1 //! The Rust Linkage Model and Symbol Names
2 //! =======================================
4 //! The semantic model of Rust linkage is, broadly, that "there's no global
5 //! namespace" between crates. Our aim is to preserve the illusion of this
6 //! model despite the fact that it's not *quite* possible to implement on
7 //! modern linkers. We initially didn't use system linkers at all, but have
8 //! been convinced of their utility.
10 //! There are a few issues to handle:
12 //! - Linkers operate on a flat namespace, so we have to flatten names.
13 //! We do this using the C++ namespace-mangling technique. Foo::bar
16 //! - Symbols for distinct items with the same *name* need to get different
17 //! linkage-names. Examples of this are monomorphizations of functions or
18 //! items within anonymous scopes that end up having the same path.
20 //! - Symbols in different crates but with same names "within" the crate need
21 //! to get different linkage-names.
23 //! - Symbol names should be deterministic: Two consecutive runs of the
24 //! compiler over the same code base should produce the same symbol names for
27 //! - Symbol names should not depend on any global properties of the code base,
28 //! so that small modifications to the code base do not result in all symbols
29 //! changing. In previous versions of the compiler, symbol names incorporated
30 //! the SVH (Stable Version Hash) of the crate. This scheme turned out to be
31 //! infeasible when used in conjunction with incremental compilation because
32 //! small code changes would invalidate all symbols generated previously.
34 //! - Even symbols from different versions of the same crate should be able to
35 //! live next to each other without conflict.
37 //! In order to fulfill the above requirements the following scheme is used by
40 //! The main tool for avoiding naming conflicts is the incorporation of a 64-bit
41 //! hash value into every exported symbol name. Anything that makes a difference
42 //! to the symbol being named, but does not show up in the regular path needs to
43 //! be fed into this hash:
45 //! - Different monomorphizations of the same item have the same path but differ
46 //! in their concrete type parameters, so these parameters are part of the
47 //! data being digested for the symbol hash.
49 //! - Rust allows items to be defined in anonymous scopes, such as in
50 //! `fn foo() { { fn bar() {} } { fn bar() {} } }`. Both `bar` functions have
51 //! the path `foo::bar`, since the anonymous scopes do not contribute to the
52 //! path of an item. The compiler already handles this case via so-called
53 //! disambiguating `DefPaths` which use indices to distinguish items with the
54 //! same name. The DefPaths of the functions above are thus `foo[0]::bar[0]`
55 //! and `foo[0]::bar[1]`. In order to incorporate this disambiguation
56 //! information into the symbol name too, these indices are fed into the
57 //! symbol hash, so that the above two symbols would end up with different
60 //! The two measures described above suffice to avoid intra-crate conflicts. In
61 //! order to also avoid inter-crate conflicts two more measures are taken:
63 //! - The name of the crate containing the symbol is prepended to the symbol
64 //! name, i.e., symbols are "crate qualified". For example, a function `foo` in
65 //! module `bar` in crate `baz` would get a symbol name like
66 //! `baz::bar::foo::{hash}` instead of just `bar::foo::{hash}`. This avoids
67 //! simple conflicts between functions from different crates.
69 //! - In order to be able to also use symbols from two versions of the same
70 //! crate (which naturally also have the same name), a stronger measure is
71 //! required: The compiler accepts an arbitrary "disambiguator" value via the
72 //! `-C metadata` command-line argument. This disambiguator is then fed into
73 //! the symbol hash of every exported item. Consequently, the symbols in two
74 //! identical crates but with different disambiguators are not in conflict
75 //! with each other. This facility is mainly intended to be used by build
78 //! A note on symbol name stability
79 //! -------------------------------
80 //! Previous versions of the compiler resorted to feeding NodeIds into the
81 //! symbol hash in order to disambiguate between items with the same path. The
82 //! current version of the name generation algorithm takes great care not to do
83 //! that, since NodeIds are notoriously unstable: A small change to the
84 //! code base will offset all NodeIds after the change and thus, much as using
85 //! the SVH in the hash, invalidate an unbounded number of symbol names. This
86 //! makes re-using previously compiled code for incremental compilation
87 //! virtually impossible. Thus, symbol hash generation exclusively relies on
88 //! DefPaths which are much more robust in the face of changes to the code base.
90 use rustc::hir::def_id::LOCAL_CRATE;
91 use rustc::hir::CodegenFnAttrFlags;
93 use rustc::mir::mono::{InstantiationMode, MonoItem};
94 use rustc::session::config::SymbolManglingVersion;
95 use rustc::ty::query::Providers;
96 use rustc::ty::{self, Instance, TyCtxt};
98 use rustc_span::symbol::Symbol;
105 pub fn provide(providers: &mut Providers<'_>) {
106 *providers = Providers {
107 symbol_name: |tcx, instance| ty::SymbolName { name: symbol_name(tcx, instance) },
113 fn symbol_name(tcx: TyCtxt<'tcx>, instance: Instance<'tcx>) -> Symbol {
114 let def_id = instance.def_id();
115 let substs = instance.substs;
117 debug!("symbol_name(def_id={:?}, substs={:?})", def_id, substs);
119 let hir_id = tcx.hir().as_local_hir_id(def_id);
121 if def_id.is_local() {
122 if tcx.plugin_registrar_fn(LOCAL_CRATE) == Some(def_id) {
123 let disambiguator = tcx.sess.local_crate_disambiguator();
124 return Symbol::intern(&tcx.sess.generate_plugin_registrar_symbol(disambiguator));
126 if tcx.proc_macro_decls_static(LOCAL_CRATE) == Some(def_id) {
127 let disambiguator = tcx.sess.local_crate_disambiguator();
128 return Symbol::intern(&tcx.sess.generate_proc_macro_decls_symbol(disambiguator));
132 // FIXME(eddyb) Precompute a custom symbol name based on attributes.
133 let is_foreign = if let Some(id) = hir_id {
134 match tcx.hir().get(id) {
135 Node::ForeignItem(_) => true,
139 tcx.is_foreign_item(def_id)
142 let attrs = tcx.codegen_fn_attrs(def_id);
144 // Foreign items by default use no mangling for their symbol name. There's a
145 // few exceptions to this rule though:
147 // * This can be overridden with the `#[link_name]` attribute
149 // * On the wasm32 targets there is a bug (or feature) in LLD [1] where the
150 // same-named symbol when imported from different wasm modules will get
151 // hooked up incorectly. As a result foreign symbols, on the wasm target,
152 // with a wasm import module, get mangled. Additionally our codegen will
153 // deduplicate symbols based purely on the symbol name, but for wasm this
154 // isn't quite right because the same-named symbol on wasm can come from
155 // different modules. For these reasons if `#[link(wasm_import_module)]`
156 // is present we mangle everything on wasm because the demangled form will
157 // show up in the `wasm-import-name` custom attribute in LLVM IR.
159 // [1]: https://bugs.llvm.org/show_bug.cgi?id=44316
161 if tcx.sess.target.target.arch != "wasm32"
162 || !tcx.wasm_import_module_map(def_id.krate).contains_key(&def_id)
164 if let Some(name) = attrs.link_name {
167 return tcx.item_name(def_id);
171 if let Some(name) = attrs.export_name {
176 if attrs.flags.contains(CodegenFnAttrFlags::NO_MANGLE) {
178 return tcx.item_name(def_id);
181 let is_generic = substs.non_erasable_generics().next().is_some();
182 let avoid_cross_crate_conflicts =
183 // If this is an instance of a generic function, we also hash in
184 // the ID of the instantiating crate. This avoids symbol conflicts
185 // in case the same instances is emitted in two crates of the same
189 // If we're dealing with an instance of a function that's inlined from
190 // another crate but we're marking it as globally shared to our
191 // compliation (aka we're not making an internal copy in each of our
192 // codegen units) then this symbol may become an exported (but hidden
193 // visibility) symbol. This means that multiple crates may do the same
194 // and we want to be sure to avoid any symbol conflicts here.
195 match MonoItem::Fn(instance).instantiation_mode(tcx) {
196 InstantiationMode::GloballyShared { may_conflict: true } => true,
200 let instantiating_crate = if avoid_cross_crate_conflicts {
202 if !def_id.is_local() && tcx.sess.opts.share_generics() {
203 // If we are re-using a monomorphization from another crate,
204 // we have to compute the symbol hash accordingly.
205 let upstream_monomorphizations = tcx.upstream_monomorphizations_for(def_id);
207 upstream_monomorphizations
208 .and_then(|monos| monos.get(&substs).cloned())
209 .unwrap_or(LOCAL_CRATE)
220 // Pick the crate responsible for the symbol mangling version, which has to:
221 // 1. be stable for each instance, whether it's being defined or imported
222 // 2. obey each crate's own `-Z symbol-mangling-version`, as much as possible
223 // We solve these as follows:
224 // 1. because symbol names depend on both `def_id` and `instantiating_crate`,
225 // both their `CrateNum`s are stable for any given instance, so we can pick
226 // either and have a stable choice of symbol mangling version
227 // 2. we favor `instantiating_crate` where possible (i.e. when `Some`)
228 let mangling_version_crate = instantiating_crate.unwrap_or(def_id.krate);
229 let mangling_version = if mangling_version_crate == LOCAL_CRATE {
230 tcx.sess.opts.debugging_opts.symbol_mangling_version
232 tcx.symbol_mangling_version(mangling_version_crate)
235 let mangled = match mangling_version {
236 SymbolManglingVersion::Legacy => legacy::mangle(tcx, instance, instantiating_crate),
237 SymbolManglingVersion::V0 => v0::mangle(tcx, instance, instantiating_crate),
240 Symbol::intern(&mangled)