Merge pull request #3464 from swiftwasm/main

[pull] swiftwasm from main

Author: kateinoigakukun

docs/SIL.rst

@@ -3493,11 +3493,10 @@ debug_value
 
   debug_value %1 : $Int
 
-This indicates that the value of a declaration with loadable type has changed
-value to the specified operand.  The declaration in question is identified by
-the SILLocation attached to the debug_value instruction.
-
-The operand must have loadable type.
+This indicates that the value of a declaration has changed value to the
+specified operand.  The declaration in question is identified by either the
+SILLocation attached to the debug_value instruction or the SILLocation specified
+in the advanced debug variable attributes.
 
 ::
 
@@ -3507,47 +3506,74 @@ The operand must have loadable type.
    debug-var-attr ::= 'argno' integer-literal
    debug-var-attr ::= 'implicit'
 
-::
-
-  advanced-debug-var-attr ::= '(' 'name' string-literal (',' sil-instruction-source-info)? ')'
-  advanced-debug-var-attr ::= 'type' sil-type
-
-::
-
-  debug-info-expr   ::= di-expr-operand (':' di-expr-operand)*
-  di-expr-operand   ::= di-expr-operator (':' sil-operand)*
-  di-expr-operator  ::= 'op_fragment'
-
 There are a number of attributes that provide details about the source
 variable that is being described, including the name of the
 variable. For function and closure arguments ``argno`` is the number
 of the function argument starting with 1. A compiler-generated source
 variable will be marked ``implicit`` and optimizers are free to remove
-it even in -Onone. The advanced debug variable attributes represent source
-locations and type of the source variable when it was originally declared.
-It is useful when we're indirectly associating the SSA value with the
-source variable (via di-expression, for example) in which case SSA value's
-type is different from that of source variable.
+it even in -Onone.
 
 If the '[poison]' flag is set, then all references within this debug
 value will be overwritten with a sentinel at this point in the
 program. This is used in debug builds when shortening non-trivial
 value lifetimes to ensure the debugger cannot inspect invalid
-memory. `debug_value` instructions with the poison flag are not
-generated until OSSA islowered. They are not expected to be serialized
+memory. ``debug_value`` instructions with the poison flag are not
+generated until OSSA is lowered. They are not expected to be serialized
 within the module, and the pipeline is not expected to do any
 significant code motion after lowering.
 
-Debug info expression (di-expression) is a powerful method to connect SSA
-value with the source variable in an indirect fashion. For example,
-we can use the ``op_fragment`` operator to specify that the SSA value
-is originated from a struct field inside the source variable (which has
-an aggregate data type). Di-expression in SIL works similarly to LLVM's
-``!DIExpression`` metadata. Where both of them adopt a stack based
-execution model to evaluate the expression. The biggest difference between
-them is that LLVM always represent ``!DIExpression`` elements as 64-bit
-integers, while SIL's di-expression can have elements with various types,
-like AST nodes or strings. Here is an example::
+::
+
+  advanced-debug-var-attr ::= '(' 'name' string-literal (',' sil-instruction-source-info)? ')'
+  advanced-debug-var-attr ::= 'type' sil-type
+
+Advanced debug variable attributes represent source locations and the type of
+the source variable when it was originally declared. It is useful when
+we're indirectly associating the SSA value with the source variable
+(via SIL DIExpression, for example) in which case SSA value's type is different
+from that of source variable.
+
+::
+
+  debug-info-expr   ::= di-expr-operand (':' di-expr-operand)*
+  di-expr-operand   ::= di-expr-operator (':' sil-operand)*
+  di-expr-operator  ::= 'op_fragment'
+  di-expr-operator  ::= 'op_deref'
+
+SIL debug info expression (SIL DIExpression) is a powerful method to connect SSA
+value with the source variable in an indirect fashion. Di-expression in SIL
+uses a stack based execution model to evaluate the expression and apply on
+the associated (SIL) SSA value before connecting it with the debug variable.
+For instance, given the following SIL code::
+
+  debug_value %a : $*Int, name "x", expr op_deref
+
+It means: "You can get the value of source variable 'x' by *dereferencing*
+SSA value ``%a``". The ``op_deref`` is a SIL DIExpression operator that represents
+"dereference". If there are multiple SIL DIExpression operators (or arguments), they
+are evaluated from left to right::
+
+  debug_value %b : $**Int, name "y", expr op_deref:op_deref
+
+In the snippet above, two ``op_deref`` operators will be applied on SSA value
+``%b`` sequentially.
+
+Note that normally when the SSA value has an address type, there will be a ``op_deref``
+in the SIL DIExpression. Because there is no pointer in Swift so you always need to
+dereference an address-type SSA value to get the value of a source variable.
+However, if the SSA value is a ``alloc_stack``, the ``debug_value`` is used to indicate
+the *declaration* of a source variable. Or, you can say, used to specify the location
+(memory address) of the source variable. Therefore, we don't need to add a ``op_deref``
+in this case::
+
+  %a = alloc_stack $Int, ...
+  debug_value %a : $*Int, name "my_var"
+
+
+The ``op_fragment`` operator is used to specify the SSA value of a specific
+field in an aggregate-type source variable. This SIL DIExpression operator takes
+a field declaration -- which references the desired sub-field in source variable
+-- as its argument. Here is an example::
 
   struct MyStruct {
     var x: Int
@@ -3556,23 +3582,13 @@ like AST nodes or strings. Here is an example::
   ...
   debug_value %1 : $Int, var, (name "the_struct", loc "file.swift":8:7), type $MyStruct, expr op_fragment:#MyStruct.y, loc "file.swift":9:4
 
-In the snippet above, source variable "the_struct" has an aggregate type ``$MyStruct`` and we use di-expression with ``op_fragment`` operator to associate ``%1`` to the ``y`` member variable inside "the_struct". Note that the extra source location directive follows rigt after ``name "the_struct"`` indicate that "the_struct" was originally declared in line 8, but not until line 9, the current ``debug_value`` instruction's source location, does member ``y`` got updated with SSA value ``%1``.
-
-debug_value_addr
-````````````````
-
-::
-
-  sil-instruction ::= debug_value_addr sil-operand (',' debug-var-attr)* advanced-debug-var-attr* (',' 'expr' debug-info-expr)?
-
-  debug_value_addr %7 : $*SomeProtocol
-
-This indicates that the value of a declaration with address-only type
-has changed value to the specified operand.  The declaration in
-question is identified by the SILLocation attached to the
-debug_value_addr instruction.
+In the snippet above, source variable "the_struct" has an aggregate type ``$MyStruct`` and we use a SIL DIExpression with ``op_fragment`` operator to associate ``%1`` to the ``y`` member variable (via the ``#MyStruct.y`` directive) inside "the_struct".
+Note that the extra source location directive follows rigt after ``name "the_struct"`` indicate that "the_struct" was originally declared in line 8, but not until line 9 -- the current ``debug_value`` instruction's source location -- does member ``y`` got updated with SSA value ``%1``.
 
-Note that this instruction can be replaced by ``debug_value`` + di-expression operator that is equivalent to LLVM's ``DW_OP_deref``.
+It is worth noting that a SIL DIExpression is similar to
+`!DIExpression <https://www.llvm.org/docs/LangRef.html#diexpression>`_ in LLVM debug
+info metadata. While LLVM represents ``!DIExpression`` are a list of 64-bit integers,
+SIL DIExpression can have elements with various types, like AST nodes or strings.
 
 Accessing Memory
 ~~~~~~~~~~~~~~~~

include/swift/ABI/Task.h

@@ -23,6 +23,7 @@
 #include "swift/ABI/Metadata.h"
 #include "swift/ABI/MetadataValues.h"
 #include "swift/Runtime/Config.h"
+#include "swift/Runtime/VoucherShims.h"
 #include "swift/Basic/STLExtras.h"
 #include "bitset"
 #include "queue" // TODO: remove and replace with our own mpsc
@@ -72,7 +73,13 @@ class alignas(2 * alignof(void*)) Job :
   // Derived classes can use this to store a Job Id.
   uint32_t Id = 0;
 
-  void *Reserved[2] = {};
+  /// The voucher associated with the job. Note: this is currently unused on
+  /// non-Darwin platforms, with stub implementations of the functions for
+  /// consistency.
+  voucher_t Voucher = nullptr;
+
+  /// Reserved for future use.
+  void *Reserved = nullptr;
 
   // We use this union to avoid having to do a second indirect branch
   // when resuming an asynchronous task, which we expect will be the
@@ -88,23 +95,32 @@ class alignas(2 * alignof(void*)) Job :
   Job(JobFlags flags, JobInvokeFunction *invoke,
       const HeapMetadata *metadata = &jobHeapMetadata)
       : HeapObject(metadata), Flags(flags), RunJob(invoke) {
+    Voucher = voucher_copy();
     assert(!isAsyncTask() && "wrong constructor for a task");
   }
 
   Job(JobFlags flags, TaskContinuationFunction *invoke,
-      const HeapMetadata *metadata = &jobHeapMetadata)
+      const HeapMetadata *metadata = &jobHeapMetadata,
+      bool captureCurrentVoucher = true)
       : HeapObject(metadata), Flags(flags), ResumeTask(invoke) {
+    if (captureCurrentVoucher)
+      Voucher = voucher_copy();
     assert(isAsyncTask() && "wrong constructor for a non-task job");
   }
 
   /// Create a job with "immortal" reference counts.
   /// Used for async let tasks.
   Job(JobFlags flags, TaskContinuationFunction *invoke,
-      const HeapMetadata *metadata, InlineRefCounts::Immortal_t immortal)
+      const HeapMetadata *metadata, InlineRefCounts::Immortal_t immortal,
+      bool captureCurrentVoucher = true)
       : HeapObject(metadata, immortal), Flags(flags), ResumeTask(invoke) {
+    if (captureCurrentVoucher)
+      Voucher = voucher_copy();
     assert(isAsyncTask() && "wrong constructor for a non-task job");
   }
 
+  ~Job() { swift_voucher_release(Voucher); }
+
   bool isAsyncTask() const {
     return Flags.isAsyncTask();
   }
@@ -229,8 +245,9 @@ class AsyncTask : public Job {
   /// Private.initialize separately.
   AsyncTask(const HeapMetadata *metadata, JobFlags flags,
             TaskContinuationFunction *run,
-            AsyncContext *initialContext)
-    : Job(flags, run, metadata),
+            AsyncContext *initialContext,
+            bool captureCurrentVoucher)
+    : Job(flags, run, metadata, captureCurrentVoucher),
       ResumeContext(initialContext) {
     assert(flags.isAsyncTask());
     Id = getNextTaskId();
@@ -243,8 +260,9 @@ class AsyncTask : public Job {
   AsyncTask(const HeapMetadata *metadata, InlineRefCounts::Immortal_t immortal,
             JobFlags flags,
             TaskContinuationFunction *run,
-            AsyncContext *initialContext)
-    : Job(flags, run, metadata, immortal),
+            AsyncContext *initialContext,
+            bool captureCurrentVoucher)
+    : Job(flags, run, metadata, immortal, captureCurrentVoucher),
       ResumeContext(initialContext) {
     assert(flags.isAsyncTask());
     Id = getNextTaskId();

include/swift/AST/DiagnosticsSema.def

@@ -3756,8 +3756,13 @@ NOTE(add_self_to_type,none,
 WARNING(warn_unqualified_access,none,
         "use of %0 treated as a reference to %1 in %2 %3",
         (Identifier, DescriptiveDeclKind, DescriptiveDeclKind, DeclName))
+WARNING(self_refers_to_method,none,
+       "'self' refers to the method '%0.self', which may be unexpected",
+       (StringRef))
 NOTE(fix_unqualified_access_member,none,
      "use 'self.' to silence this warning", ())
+NOTE(fix_unqualified_access_member_named_self,none,
+     "use '%0.self' to silence this warning", (StringRef))
 NOTE(fix_unqualified_access_top_level,none,
      "use '%0' to reference the %1",
      (StringRef, DescriptiveDeclKind, Identifier))

include/swift/Runtime/VoucherShims.h

@@ -0,0 +1,97 @@
+//===--- VoucherShims.h - Shims for OS vouchers --------------------*- C++ -*-//
+//
+// This source file is part of the Swift.org open source project
+//
+// Copyright (c) 2014 - 2021 Apple Inc. and the Swift project authors
+// Licensed under Apache License v2.0 with Runtime Library Exception
+//
+// See https://swift.org/LICENSE.txt for license information
+// See https://swift.org/CONTRIBUTORS.txt for the list of Swift project authors
+//
+//===----------------------------------------------------------------------===//
+//
+// Shims for interfacing with OS voucher calls.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef SWIFT_CONCURRENCY_VOUCHERSHIMS_H
+#define SWIFT_CONCURRENCY_VOUCHERSHIMS_H
+
+#include "Config.h"
+
+// swift-corelibs-libdispatch has os/voucher_private.h but it doesn't work for
+// us yet, so only look for it on Apple platforms.
+#if __APPLE__ && __has_include(<os/voucher_private.h>)
+#define SWIFT_HAS_VOUCHER_HEADER 1
+#include <os/voucher_private.h>
+#endif
+
+// A "dead" voucher pointer, indicating that a voucher has been removed from
+// a Job, distinct from a NULL voucher that could just mean no voucher was
+// present. This allows us to catch problems like adopting a voucher from the
+// same Job twice without restoring it.
+#define SWIFT_DEAD_VOUCHER ((voucher_t)-1)
+
+// The OS has voucher support if it has the header or if it has ObjC interop.
+#if SWIFT_HAS_VOUCHER_HEADER || SWIFT_OBJC_INTEROP
+#define SWIFT_HAS_VOUCHERS 1
+#endif
+
+#if SWIFT_HAS_VOUCHERS
+
+#if SWIFT_HAS_VOUCHER_HEADER
+
+static inline bool swift_voucher_needs_adopt(voucher_t _Nullable voucher) {
+  if (__builtin_available(macOS 12.0, iOS 15.0, tvOS 15.0, watchOS 8.0, *)) {
+    return voucher_needs_adopt(voucher);
+  }
+  return true;
+}
+
+#else
+
+// If the header isn't available, declare the necessary calls here.
+
+#include <os/object.h>
+
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wgnu-zero-variadic-macro-arguments"
+OS_OBJECT_DECL_CLASS(voucher);
+#pragma clang diagnostic pop
+
+extern "C" voucher_t _Nullable voucher_copy(void);
+
+// Consumes argument, returns retained value.
+extern "C" voucher_t _Nullable voucher_adopt(voucher_t _Nullable voucher);
+
+static inline bool swift_voucher_needs_adopt(voucher_t _Nullable voucher) {
+  return true;
+}
+
+#endif // __has_include(<os/voucher_private.h>)
+
+static inline void swift_voucher_release(voucher_t _Nullable voucher) {
+  // This NULL check isn't necessary, but NULL vouchers will be common, so
+  // optimize for that.
+  if (!voucher)
+    return;
+  if (voucher == SWIFT_DEAD_VOUCHER)
+    return;
+  os_release(voucher);
+}
+
+#else  // __APPLE__
+
+// Declare some do-nothing stubs for OSes without voucher support.
+typedef void *voucher_t;
+static inline voucher_t _Nullable voucher_copy(void) { return nullptr; }
+static inline voucher_t _Nullable voucher_adopt(voucher_t _Nullable voucher) {
+  return nullptr;
+}
+static inline bool swift_voucher_needs_adopt(voucher_t _Nullable voucher) {
+  return true;
+}
+static inline void swift_voucher_release(voucher_t _Nullable voucher) {}
+#endif // __APPLE__
+
+#endif

lib/APIDigester/ModuleAnalyzerNodes.cpp

@@ -1797,12 +1797,17 @@ SwiftDeclCollector::addMembersToRoot(SDKNode *Root, IterableDeclContext *Context
       Root->addChild(constructSubscriptDeclNode(SD));
     } else if (isa<PatternBindingDecl>(Member)) {
       // All containing variables should have been handled.
+    } else if (isa<EnumCaseDecl>(Member)) {
+      // All containing variables should have been handled.
+    } else if (isa<IfConfigDecl>(Member)) {
+      // All containing members should have been handled.
     } else if (isa<DestructorDecl>(Member)) {
       // deinit has no impact.
     } else if (isa<MissingMemberDecl>(Member)) {
       // avoid adding MissingMemberDecl
     } else {
-      llvm_unreachable("unhandled member decl kind.");
+      llvm::errs() << "Unhandled decl:\n";
+      Member->dump(llvm::errs());
     }
   }
 }

lib/AST/RequirementMachine/RewriteSystem.cpp

@@ -97,7 +97,11 @@ bool RewriteSystem::addRule(MutableTerm lhs, MutableTerm rhs) {
   }
 
   unsigned i = Rules.size();
-  Rules.emplace_back(Term::get(lhs, Context), Term::get(rhs, Context));
+
+  auto uniquedLHS = Term::get(lhs, Context);
+  auto uniquedRHS = Term::get(rhs, Context);
+  Rules.emplace_back(uniquedLHS, uniquedRHS);
+
   auto oldRuleID = Trie.insert(lhs.begin(), lhs.end(), i);
   if (oldRuleID) {
     llvm::errs() << "Duplicate rewrite rule!\n";
@@ -112,23 +116,7 @@ bool RewriteSystem::addRule(MutableTerm lhs, MutableTerm rhs) {
     abort();
   }
 
-  // Check if we have a rule of the form
-  //
-  //   X.[P1:T] => X.[P2:T]
-  //
-  // If so, record this rule for later. We'll try to merge the associated
-  // types in RewriteSystem::processMergedAssociatedTypes().
-  if (lhs.size() == rhs.size() &&
-      std::equal(lhs.begin(), lhs.end() - 1, rhs.begin()) &&
-      lhs.back().getKind() == Symbol::Kind::AssociatedType &&
-      rhs.back().getKind() == Symbol::Kind::AssociatedType &&
-      lhs.back().getName() == rhs.back().getName()) {
-    if (Debug.contains(DebugFlags::Merge)) {
-      llvm::dbgs() << "## Associated type merge candidate ";
-      llvm::dbgs() << lhs << " => " << rhs << "\n\n";
-    }
-    MergedAssociatedTypes.emplace_back(lhs, rhs);
-  }
+  checkMergedAssociatedType(uniquedLHS, uniquedRHS);
 
   // Tell the caller that we added a new rule.
   return true;