: F) -> T {
1225 GLOBALS.with(|globals| f(&mut *globals.symbol_interner.lock()))
1228 /// An alternative to `Symbol`, useful when the chars within the symbol need to
1229 /// be accessed. It deliberately has limited functionality and should only be
1230 /// used for temporary values.
1232 /// Because the interner outlives any thread which uses this type, we can
1233 /// safely treat `string` which points to interner data, as an immortal string,
1234 /// as long as this type never crosses between threads.
1236 // FIXME: ensure that the interner outlives any thread which uses `SymbolStr`,
1237 // by creating a new thread right after constructing the interner.
1238 #[derive(Clone, Eq, PartialOrd, Ord)]
1239 pub struct SymbolStr {
1240 string: &'static str,
1243 // This impl allows a `SymbolStr` to be directly equated with a `String` or
1245 impl<T: std::ops::Deref<Target = str>> std::cmp::PartialEq<T> for SymbolStr {
1246 fn eq(&self, other: &T) -> bool {
1247 self.string == other.deref()
1251 impl !Send for SymbolStr {}
1252 impl !Sync for SymbolStr {}
1254 /// This impl means that if `ss` is a `SymbolStr`:
1255 /// - `*ss` is a `str`;
1256 /// - `&*ss` is a `&str`;
1257 /// - `&ss as &str` is a `&str`, which means that `&ss` can be passed to a
1258 /// function expecting a `&str`.
1259 impl std::ops::Deref for SymbolStr {
1262 fn deref(&self) -> &str {
1267 impl fmt::Debug for SymbolStr {
1268 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1269 fmt::Debug::fmt(self.string, f)
1273 impl fmt::Display for SymbolStr {
1274 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1275 fmt::Display::fmt(self.string, f)
1279 impl<CTX> HashStable<CTX> for SymbolStr {
1281 fn hash_stable(&self, hcx: &mut CTX, hasher: &mut StableHasher) {
1282 self.string.hash_stable(hcx, hasher)
1286 impl<CTX> ToStableHashKey<CTX> for SymbolStr {
1287 type KeyType = SymbolStr;
1290 fn to_stable_hash_key(&self, _: &CTX) -> SymbolStr {