1 //! Mono Item Collection
2 //! ====================
4 //! This module is responsible for discovering all items that will contribute
5 //! to code generation of the crate. The important part here is that it not only
6 //! needs to find syntax-level items (functions, structs, etc) but also all
7 //! their monomorphized instantiations. Every non-generic, non-const function
8 //! maps to one LLVM artifact. Every generic function can produce
9 //! from zero to N artifacts, depending on the sets of type arguments it
10 //! is instantiated with.
11 //! This also applies to generic items from other crates: A generic definition
12 //! in crate X might produce monomorphizations that are compiled into crate Y.
13 //! We also have to collect these here.
15 //! The following kinds of "mono items" are handled here:
23 //! The following things also result in LLVM artifacts, but are not collected
24 //! here, since we instantiate them locally on demand when needed in a given
34 //! Let's define some terms first:
36 //! - A "mono item" is something that results in a function or global in
37 //! the LLVM IR of a codegen unit. Mono items do not stand on their
38 //! own, they can reference other mono items. For example, if function
39 //! `foo()` calls function `bar()` then the mono item for `foo()`
40 //! references the mono item for function `bar()`. In general, the
41 //! definition for mono item A referencing a mono item B is that
42 //! the LLVM artifact produced for A references the LLVM artifact produced
45 //! - Mono items and the references between them form a directed graph,
46 //! where the mono items are the nodes and references form the edges.
47 //! Let's call this graph the "mono item graph".
49 //! - The mono item graph for a program contains all mono items
50 //! that are needed in order to produce the complete LLVM IR of the program.
52 //! The purpose of the algorithm implemented in this module is to build the
53 //! mono item graph for the current crate. It runs in two phases:
55 //! 1. Discover the roots of the graph by traversing the HIR of the crate.
56 //! 2. Starting from the roots, find neighboring nodes by inspecting the MIR
57 //! representation of the item corresponding to a given node, until no more
58 //! new nodes are found.
60 //! ### Discovering roots
62 //! The roots of the mono item graph correspond to the public non-generic
63 //! syntactic items in the source code. We find them by walking the HIR of the
64 //! crate, and whenever we hit upon a public function, method, or static item,
65 //! we create a mono item consisting of the items DefId and, since we only
66 //! consider non-generic items, an empty type-substitution set. (In eager
67 //! collection mode, during incremental compilation, all non-generic functions
68 //! are considered as roots, as well as when the `-Clink-dead-code` option is
69 //! specified. Functions marked `#[no_mangle]` and functions called by inlinable
70 //! functions also always act as roots.)
72 //! ### Finding neighbor nodes
73 //! Given a mono item node, we can discover neighbors by inspecting its
74 //! MIR. We walk the MIR and any time we hit upon something that signifies a
75 //! reference to another mono item, we have found a neighbor. Since the
76 //! mono item we are currently at is always monomorphic, we also know the
77 //! concrete type arguments of its neighbors, and so all neighbors again will be
78 //! monomorphic. The specific forms a reference to a neighboring node can take
79 //! in MIR are quite diverse. Here is an overview:
81 //! #### Calling Functions/Methods
82 //! The most obvious form of one mono item referencing another is a
83 //! function or method call (represented by a CALL terminator in MIR). But
84 //! calls are not the only thing that might introduce a reference between two
85 //! function mono items, and as we will see below, they are just a
86 //! specialization of the form described next, and consequently will not get any
87 //! special treatment in the algorithm.
89 //! #### Taking a reference to a function or method
90 //! A function does not need to actually be called in order to be a neighbor of
91 //! another function. It suffices to just take a reference in order to introduce
92 //! an edge. Consider the following example:
95 //! # use core::fmt::Display;
96 //! fn print_val<T: Display>(x: T) {
97 //! println!("{}", x);
100 //! fn call_fn(f: &dyn Fn(i32), x: i32) {
105 //! let print_i32 = print_val::<i32>;
106 //! call_fn(&print_i32, 0);
109 //! The MIR of none of these functions will contain an explicit call to
110 //! `print_val::<i32>`. Nonetheless, in order to mono this program, we need
111 //! an instance of this function. Thus, whenever we encounter a function or
112 //! method in operand position, we treat it as a neighbor of the current
113 //! mono item. Calls are just a special case of that.
116 //! In a way, closures are a simple case. Since every closure object needs to be
117 //! constructed somewhere, we can reliably discover them by observing
118 //! `RValue::Aggregate` expressions with `AggregateKind::Closure`. This is also
119 //! true for closures inlined from other crates.
122 //! Drop glue mono items are introduced by MIR drop-statements. The
123 //! generated mono item will again have drop-glue item neighbors if the
124 //! type to be dropped contains nested values that also need to be dropped. It
125 //! might also have a function item neighbor for the explicit `Drop::drop`
126 //! implementation of its type.
128 //! #### Unsizing Casts
129 //! A subtle way of introducing neighbor edges is by casting to a trait object.
130 //! Since the resulting fat-pointer contains a reference to a vtable, we need to
131 //! instantiate all object-save methods of the trait, as we need to store
132 //! pointers to these functions even if they never get called anywhere. This can
133 //! be seen as a special case of taking a function reference.
136 //! Since `Box` expression have special compiler support, no explicit calls to
137 //! `exchange_malloc()` and `box_free()` may show up in MIR, even if the
138 //! compiler will generate them. We have to observe `Rvalue::Box` expressions
139 //! and Box-typed drop-statements for that purpose.
142 //! Interaction with Cross-Crate Inlining
143 //! -------------------------------------
144 //! The binary of a crate will not only contain machine code for the items
145 //! defined in the source code of that crate. It will also contain monomorphic
146 //! instantiations of any extern generic functions and of functions marked with
148 //! The collection algorithm handles this more or less mono. If it is
149 //! about to create a mono item for something with an external `DefId`,
150 //! it will take a look if the MIR for that item is available, and if so just
151 //! proceed normally. If the MIR is not available, it assumes that the item is
152 //! just linked to and no node is created; which is exactly what we want, since
153 //! no machine code should be generated in the current crate for such an item.
155 //! Eager and Lazy Collection Mode
156 //! ------------------------------
157 //! Mono item collection can be performed in one of two modes:
159 //! - Lazy mode means that items will only be instantiated when actually
160 //! referenced. The goal is to produce the least amount of machine code
163 //! - Eager mode is meant to be used in conjunction with incremental compilation
164 //! where a stable set of mono items is more important than a minimal
165 //! one. Thus, eager mode will instantiate drop-glue for every drop-able type
166 //! in the crate, even if no drop call for that type exists (yet). It will
167 //! also instantiate default implementations of trait methods, something that
168 //! otherwise is only done on demand.
173 //! Some things are not yet fully implemented in the current version of this
177 //! Ideally, no mono item should be generated for const fns unless there
178 //! is a call to them that cannot be evaluated at compile time. At the moment
179 //! this is not implemented however: a mono item will be produced
180 //! regardless of whether it is actually needed or not.
182 use rustc_data_structures::fx::{FxHashMap, FxHashSet};
183 use rustc_data_structures::sync::{par_iter, MTLock, MTRef, ParallelIterator};
184 use rustc_hir as hir;
185 use rustc_hir::def::DefKind;
186 use rustc_hir::def_id::{DefId, DefIdMap, LocalDefId};
187 use rustc_hir::lang_items::LangItem;
188 use rustc_index::bit_set::GrowableBitSet;
189 use rustc_middle::mir::interpret::{AllocId, ConstValue};
190 use rustc_middle::mir::interpret::{ErrorHandled, GlobalAlloc, Scalar};
191 use rustc_middle::mir::mono::{InstantiationMode, MonoItem};
192 use rustc_middle::mir::visit::Visitor as MirVisitor;
193 use rustc_middle::mir::{self, Local, Location};
194 use rustc_middle::ty::adjustment::{CustomCoerceUnsized, PointerCast};
195 use rustc_middle::ty::print::with_no_trimmed_paths;
196 use rustc_middle::ty::subst::{GenericArgKind, InternalSubsts};
197 use rustc_middle::ty::{self, GenericParamDefKind, Instance, Ty, TyCtxt, TypeFoldable, VtblEntry};
198 use rustc_middle::{middle::codegen_fn_attrs::CodegenFnAttrFlags, mir::visit::TyContext};
199 use rustc_session::config::EntryFnType;
200 use rustc_session::lint::builtin::LARGE_ASSIGNMENTS;
201 use rustc_session::Limit;
202 use rustc_span::source_map::{dummy_spanned, respan, Span, Spanned, DUMMY_SP};
203 use rustc_target::abi::Size;
206 use std::path::PathBuf;
209 pub enum MonoItemCollectionMode {
214 /// Maps every mono item to all mono items it references in its
216 pub struct InliningMap<'tcx> {
217 // Maps a source mono item to the range of mono items
219 // The range selects elements within the `targets` vecs.
220 index: FxHashMap<MonoItem<'tcx>, Range<usize>>,
221 targets: Vec<MonoItem<'tcx>>,
223 // Contains one bit per mono item in the `targets` field. That bit
224 // is true if that mono item needs to be inlined into every CGU.
225 inlines: GrowableBitSet<usize>,
228 /// Struct to store mono items in each collecting and if they should
229 /// be inlined. We call `instantiation_mode` to get their inlining
230 /// status when inserting new elements, which avoids calling it in
231 /// `inlining_map.lock_mut()`. See the `collect_items_rec` implementation
233 struct MonoItems<'tcx> {
234 // If this is false, we do not need to compute whether items
235 // will need to be inlined.
236 compute_inlining: bool,
238 // The TyCtxt used to determine whether the a item should
242 // The collected mono items. The bool field in each element
243 // indicates whether this element should be inlined.
244 items: Vec<(Spanned<MonoItem<'tcx>>, bool /*inlined*/)>,
247 impl<'tcx> MonoItems<'tcx> {
249 fn push(&mut self, item: Spanned<MonoItem<'tcx>>) {
254 fn extend<T: IntoIterator<Item = Spanned<MonoItem<'tcx>>>>(&mut self, iter: T) {
255 self.items.extend(iter.into_iter().map(|mono_item| {
256 let inlined = if !self.compute_inlining {
259 mono_item.node.instantiation_mode(self.tcx) == InstantiationMode::LocalCopy
266 impl<'tcx> InliningMap<'tcx> {
267 fn new() -> InliningMap<'tcx> {
269 index: FxHashMap::default(),
271 inlines: GrowableBitSet::with_capacity(1024),
275 fn record_accesses<'a>(
277 source: MonoItem<'tcx>,
278 new_targets: &'a [(Spanned<MonoItem<'tcx>>, bool)],
282 let start_index = self.targets.len();
283 let new_items_count = new_targets.len();
284 let new_items_count_total = new_items_count + self.targets.len();
286 self.targets.reserve(new_items_count);
287 self.inlines.ensure(new_items_count_total);
289 for (i, (Spanned { node: mono_item, .. }, inlined)) in new_targets.into_iter().enumerate() {
290 self.targets.push(*mono_item);
292 self.inlines.insert(i + start_index);
296 let end_index = self.targets.len();
297 assert!(self.index.insert(source, start_index..end_index).is_none());
300 // Internally iterate over all items referenced by `source` which will be
301 // made available for inlining.
302 pub fn with_inlining_candidates<F>(&self, source: MonoItem<'tcx>, mut f: F)
304 F: FnMut(MonoItem<'tcx>),
306 if let Some(range) = self.index.get(&source) {
307 for (i, candidate) in self.targets[range.clone()].iter().enumerate() {
308 if self.inlines.contains(range.start + i) {
315 // Internally iterate over all items and the things each accesses.
316 pub fn iter_accesses<F>(&self, mut f: F)
318 F: FnMut(MonoItem<'tcx>, &[MonoItem<'tcx>]),
320 for (&accessor, range) in &self.index {
321 f(accessor, &self.targets[range.clone()])
326 pub fn collect_crate_mono_items(
328 mode: MonoItemCollectionMode,
329 ) -> (FxHashSet<MonoItem<'_>>, InliningMap<'_>) {
330 let _prof_timer = tcx.prof.generic_activity("monomorphization_collector");
333 tcx.sess.time("monomorphization_collector_root_collections", || collect_roots(tcx, mode));
335 debug!("building mono item graph, beginning at roots");
337 let mut visited = MTLock::new(FxHashSet::default());
338 let mut inlining_map = MTLock::new(InliningMap::new());
339 let recursion_limit = tcx.recursion_limit();
342 let visited: MTRef<'_, _> = &mut visited;
343 let inlining_map: MTRef<'_, _> = &mut inlining_map;
345 tcx.sess.time("monomorphization_collector_graph_walk", || {
346 par_iter(roots).for_each(|root| {
347 let mut recursion_depths = DefIdMap::default();
352 &mut recursion_depths,
360 (visited.into_inner(), inlining_map.into_inner())
363 // Find all non-generic items by walking the HIR. These items serve as roots to
364 // start monomorphizing from.
365 fn collect_roots(tcx: TyCtxt<'_>, mode: MonoItemCollectionMode) -> Vec<MonoItem<'_>> {
366 debug!("collecting roots");
367 let mut roots = MonoItems { compute_inlining: false, tcx, items: Vec::new() };
370 let entry_fn = tcx.entry_fn(());
372 debug!("collect_roots: entry_fn = {:?}", entry_fn);
374 let mut collector = RootCollector { tcx, mode, entry_fn, output: &mut roots };
376 let crate_items = tcx.hir_crate_items(());
378 for id in crate_items.items() {
379 collector.process_item(id);
382 for id in crate_items.impl_items() {
383 collector.process_impl_item(id);
386 collector.push_extra_entry_roots();
389 // We can only codegen items that are instantiable - items all of
390 // whose predicates hold. Luckily, items that aren't instantiable
391 // can't actually be used, so we can just skip codegenning them.
395 .filter_map(|(Spanned { node: mono_item, .. }, _)| {
396 mono_item.is_instantiable(tcx).then_some(mono_item)
401 /// Collect all monomorphized items reachable from `starting_point`, and emit a note diagnostic if a
402 /// post-monorphization error is encountered during a collection step.
403 fn collect_items_rec<'tcx>(
405 starting_point: Spanned<MonoItem<'tcx>>,
406 visited: MTRef<'_, MTLock<FxHashSet<MonoItem<'tcx>>>>,
407 recursion_depths: &mut DefIdMap<usize>,
408 recursion_limit: Limit,
409 inlining_map: MTRef<'_, MTLock<InliningMap<'tcx>>>,
411 if !visited.lock_mut().insert(starting_point.node) {
412 // We've been here already, no need to search again.
415 debug!("BEGIN collect_items_rec({})", starting_point.node);
417 let mut neighbors = MonoItems { compute_inlining: true, tcx, items: Vec::new() };
418 let recursion_depth_reset;
421 // Post-monomorphization errors MVP
423 // We can encounter errors while monomorphizing an item, but we don't have a good way of
424 // showing a complete stack of spans ultimately leading to collecting the erroneous one yet.
425 // (It's also currently unclear exactly which diagnostics and information would be interesting
426 // to report in such cases)
428 // This leads to suboptimal error reporting: a post-monomorphization error (PME) will be
429 // shown with just a spanned piece of code causing the error, without information on where
430 // it was called from. This is especially obscure if the erroneous mono item is in a
431 // dependency. See for example issue #85155, where, before minimization, a PME happened two
432 // crates downstream from libcore's stdarch, without a way to know which dependency was the
435 // If such an error occurs in the current crate, its span will be enough to locate the
436 // source. If the cause is in another crate, the goal here is to quickly locate which mono
437 // item in the current crate is ultimately responsible for causing the error.
439 // To give at least _some_ context to the user: while collecting mono items, we check the
440 // error count. If it has changed, a PME occurred, and we trigger some diagnostics about the
441 // current step of mono items collection.
443 // FIXME: don't rely on global state, instead bubble up errors. Note: this is very hard to do.
444 let error_count = tcx.sess.diagnostic().err_count();
446 match starting_point.node {
447 MonoItem::Static(def_id) => {
448 let instance = Instance::mono(tcx, def_id);
450 // Sanity check whether this ended up being collected accidentally
451 debug_assert!(should_codegen_locally(tcx, &instance));
453 let ty = instance.ty(tcx, ty::ParamEnv::reveal_all());
454 visit_drop_use(tcx, ty, true, starting_point.span, &mut neighbors);
456 recursion_depth_reset = None;
458 if let Ok(alloc) = tcx.eval_static_initializer(def_id) {
459 for &id in alloc.inner().relocations().values() {
460 collect_miri(tcx, id, &mut neighbors);
464 MonoItem::Fn(instance) => {
465 // Sanity check whether this ended up being collected accidentally
466 debug_assert!(should_codegen_locally(tcx, &instance));
468 // Keep track of the monomorphization recursion depth
469 recursion_depth_reset = Some(check_recursion_limit(
476 check_type_length_limit(tcx, instance);
478 rustc_data_structures::stack::ensure_sufficient_stack(|| {
479 collect_neighbours(tcx, instance, &mut neighbors);
482 MonoItem::GlobalAsm(item_id) => {
483 recursion_depth_reset = None;
485 let item = tcx.hir().item(item_id);
486 if let hir::ItemKind::GlobalAsm(asm) = item.kind {
487 for (op, op_sp) in asm.operands {
489 hir::InlineAsmOperand::Const { .. } => {
490 // Only constants which resolve to a plain integer
491 // are supported. Therefore the value should not
492 // depend on any other items.
494 hir::InlineAsmOperand::SymFn { anon_const } => {
496 tcx.typeck_body(anon_const.body).node_type(anon_const.hir_id);
497 visit_fn_use(tcx, fn_ty, false, *op_sp, &mut neighbors);
499 hir::InlineAsmOperand::SymStatic { path: _, def_id } => {
500 let instance = Instance::mono(tcx, *def_id);
501 if should_codegen_locally(tcx, &instance) {
502 trace!("collecting static {:?}", def_id);
503 neighbors.push(dummy_spanned(MonoItem::Static(*def_id)));
506 hir::InlineAsmOperand::In { .. }
507 | hir::InlineAsmOperand::Out { .. }
508 | hir::InlineAsmOperand::InOut { .. }
509 | hir::InlineAsmOperand::SplitInOut { .. } => {
510 span_bug!(*op_sp, "invalid operand type for global_asm!")
515 span_bug!(item.span, "Mismatch between hir::Item type and MonoItem type")
520 // Check for PMEs and emit a diagnostic if one happened. To try to show relevant edges of the
522 if tcx.sess.diagnostic().err_count() > error_count
523 && starting_point.node.is_generic_fn()
524 && starting_point.node.is_user_defined()
526 let formatted_item = with_no_trimmed_paths!(starting_point.node.to_string());
527 tcx.sess.span_note_without_error(
529 &format!("the above error was encountered while instantiating `{}`", formatted_item),
532 inlining_map.lock_mut().record_accesses(starting_point.node, &neighbors.items);
534 for (neighbour, _) in neighbors.items {
535 collect_items_rec(tcx, neighbour, visited, recursion_depths, recursion_limit, inlining_map);
538 if let Some((def_id, depth)) = recursion_depth_reset {
539 recursion_depths.insert(def_id, depth);
542 debug!("END collect_items_rec({})", starting_point.node);
545 /// Format instance name that is already known to be too long for rustc.
546 /// Show only the first and last 32 characters to avoid blasting
547 /// the user's terminal with thousands of lines of type-name.
549 /// If the type name is longer than before+after, it will be written to a file.
550 fn shrunk_instance_name<'tcx>(
552 instance: &Instance<'tcx>,
555 ) -> (String, Option<PathBuf>) {
556 let s = instance.to_string();
558 // Only use the shrunk version if it's really shorter.
559 // This also avoids the case where before and after slices overlap.
560 if s.chars().nth(before + after + 1).is_some() {
561 // An iterator of all byte positions including the end of the string.
562 let positions = || s.char_indices().map(|(i, _)| i).chain(iter::once(s.len()));
564 let shrunk = format!(
565 "{before}...{after}",
566 before = &s[..positions().nth(before).unwrap_or(s.len())],
567 after = &s[positions().rev().nth(after).unwrap_or(0)..],
570 let path = tcx.output_filenames(()).temp_path_ext("long-type.txt", None);
571 let written_to_path = std::fs::write(&path, s).ok().map(|_| path);
573 (shrunk, written_to_path)
579 fn check_recursion_limit<'tcx>(
581 instance: Instance<'tcx>,
583 recursion_depths: &mut DefIdMap<usize>,
584 recursion_limit: Limit,
585 ) -> (DefId, usize) {
586 let def_id = instance.def_id();
587 let recursion_depth = recursion_depths.get(&def_id).cloned().unwrap_or(0);
588 debug!(" => recursion depth={}", recursion_depth);
590 let adjusted_recursion_depth = if Some(def_id) == tcx.lang_items().drop_in_place_fn() {
591 // HACK: drop_in_place creates tight monomorphization loops. Give
598 // Code that needs to instantiate the same function recursively
599 // more than the recursion limit is assumed to be causing an
600 // infinite expansion.
601 if !recursion_limit.value_within_limit(adjusted_recursion_depth) {
602 let (shrunk, written_to_path) = shrunk_instance_name(tcx, &instance, 32, 32);
603 let error = format!("reached the recursion limit while instantiating `{}`", shrunk);
604 let mut err = tcx.sess.struct_span_fatal(span, &error);
606 tcx.def_span(def_id),
607 &format!("`{}` defined here", tcx.def_path_str(def_id)),
609 if let Some(path) = written_to_path {
610 err.note(&format!("the full type name has been written to '{}'", path.display()));
615 recursion_depths.insert(def_id, recursion_depth + 1);
617 (def_id, recursion_depth)
620 fn check_type_length_limit<'tcx>(tcx: TyCtxt<'tcx>, instance: Instance<'tcx>) {
621 let type_length = instance
624 .flat_map(|arg| arg.walk())
625 .filter(|arg| match arg.unpack() {
626 GenericArgKind::Type(_) | GenericArgKind::Const(_) => true,
627 GenericArgKind::Lifetime(_) => false,
630 debug!(" => type length={}", type_length);
632 // Rust code can easily create exponentially-long types using only a
633 // polynomial recursion depth. Even with the default recursion
634 // depth, you can easily get cases that take >2^60 steps to run,
635 // which means that rustc basically hangs.
637 // Bail out in these cases to avoid that bad user experience.
638 if !tcx.type_length_limit().value_within_limit(type_length) {
639 let (shrunk, written_to_path) = shrunk_instance_name(tcx, &instance, 32, 32);
640 let msg = format!("reached the type-length limit while instantiating `{}`", shrunk);
641 let mut diag = tcx.sess.struct_span_fatal(tcx.def_span(instance.def_id()), &msg);
642 if let Some(path) = written_to_path {
643 diag.note(&format!("the full type name has been written to '{}'", path.display()));
646 "consider adding a `#![type_length_limit=\"{}\"]` attribute to your crate",
653 struct MirNeighborCollector<'a, 'tcx> {
655 body: &'a mir::Body<'tcx>,
656 output: &'a mut MonoItems<'tcx>,
657 instance: Instance<'tcx>,
660 impl<'a, 'tcx> MirNeighborCollector<'a, 'tcx> {
661 pub fn monomorphize<T>(&self, value: T) -> T
663 T: TypeFoldable<'tcx>,
665 debug!("monomorphize: self.instance={:?}", self.instance);
666 self.instance.subst_mir_and_normalize_erasing_regions(
668 ty::ParamEnv::reveal_all(),
674 impl<'a, 'tcx> MirVisitor<'tcx> for MirNeighborCollector<'a, 'tcx> {
675 fn visit_rvalue(&mut self, rvalue: &mir::Rvalue<'tcx>, location: Location) {
676 debug!("visiting rvalue {:?}", *rvalue);
678 let span = self.body.source_info(location).span;
681 // When doing an cast from a regular pointer to a fat pointer, we
682 // have to instantiate all methods of the trait being cast to, so we
683 // can build the appropriate vtable.
685 mir::CastKind::Pointer(PointerCast::Unsize),
689 let target_ty = self.monomorphize(target_ty);
690 let source_ty = operand.ty(self.body, self.tcx);
691 let source_ty = self.monomorphize(source_ty);
692 let (source_ty, target_ty) =
693 find_vtable_types_for_unsizing(self.tcx, source_ty, target_ty);
694 // This could also be a different Unsize instruction, like
695 // from a fixed sized array to a slice. But we are only
696 // interested in things that produce a vtable.
697 if target_ty.is_trait() && !source_ty.is_trait() {
698 create_mono_items_for_vtable_methods(
708 mir::CastKind::Pointer(PointerCast::ReifyFnPointer),
712 let fn_ty = operand.ty(self.body, self.tcx);
713 let fn_ty = self.monomorphize(fn_ty);
714 visit_fn_use(self.tcx, fn_ty, false, span, &mut self.output);
717 mir::CastKind::Pointer(PointerCast::ClosureFnPointer(_)),
721 let source_ty = operand.ty(self.body, self.tcx);
722 let source_ty = self.monomorphize(source_ty);
723 match *source_ty.kind() {
724 ty::Closure(def_id, substs) => {
725 let instance = Instance::resolve_closure(
729 ty::ClosureKind::FnOnce,
731 if should_codegen_locally(self.tcx, &instance) {
732 self.output.push(create_fn_mono_item(self.tcx, instance, span));
738 mir::Rvalue::ThreadLocalRef(def_id) => {
739 assert!(self.tcx.is_thread_local_static(def_id));
740 let instance = Instance::mono(self.tcx, def_id);
741 if should_codegen_locally(self.tcx, &instance) {
742 trace!("collecting thread-local static {:?}", def_id);
743 self.output.push(respan(span, MonoItem::Static(def_id)));
746 _ => { /* not interesting */ }
749 self.super_rvalue(rvalue, location);
752 /// This does not walk the constant, as it has been handled entirely here and trying
753 /// to walk it would attempt to evaluate the `ty::Const` inside, which doesn't necessarily
754 /// work, as some constants cannot be represented in the type system.
755 fn visit_constant(&mut self, constant: &mir::Constant<'tcx>, location: Location) {
756 let literal = self.monomorphize(constant.literal);
757 let val = match literal {
758 mir::ConstantKind::Val(val, _) => val,
759 mir::ConstantKind::Ty(ct) => match ct.val() {
760 ty::ConstKind::Value(val) => val,
761 ty::ConstKind::Unevaluated(ct) => {
762 let param_env = ty::ParamEnv::reveal_all();
763 match self.tcx.const_eval_resolve(param_env, ct, None) {
764 // The `monomorphize` call should have evaluated that constant already.
766 Err(ErrorHandled::Reported(_) | ErrorHandled::Linted) => return,
767 Err(ErrorHandled::TooGeneric) => span_bug!(
768 self.body.source_info(location).span,
769 "collection encountered polymorphic constant: {:?}",
777 collect_const_value(self.tcx, val, self.output);
778 self.visit_ty(literal.ty(), TyContext::Location(location));
781 fn visit_const(&mut self, constant: ty::Const<'tcx>, location: Location) {
782 debug!("visiting const {:?} @ {:?}", constant, location);
784 let substituted_constant = self.monomorphize(constant);
785 let param_env = ty::ParamEnv::reveal_all();
787 match substituted_constant.val() {
788 ty::ConstKind::Value(val) => collect_const_value(self.tcx, val, self.output),
789 ty::ConstKind::Unevaluated(unevaluated) => {
790 match self.tcx.const_eval_resolve(param_env, unevaluated, None) {
791 // The `monomorphize` call should have evaluated that constant already.
792 Ok(val) => span_bug!(
793 self.body.source_info(location).span,
794 "collection encountered the unevaluated constant {} which evaluated to {:?}",
795 substituted_constant,
798 Err(ErrorHandled::Reported(_) | ErrorHandled::Linted) => {}
799 Err(ErrorHandled::TooGeneric) => span_bug!(
800 self.body.source_info(location).span,
801 "collection encountered polymorphic constant: {}",
809 self.super_const(constant);
812 fn visit_terminator(&mut self, terminator: &mir::Terminator<'tcx>, location: Location) {
813 debug!("visiting terminator {:?} @ {:?}", terminator, location);
814 let source = self.body.source_info(location).span;
817 match terminator.kind {
818 mir::TerminatorKind::Call { ref func, .. } => {
819 let callee_ty = func.ty(self.body, tcx);
820 let callee_ty = self.monomorphize(callee_ty);
821 visit_fn_use(self.tcx, callee_ty, true, source, &mut self.output);
823 mir::TerminatorKind::Drop { ref place, .. }
824 | mir::TerminatorKind::DropAndReplace { ref place, .. } => {
825 let ty = place.ty(self.body, self.tcx).ty;
826 let ty = self.monomorphize(ty);
827 visit_drop_use(self.tcx, ty, true, source, self.output);
829 mir::TerminatorKind::InlineAsm { ref operands, .. } => {
832 mir::InlineAsmOperand::SymFn { ref value } => {
833 let fn_ty = self.monomorphize(value.literal.ty());
834 visit_fn_use(self.tcx, fn_ty, false, source, &mut self.output);
836 mir::InlineAsmOperand::SymStatic { def_id } => {
837 let instance = Instance::mono(self.tcx, def_id);
838 if should_codegen_locally(self.tcx, &instance) {
839 trace!("collecting asm sym static {:?}", def_id);
840 self.output.push(respan(source, MonoItem::Static(def_id)));
847 mir::TerminatorKind::Assert { ref msg, .. } => {
848 let lang_item = match msg {
849 mir::AssertKind::BoundsCheck { .. } => LangItem::PanicBoundsCheck,
850 _ => LangItem::Panic,
852 let instance = Instance::mono(tcx, tcx.require_lang_item(lang_item, Some(source)));
853 if should_codegen_locally(tcx, &instance) {
854 self.output.push(create_fn_mono_item(tcx, instance, source));
857 mir::TerminatorKind::Abort { .. } => {
858 let instance = Instance::mono(
860 tcx.require_lang_item(LangItem::PanicNoUnwind, Some(source)),
862 if should_codegen_locally(tcx, &instance) {
863 self.output.push(create_fn_mono_item(tcx, instance, source));
866 mir::TerminatorKind::Goto { .. }
867 | mir::TerminatorKind::SwitchInt { .. }
868 | mir::TerminatorKind::Resume
869 | mir::TerminatorKind::Return
870 | mir::TerminatorKind::Unreachable => {}
871 mir::TerminatorKind::GeneratorDrop
872 | mir::TerminatorKind::Yield { .. }
873 | mir::TerminatorKind::FalseEdge { .. }
874 | mir::TerminatorKind::FalseUnwind { .. } => bug!(),
877 self.super_terminator(terminator, location);
880 fn visit_operand(&mut self, operand: &mir::Operand<'tcx>, location: Location) {
881 self.super_operand(operand, location);
882 let limit = self.tcx.move_size_limit().0;
886 let limit = Size::from_bytes(limit);
887 let ty = operand.ty(self.body, self.tcx);
888 let ty = self.monomorphize(ty);
889 let layout = self.tcx.layout_of(ty::ParamEnv::reveal_all().and(ty));
890 if let Ok(layout) = layout {
891 if layout.size > limit {
893 let source_info = self.body.source_info(location);
894 debug!(?source_info);
895 let lint_root = source_info.scope.lint_root(&self.body.source_scopes);
897 let Some(lint_root) = lint_root else {
898 // This happens when the issue is in a function from a foreign crate that
899 // we monomorphized in the current crate. We can't get a `HirId` for things
901 // FIXME: Find out where to report the lint on. Maybe simply crate-level lint root
902 // but correct span? This would make the lint at least accept crate-level lint attributes.
905 self.tcx.struct_span_lint_hir(
910 let mut err = lint.build(&format!("moving {} bytes", layout.size.bytes()));
911 err.span_label(source_info.span, "value moved from here");
912 err.note(&format!(r#"The current maximum size is {}, but it can be customized with the move_size_limit attribute: `#![move_size_limit = "..."]`"#, limit.bytes()));
922 _place_local: &Local,
923 _context: mir::visit::PlaceContext,
929 fn visit_drop_use<'tcx>(
932 is_direct_call: bool,
934 output: &mut MonoItems<'tcx>,
936 let instance = Instance::resolve_drop_in_place(tcx, ty);
937 visit_instance_use(tcx, instance, is_direct_call, source, output);
940 fn visit_fn_use<'tcx>(
943 is_direct_call: bool,
945 output: &mut MonoItems<'tcx>,
947 if let ty::FnDef(def_id, substs) = *ty.kind() {
948 let instance = if is_direct_call {
949 ty::Instance::resolve(tcx, ty::ParamEnv::reveal_all(), def_id, substs).unwrap().unwrap()
951 ty::Instance::resolve_for_fn_ptr(tcx, ty::ParamEnv::reveal_all(), def_id, substs)
954 visit_instance_use(tcx, instance, is_direct_call, source, output);
958 fn visit_instance_use<'tcx>(
960 instance: ty::Instance<'tcx>,
961 is_direct_call: bool,
963 output: &mut MonoItems<'tcx>,
965 debug!("visit_item_use({:?}, is_direct_call={:?})", instance, is_direct_call);
966 if !should_codegen_locally(tcx, &instance) {
971 ty::InstanceDef::Virtual(..) | ty::InstanceDef::Intrinsic(_) => {
973 bug!("{:?} being reified", instance);
976 ty::InstanceDef::DropGlue(_, None) => {
977 // Don't need to emit noop drop glue if we are calling directly.
979 output.push(create_fn_mono_item(tcx, instance, source));
982 ty::InstanceDef::DropGlue(_, Some(_))
983 | ty::InstanceDef::VtableShim(..)
984 | ty::InstanceDef::ReifyShim(..)
985 | ty::InstanceDef::ClosureOnceShim { .. }
986 | ty::InstanceDef::Item(..)
987 | ty::InstanceDef::FnPtrShim(..)
988 | ty::InstanceDef::CloneShim(..) => {
989 output.push(create_fn_mono_item(tcx, instance, source));
994 /// Returns `true` if we should codegen an instance in the local crate, or returns `false` if we
995 /// can just link to the upstream crate and therefore don't need a mono item.
996 fn should_codegen_locally<'tcx>(tcx: TyCtxt<'tcx>, instance: &Instance<'tcx>) -> bool {
997 let Some(def_id) = instance.def.def_id_if_not_guaranteed_local_codegen() else {
1001 if tcx.is_foreign_item(def_id) {
1002 // Foreign items are always linked against, there's no way of instantiating them.
1006 if def_id.is_local() {
1007 // Local items cannot be referred to locally without monomorphizing them locally.
1011 if tcx.is_reachable_non_generic(def_id)
1012 || instance.polymorphize(tcx).upstream_monomorphization(tcx).is_some()
1014 // We can link to the item in question, no instance needed in this crate.
1018 if !tcx.is_mir_available(def_id) {
1019 bug!("no MIR available for {:?}", def_id);
1025 /// For a given pair of source and target type that occur in an unsizing coercion,
1026 /// this function finds the pair of types that determines the vtable linking
1029 /// For example, the source type might be `&SomeStruct` and the target type
1030 /// might be `&SomeTrait` in a cast like:
1032 /// let src: &SomeStruct = ...;
1033 /// let target = src as &SomeTrait;
1035 /// Then the output of this function would be (SomeStruct, SomeTrait) since for
1036 /// constructing the `target` fat-pointer we need the vtable for that pair.
1038 /// Things can get more complicated though because there's also the case where
1039 /// the unsized type occurs as a field:
1042 /// struct ComplexStruct<T: ?Sized> {
1049 /// In this case, if `T` is sized, `&ComplexStruct<T>` is a thin pointer. If `T`
1050 /// is unsized, `&SomeStruct` is a fat pointer, and the vtable it points to is
1051 /// for the pair of `T` (which is a trait) and the concrete type that `T` was
1052 /// originally coerced from:
1054 /// let src: &ComplexStruct<SomeStruct> = ...;
1055 /// let target = src as &ComplexStruct<SomeTrait>;
1057 /// Again, we want this `find_vtable_types_for_unsizing()` to provide the pair
1058 /// `(SomeStruct, SomeTrait)`.
1060 /// Finally, there is also the case of custom unsizing coercions, e.g., for
1061 /// smart pointers such as `Rc` and `Arc`.
1062 fn find_vtable_types_for_unsizing<'tcx>(
1064 source_ty: Ty<'tcx>,
1065 target_ty: Ty<'tcx>,
1066 ) -> (Ty<'tcx>, Ty<'tcx>) {
1067 let ptr_vtable = |inner_source: Ty<'tcx>, inner_target: Ty<'tcx>| {
1068 let param_env = ty::ParamEnv::reveal_all();
1069 let type_has_metadata = |ty: Ty<'tcx>| -> bool {
1070 if ty.is_sized(tcx.at(DUMMY_SP), param_env) {
1073 let tail = tcx.struct_tail_erasing_lifetimes(ty, param_env);
1075 ty::Foreign(..) => false,
1076 ty::Str | ty::Slice(..) | ty::Dynamic(..) => true,
1077 _ => bug!("unexpected unsized tail: {:?}", tail),
1080 if type_has_metadata(inner_source) {
1081 (inner_source, inner_target)
1083 tcx.struct_lockstep_tails_erasing_lifetimes(inner_source, inner_target, param_env)
1087 match (&source_ty.kind(), &target_ty.kind()) {
1088 (&ty::Ref(_, a, _), &ty::Ref(_, b, _) | &ty::RawPtr(ty::TypeAndMut { ty: b, .. }))
1089 | (&ty::RawPtr(ty::TypeAndMut { ty: a, .. }), &ty::RawPtr(ty::TypeAndMut { ty: b, .. })) => {
1092 (&ty::Adt(def_a, _), &ty::Adt(def_b, _)) if def_a.is_box() && def_b.is_box() => {
1093 ptr_vtable(source_ty.boxed_ty(), target_ty.boxed_ty())
1096 (&ty::Adt(source_adt_def, source_substs), &ty::Adt(target_adt_def, target_substs)) => {
1097 assert_eq!(source_adt_def, target_adt_def);
1099 let CustomCoerceUnsized::Struct(coerce_index) =
1100 crate::custom_coerce_unsize_info(tcx, source_ty, target_ty);
1102 let source_fields = &source_adt_def.non_enum_variant().fields;
1103 let target_fields = &target_adt_def.non_enum_variant().fields;
1106 coerce_index < source_fields.len() && source_fields.len() == target_fields.len()
1109 find_vtable_types_for_unsizing(
1111 source_fields[coerce_index].ty(tcx, source_substs),
1112 target_fields[coerce_index].ty(tcx, target_substs),
1116 "find_vtable_types_for_unsizing: invalid coercion {:?} -> {:?}",
1123 fn create_fn_mono_item<'tcx>(
1125 instance: Instance<'tcx>,
1127 ) -> Spanned<MonoItem<'tcx>> {
1128 debug!("create_fn_mono_item(instance={})", instance);
1130 let def_id = instance.def_id();
1131 if tcx.sess.opts.debugging_opts.profile_closures && def_id.is_local() && tcx.is_closure(def_id)
1133 crate::util::dump_closure_profile(tcx, instance);
1136 respan(source, MonoItem::Fn(instance.polymorphize(tcx)))
1139 /// Creates a `MonoItem` for each method that is referenced by the vtable for
1140 /// the given trait/impl pair.
1141 fn create_mono_items_for_vtable_methods<'tcx>(
1146 output: &mut MonoItems<'tcx>,
1148 assert!(!trait_ty.has_escaping_bound_vars() && !impl_ty.has_escaping_bound_vars());
1150 if let ty::Dynamic(ref trait_ty, ..) = trait_ty.kind() {
1151 if let Some(principal) = trait_ty.principal() {
1152 let poly_trait_ref = principal.with_self_ty(tcx, impl_ty);
1153 assert!(!poly_trait_ref.has_escaping_bound_vars());
1155 // Walk all methods of the trait, including those of its supertraits
1156 let entries = tcx.vtable_entries(poly_trait_ref);
1157 let methods = entries
1159 .filter_map(|entry| match entry {
1160 VtblEntry::MetadataDropInPlace
1161 | VtblEntry::MetadataSize
1162 | VtblEntry::MetadataAlign
1163 | VtblEntry::Vacant => None,
1164 VtblEntry::TraitVPtr(_) => {
1165 // all super trait items already covered, so skip them.
1168 VtblEntry::Method(instance) => {
1169 Some(*instance).filter(|instance| should_codegen_locally(tcx, instance))
1172 .map(|item| create_fn_mono_item(tcx, item, source));
1173 output.extend(methods);
1176 // Also add the destructor.
1177 visit_drop_use(tcx, impl_ty, false, source, output);
1181 //=-----------------------------------------------------------------------------
1183 //=-----------------------------------------------------------------------------
1185 struct RootCollector<'a, 'tcx> {
1187 mode: MonoItemCollectionMode,
1188 output: &'a mut MonoItems<'tcx>,
1189 entry_fn: Option<(DefId, EntryFnType)>,
1192 impl<'v> RootCollector<'_, 'v> {
1193 fn process_item(&mut self, id: hir::ItemId) {
1194 match self.tcx.def_kind(id.def_id) {
1195 DefKind::Enum | DefKind::Struct | DefKind::Union => {
1196 let item = self.tcx.hir().item(id);
1198 hir::ItemKind::Enum(_, ref generics)
1199 | hir::ItemKind::Struct(_, ref generics)
1200 | hir::ItemKind::Union(_, ref generics) => {
1201 if generics.params.is_empty() {
1202 if self.mode == MonoItemCollectionMode::Eager {
1204 "RootCollector: ADT drop-glue for {}",
1205 self.tcx.def_path_str(item.def_id.to_def_id())
1209 Instance::new(item.def_id.to_def_id(), InternalSubsts::empty())
1210 .ty(self.tcx, ty::ParamEnv::reveal_all());
1211 visit_drop_use(self.tcx, ty, true, DUMMY_SP, self.output);
1218 DefKind::GlobalAsm => {
1220 "RootCollector: ItemKind::GlobalAsm({})",
1221 self.tcx.def_path_str(id.def_id.to_def_id())
1223 self.output.push(dummy_spanned(MonoItem::GlobalAsm(id)));
1225 DefKind::Static(..) => {
1227 "RootCollector: ItemKind::Static({})",
1228 self.tcx.def_path_str(id.def_id.to_def_id())
1230 self.output.push(dummy_spanned(MonoItem::Static(id.def_id.to_def_id())));
1233 // const items only generate mono items if they are
1234 // actually used somewhere. Just declaring them is insufficient.
1236 // but even just declaring them must collect the items they refer to
1237 if let Ok(val) = self.tcx.const_eval_poly(id.def_id.to_def_id()) {
1238 collect_const_value(self.tcx, val, &mut self.output);
1242 if self.mode == MonoItemCollectionMode::Eager {
1243 let item = self.tcx.hir().item(id);
1244 create_mono_items_for_default_impls(self.tcx, item, self.output);
1248 self.push_if_root(id.def_id);
1254 fn process_impl_item(&mut self, id: hir::ImplItemId) {
1255 if matches!(self.tcx.def_kind(id.def_id), DefKind::AssocFn) {
1256 self.push_if_root(id.def_id);
1260 fn is_root(&self, def_id: LocalDefId) -> bool {
1261 !item_requires_monomorphization(self.tcx, def_id)
1262 && match self.mode {
1263 MonoItemCollectionMode::Eager => true,
1264 MonoItemCollectionMode::Lazy => {
1265 self.entry_fn.and_then(|(id, _)| id.as_local()) == Some(def_id)
1266 || self.tcx.is_reachable_non_generic(def_id)
1269 .codegen_fn_attrs(def_id)
1271 .contains(CodegenFnAttrFlags::RUSTC_STD_INTERNAL_SYMBOL)
1276 /// If `def_id` represents a root, pushes it onto the list of
1277 /// outputs. (Note that all roots must be monomorphic.)
1278 fn push_if_root(&mut self, def_id: LocalDefId) {
1279 if self.is_root(def_id) {
1280 debug!("RootCollector::push_if_root: found root def_id={:?}", def_id);
1282 let instance = Instance::mono(self.tcx, def_id.to_def_id());
1283 self.output.push(create_fn_mono_item(self.tcx, instance, DUMMY_SP));
1287 /// As a special case, when/if we encounter the
1288 /// `main()` function, we also have to generate a
1289 /// monomorphized copy of the start lang item based on
1290 /// the return type of `main`. This is not needed when
1291 /// the user writes their own `start` manually.
1292 fn push_extra_entry_roots(&mut self) {
1293 let Some((main_def_id, EntryFnType::Main)) = self.entry_fn else {
1297 let start_def_id = match self.tcx.lang_items().require(LangItem::Start) {
1299 Err(err) => self.tcx.sess.fatal(&err),
1301 let main_ret_ty = self.tcx.fn_sig(main_def_id).output();
1303 // Given that `main()` has no arguments,
1304 // then its return type cannot have
1305 // late-bound regions, since late-bound
1306 // regions must appear in the argument
1308 let main_ret_ty = self.tcx.normalize_erasing_regions(
1309 ty::ParamEnv::reveal_all(),
1310 main_ret_ty.no_bound_vars().unwrap(),
1313 let start_instance = Instance::resolve(
1315 ty::ParamEnv::reveal_all(),
1317 self.tcx.intern_substs(&[main_ret_ty.into()]),
1322 self.output.push(create_fn_mono_item(self.tcx, start_instance, DUMMY_SP));
1326 fn item_requires_monomorphization(tcx: TyCtxt<'_>, def_id: LocalDefId) -> bool {
1327 let generics = tcx.generics_of(def_id);
1328 generics.requires_monomorphization(tcx)
1331 fn create_mono_items_for_default_impls<'tcx>(
1333 item: &'tcx hir::Item<'tcx>,
1334 output: &mut MonoItems<'tcx>,
1337 hir::ItemKind::Impl(ref impl_) => {
1338 for param in impl_.generics.params {
1340 hir::GenericParamKind::Lifetime { .. } => {}
1341 hir::GenericParamKind::Type { .. } | hir::GenericParamKind::Const { .. } => {
1348 "create_mono_items_for_default_impls(item={})",
1349 tcx.def_path_str(item.def_id.to_def_id())
1352 if let Some(trait_ref) = tcx.impl_trait_ref(item.def_id) {
1353 let param_env = ty::ParamEnv::reveal_all();
1354 let trait_ref = tcx.normalize_erasing_regions(param_env, trait_ref);
1355 let overridden_methods = tcx.impl_item_implementor_ids(item.def_id);
1356 for method in tcx.provided_trait_methods(trait_ref.def_id) {
1357 if overridden_methods.contains_key(&method.def_id) {
1361 if tcx.generics_of(method.def_id).own_requires_monomorphization() {
1366 InternalSubsts::for_item(tcx, method.def_id, |param, _| match param.kind {
1367 GenericParamDefKind::Lifetime => tcx.lifetimes.re_erased.into(),
1368 GenericParamDefKind::Type { .. }
1369 | GenericParamDefKind::Const { .. } => {
1370 trait_ref.substs[param.index as usize]
1373 let instance = ty::Instance::resolve(tcx, param_env, method.def_id, substs)
1377 let mono_item = create_fn_mono_item(tcx, instance, DUMMY_SP);
1378 if mono_item.node.is_instantiable(tcx) && should_codegen_locally(tcx, &instance)
1380 output.push(mono_item);
1389 /// Scans the miri alloc in order to find function calls, closures, and drop-glue.
1390 fn collect_miri<'tcx>(tcx: TyCtxt<'tcx>, alloc_id: AllocId, output: &mut MonoItems<'tcx>) {
1391 match tcx.global_alloc(alloc_id) {
1392 GlobalAlloc::Static(def_id) => {
1393 assert!(!tcx.is_thread_local_static(def_id));
1394 let instance = Instance::mono(tcx, def_id);
1395 if should_codegen_locally(tcx, &instance) {
1396 trace!("collecting static {:?}", def_id);
1397 output.push(dummy_spanned(MonoItem::Static(def_id)));
1400 GlobalAlloc::Memory(alloc) => {
1401 trace!("collecting {:?} with {:#?}", alloc_id, alloc);
1402 for &inner in alloc.inner().relocations().values() {
1403 rustc_data_structures::stack::ensure_sufficient_stack(|| {
1404 collect_miri(tcx, inner, output);
1408 GlobalAlloc::Function(fn_instance) => {
1409 if should_codegen_locally(tcx, &fn_instance) {
1410 trace!("collecting {:?} with {:#?}", alloc_id, fn_instance);
1411 output.push(create_fn_mono_item(tcx, fn_instance, DUMMY_SP));
1417 /// Scans the MIR in order to find function calls, closures, and drop-glue.
1418 fn collect_neighbours<'tcx>(
1420 instance: Instance<'tcx>,
1421 output: &mut MonoItems<'tcx>,
1423 debug!("collect_neighbours: {:?}", instance.def_id());
1424 let body = tcx.instance_mir(instance.def);
1426 MirNeighborCollector { tcx, body: &body, output, instance }.visit_body(&body);
1429 fn collect_const_value<'tcx>(
1431 value: ConstValue<'tcx>,
1432 output: &mut MonoItems<'tcx>,
1435 ConstValue::Scalar(Scalar::Ptr(ptr, _size)) => collect_miri(tcx, ptr.provenance, output),
1436 ConstValue::Slice { data: alloc, start: _, end: _ } | ConstValue::ByRef { alloc, .. } => {
1437 for &id in alloc.inner().relocations().values() {
1438 collect_miri(tcx, id, output);