[llvm] r340663 - Add data structure to form equivalence classes of mangled names.

Chandler Carruth via llvm-commits llvm-commits at lists.llvm.org
Sun Aug 26 03:33:36 PDT 2018


Since there are a bunch of commits on top of this, I tried to fix the issue
with r340700.

That in turn seems to uncover a serious MSVC bug hit by this code:
http://lab.llvm.org:8011/builders/llvm-clang-x86_64-expensive-checks-win/builds/12026/steps/build-unified-tree/logs/stdio

I think I have worked around *this* with r340702.

But generally, it feels like you're trying to write a value-parameterized
test, and GoogleTest has special facilities to aid in that:
https://github.com/google/googletest/blob/master/googletest/docs/advanced.md#value-parameterized-tests

The syntax is a little clunky but I think it provides much nicer results by
allowing you to break down the test a bit more. Anyways, maybe it doesn't
work for you precise case, but something to consider.

On Sun, Aug 26, 2018 at 1:48 AM Chandler Carruth <chandlerc at gmail.com>
wrote:

> Bots have been broken for a long time now. I'm going to revert this to get
> the bots green and Richard can re-land...
>
> On Sat, Aug 25, 2018 at 4:20 PM via llvm-commits <
> llvm-commits at lists.llvm.org> wrote:
>
>> Hi Richard, your commit is causing a build failure on the PS4 Windows
>> bot. Can you please take a look?
>>
>>
>> http://lab.llvm.org:8011/builders/llvm-clang-lld-x86_64-scei-ps4-windows10pro-fast/builds/19356/steps/build/logs/stdio
>>
>> FAILED:
>> lib/Support/CMakeFiles/LLVMSupport.dir/ItaniumManglingCanonicalizer.cpp.obj
>> C:\PROGRA~2\MICROS~1.0\VC\bin\cl.exe  /nologo /TP -DGTEST_HAS_RTTI=0
>> -DUNICODE -D_CRT_NONSTDC_NO_DEPRECATE -D_CRT_NONSTDC_NO_WARNINGS
>> -D_CRT_SECURE_NO_DEPRECATE -D_CRT_SECURE_NO_WARNINGS -D_FILE_OFFSET_BITS=64
>> -D_HAS_EXCEPTIONS=0 -D_LARGEFILE_SOURCE -D_SCL_SECURE_NO_DEPRECATE
>> -D_SCL_SECURE_NO_WARNINGS -D_UNICODE -D__STDC_CONSTANT_MACROS
>> -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -Ilib\Support
>> -IC:\ps4-buildslave2\llvm-clang-lld-x86_64-scei-ps4-windows10pro-fast\llvm.src\lib\Support
>> -Iinclude
>> -IC:\ps4-buildslave2\llvm-clang-lld-x86_64-scei-ps4-windows10pro-fast\llvm.src\include
>> /DWIN32 /D_WINDOWS   /Zc:inline /Zc:strictStrings /Oi /Zc:rvalueCast /W4
>> -wd4141 -wd4146 -wd4180 -wd4244 -wd4258 -wd4267 -wd4291 -wd4345 -wd4351
>> -wd4355 -wd4456 -wd4457 -wd4458 -wd4459 -wd4503 -wd4624 -wd4722 -wd4800
>> -wd4100 -wd4127 -wd4512 -wd4505 -wd4610 -wd4510 -wd4702 -wd4245 -wd4706
>> -wd4310 -wd4701 -wd4703 -wd4389 -wd4611 -wd4805 -wd4204 -wd4577 -wd4091
>> -wd4592 -wd4319 -wd4324 -w14062 -we4238 /MD /O2 /Ob2   -UNDEBUG  /EHs-c-
>> /GR- /showIncludes
>> /Folib\Support\CMakeFiles\LLVMSupport.dir\ItaniumManglingCanonicalizer.cpp.obj
>> /Fdlib\Support\CMakeFiles\LLVMSupport.dir\LLVMSupport.pdb /FS -c
>> C:\ps4-buildslave2\llvm-clang-lld-x86_64-scei-ps4-windows10pro-fast\llvm.src\lib\Support\ItaniumManglingCanonicalizer.cpp
>> C:\ps4-buildslave2\llvm-clang-lld-x86_64-scei-ps4-windows10pro-fast\llvm.src\lib\Support\ItaniumManglingCanonicalizer.cpp(146):
>> error C2910:
>> '`anonymous-namespace'::FoldingNodeAllocator::getOrCreateNode': cannot be
>> explicitly specialized
>>
>> Douglas Yung
>>
>> > -----Original Message-----
>> > From: llvm-commits [mailto:llvm-commits-bounces at lists.llvm.org] On
>> > Behalf Of Richard Smith via llvm-commits
>> > Sent: Friday, August 24, 2018 15:32
>> > To: llvm-commits at lists.llvm.org
>> > Subject: [llvm] r340663 - Add data structure to form equivalence
>> > classes of mangled names.
>> >
>> > Author: rsmith
>> > Date: Fri Aug 24 15:31:51 2018
>> > New Revision: 340663
>> >
>> > URL: http://llvm.org/viewvc/llvm-project?rev=340663&view=rev
>> > Log:
>> > Add data structure to form equivalence classes of mangled names.
>> >
>> > Summary:
>> > Given a set of equivalent name fragments, this mechanism determines
>> > whether two
>> > mangled names are equivalent. The intent is to use this for fuzzy
>> > matching of
>> > profile data against the program after certain refactorings are
>> > performed.
>> >
>> > Reviewers: erik.pilkington, dlj
>> >
>> > Subscribers: mgorny, llvm-commits
>> >
>> > Differential Revision: https://reviews.llvm.org/D50935
>> >
>> > Added:
>> >     llvm/trunk/include/llvm/Support/ItaniumManglingCanonicalizer.h
>> >     llvm/trunk/lib/Support/ItaniumManglingCanonicalizer.cpp
>> >     llvm/trunk/unittests/Support/ItaniumManglingCanonicalizerTest.cpp
>> > Modified:
>> >     llvm/trunk/include/llvm/Demangle/ItaniumDemangle.h
>> >     llvm/trunk/lib/Support/CMakeLists.txt
>> >     llvm/trunk/unittests/Support/CMakeLists.txt
>> >
>> > Modified: llvm/trunk/include/llvm/Demangle/ItaniumDemangle.h
>> > URL: http://llvm.org/viewvc/llvm-
>> > project/llvm/trunk/include/llvm/Demangle/ItaniumDemangle.h?rev=340663&r
>> > 1=340662&r2=340663&view=diff
>> > =======================================================================
>> > =======
>> > --- llvm/trunk/include/llvm/Demangle/ItaniumDemangle.h (original)
>> > +++ llvm/trunk/include/llvm/Demangle/ItaniumDemangle.h Fri Aug 24
>> > 15:31:51 2018
>> > @@ -2160,7 +2160,7 @@ struct Db {
>> >      ASTAllocator.reset();
>> >    }
>> >
>> > -  template <class T, class... Args> T *make(Args &&... args) {
>> > +  template <class T, class... Args> Node *make(Args &&... args) {
>> >      return ASTAllocator.template
>> > makeNode<T>(std::forward<Args>(args)...);
>> >    }
>> >
>> > @@ -4948,8 +4948,11 @@ template<typename Alloc> Node *Db<Alloc>
>> >    // <template-arg> further ahead in the mangled name (currently just
>> > conversion
>> >    // operator types), then we should only look it up in the right
>> > context.
>> >    if (PermitForwardTemplateReferences) {
>> > -
>> > ForwardTemplateRefs.push_back(make<ForwardTemplateReference>(Index));
>> > -    return ForwardTemplateRefs.back();
>> > +    Node *ForwardRef = make<ForwardTemplateReference>(Index);
>> > +    assert(ForwardRef->getKind() == Node::KForwardTemplateReference);
>> > +    ForwardTemplateRefs.push_back(
>> > +        static_cast<ForwardTemplateReference *>(ForwardRef));
>> > +    return ForwardRef;
>> >    }
>> >
>> >    if (Index >= TemplateParams.size())
>> >
>> > Added: llvm/trunk/include/llvm/Support/ItaniumManglingCanonicalizer.h
>> > URL: http://llvm.org/viewvc/llvm-
>> > project/llvm/trunk/include/llvm/Support/ItaniumManglingCanonicalizer.h?
>> > rev=340663&view=auto
>> > =======================================================================
>> > =======
>> > --- llvm/trunk/include/llvm/Support/ItaniumManglingCanonicalizer.h
>> > (added)
>> > +++ llvm/trunk/include/llvm/Support/ItaniumManglingCanonicalizer.h Fri
>> > Aug 24 15:31:51 2018
>> > @@ -0,0 +1,87 @@
>> > +//===--- ItaniumManglingCanonicalizer.h -------------------------*-
>> > C++ -*-===//
>> > +//
>> > +//                     The LLVM Compiler Infrastructure
>> > +//
>> > +// This file is distributed under the University of Illinois Open
>> > Source
>> > +// License. See LICENSE.TXT for details.
>> > +//
>> > +//===-----------------------------------------------------------------
>> > -----===//
>> > +//
>> > +// This file defines a class for computing equivalence classes of
>> > mangled names
>> > +// given a set of equivalences between name fragments.
>> > +//
>> > +//===-----------------------------------------------------------------
>> > -----===//
>> > +
>> > +#ifndef LLVM_SUPPORT_ITANIUMMANGLINGCANONICALIZER_H
>> > +#define LLVM_SUPPORT_ITANIUMMANGLINGCANONICALIZER_H
>> > +
>> > +#include "llvm/ADT/StringRef.h"
>> > +
>> > +#include <cstddef>
>> > +
>> > +namespace llvm {
>> > +/// Canonicalizer for mangled names.
>> > +///
>> > +/// This class allows specifying a list of "equivalent" manglings. For
>> > example,
>> > +/// you can specify that Ss is equivalent to
>> > +///   NSt3__112basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEE
>> > +/// and then manglings that refer to libstdc++'s 'std::string' will be
>> > +/// considered equivalent to manglings that are the same except that
>> > they refer
>> > +/// to libc++'s 'std::string'.
>> > +///
>> > +/// This can be used when data (eg, profiling data) is available for a
>> > version
>> > +/// of a program built in a different configuration, with
>> > correspondingly
>> > +/// different manglings.
>> > +class ItaniumManglingCanonicalizer {
>> > +public:
>> > +  ItaniumManglingCanonicalizer();
>> > +  ItaniumManglingCanonicalizer(const ItaniumManglingCanonicalizer &) =
>> > delete;
>> > +  void operator=(const ItaniumManglingCanonicalizer &) = delete;
>> > +  ~ItaniumManglingCanonicalizer();
>> > +
>> > +  enum class EquivalenceError {
>> > +    Success,
>> > +
>> > +    /// Both the equivalent manglings have already been used as
>> > components of
>> > +    /// some other mangling we've looked at. It's too late to add this
>> > +    /// equivalence.
>> > +    ManglingAlreadyUsed,
>> > +
>> > +    /// The first equivalent mangling is invalid.
>> > +    InvalidFirstMangling,
>> > +
>> > +    /// The second equivalent mangling is invalid.
>> > +    InvalidSecondMangling,
>> > +  };
>> > +
>> > +  enum class FragmentKind {
>> > +    /// The mangling fragment is a <name> (or a predefined
>> > <substitution>).
>> > +    Name,
>> > +    /// The mangling fragment is a <type>.
>> > +    Type,
>> > +    /// The mangling fragment is an <encoding>.
>> > +    Encoding,
>> > +  };
>> > +
>> > +  /// Add an equivalence between \p First and \p Second. Both
>> > manglings must
>> > +  /// live at least as long as the canonicalizer.
>> > +  EquivalenceError addEquivalence(FragmentKind Kind, StringRef First,
>> > +                                  StringRef Second);
>> > +
>> > +  using Key = uintptr_t;
>> > +
>> > +  /// Form a canonical key for the specified mangling. They key will
>> > be the
>> > +  /// same for all equivalent manglings, and different for any two
>> > +  /// non-equivalent manglings, but is otherwise unspecified.
>> > +  ///
>> > +  /// Returns Key() if (and only if) the mangling is not a valid
>> > Itanium C++
>> > +  /// ABI mangling.
>> > +  Key canonicalize(StringRef Mangling);
>> > +
>> > +private:
>> > +  struct Impl;
>> > +  Impl *P;
>> > +};
>> > +} // namespace llvm
>> > +
>> > +#endif // LLVM_SUPPORT_ITANIUMMANGLINGCANONICALIZER_H
>> >
>> > Modified: llvm/trunk/lib/Support/CMakeLists.txt
>> > URL: http://llvm.org/viewvc/llvm-
>> > project/llvm/trunk/lib/Support/CMakeLists.txt?rev=340663&r1=340662&r2=3
>> > 40663&view=diff
>> > =======================================================================
>> > =======
>> > --- llvm/trunk/lib/Support/CMakeLists.txt (original)
>> > +++ llvm/trunk/lib/Support/CMakeLists.txt Fri Aug 24 15:31:51 2018
>> > @@ -83,6 +83,7 @@ add_llvm_library(LLVMSupport
>> >    InitLLVM.cpp
>> >    IntEqClasses.cpp
>> >    IntervalMap.cpp
>> > +  ItaniumManglingCanonicalizer.cpp
>> >    JamCRC.cpp
>> >    JSON.cpp
>> >    KnownBits.cpp
>> >
>> > Added: llvm/trunk/lib/Support/ItaniumManglingCanonicalizer.cpp
>> > URL: http://llvm.org/viewvc/llvm-
>> > project/llvm/trunk/lib/Support/ItaniumManglingCanonicalizer.cpp?rev=340
>> > 663&view=auto
>> > =======================================================================
>> > =======
>> > --- llvm/trunk/lib/Support/ItaniumManglingCanonicalizer.cpp (added)
>> > +++ llvm/trunk/lib/Support/ItaniumManglingCanonicalizer.cpp Fri Aug 24
>> > 15:31:51 2018
>> > @@ -0,0 +1,307 @@
>> > +//===----------------- ItaniumManglingCanonicalizer.cpp --------------
>> > -----===//
>> > +//
>> > +//                     The LLVM Compiler Infrastructure
>> > +//
>> > +// This file is dual licensed under the MIT and the University of
>> > Illinois Open
>> > +// Source Licenses. See LICENSE.TXT for details.
>> > +//
>> > +//===-----------------------------------------------------------------
>> > -----===//
>> > +
>> > +#include "llvm/Support/ItaniumManglingCanonicalizer.h"
>> > +
>> > +#include "llvm/ADT/FoldingSet.h"
>> > +#include "llvm/ADT/StringRef.h"
>> > +#include "llvm/Demangle/ItaniumDemangle.h"
>> > +#include "llvm/Support/Allocator.h"
>> > +
>> > +#include "llvm/ADT/DenseMap.h"
>> > +#include "llvm/ADT/FoldingSet.h"
>> > +#include "llvm/ADT/StringRef.h"
>> > +
>> > +using namespace llvm;
>> > +using llvm::itanium_demangle::ForwardTemplateReference;
>> > +using llvm::itanium_demangle::Node;
>> > +using llvm::itanium_demangle::NodeKind;
>> > +
>> > +namespace {
>> > +struct FoldingSetNodeIDBuilder {
>> > +  llvm::FoldingSetNodeID &ID;
>> > +  void operator()(const Node *P) { ID.AddPointer(P); }
>> > +  void operator()(StringView Str) {
>> > +    ID.AddString(llvm::StringRef(Str.begin(), Str.size()));
>> > +  }
>> > +  template<typename T>
>> > +  typename std::enable_if<std::is_integral<T>::value ||
>> > +                          std::is_enum<T>::value>::type
>> > +  operator()(T V) {
>> > +    ID.AddInteger((unsigned long long)V);
>> > +  }
>> > +  void operator()(itanium_demangle::NodeOrString NS) {
>> > +    if (NS.isNode()) {
>> > +      ID.AddInteger(0);
>> > +      (*this)(NS.asNode());
>> > +    } else if (NS.isString()) {
>> > +      ID.AddInteger(1);
>> > +      (*this)(NS.asString());
>> > +    } else {
>> > +      ID.AddInteger(2);
>> > +    }
>> > +  }
>> > +  void operator()(itanium_demangle::NodeArray A) {
>> > +    ID.AddInteger(A.size());
>> > +    for (const Node *N : A)
>> > +      (*this)(N);
>> > +  }
>> > +};
>> > +
>> > +template<typename ...T>
>> > +void profileCtor(llvm::FoldingSetNodeID &ID, Node::Kind K, T ...V) {
>> > +  FoldingSetNodeIDBuilder Builder = {ID};
>> > +  Builder(K);
>> > +  int VisitInOrder[] = {
>> > +    (Builder(V), 0) ...,
>> > +    0 // Avoid empty array if there are no arguments.
>> > +  };
>> > +  (void)VisitInOrder;
>> > +}
>> > +
>> > +// FIXME: Convert this to a generic lambda when possible.
>> > +template<typename NodeT> struct ProfileSpecificNode {
>> > +  FoldingSetNodeID &ID;
>> > +  template<typename ...T> void operator()(T ...V) {
>> > +    profileCtor(ID, NodeKind<NodeT>::Kind, V...);
>> > +  }
>> > +};
>> > +
>> > +struct ProfileNode {
>> > +  FoldingSetNodeID &ID;
>> > +  template<typename NodeT> void operator()(const NodeT *N) {
>> > +    N->match(ProfileSpecificNode<NodeT>{ID});
>> > +  }
>> > +};
>> > +
>> > +template<> void ProfileNode::operator()(const ForwardTemplateReference
>> > *N) {
>> > +  llvm_unreachable("should never canonicalize a
>> > ForwardTemplateReference");
>> > +};
>> > +
>> > +void profileNode(llvm::FoldingSetNodeID &ID, const Node *N) {
>> > +  N->visit(ProfileNode{ID});
>> > +}
>> > +
>> > +class FoldingNodeAllocator {
>> > +  class alignas(alignof(Node *)) NodeHeader : public
>> > llvm::FoldingSetNode {
>> > +  public:
>> > +    // 'Node' in this context names the injected-class-name of the
>> > base class.
>> > +    itanium_demangle::Node *getNode() {
>> > +      return reinterpret_cast<itanium_demangle::Node *>(this + 1);
>> > +    }
>> > +    void Profile(llvm::FoldingSetNodeID &ID) { profileNode(ID,
>> > getNode()); }
>> > +  };
>> > +
>> > +  BumpPtrAllocator RawAlloc;
>> > +  llvm::FoldingSet<NodeHeader> Nodes;
>> > +
>> > +public:
>> > +  void reset() {}
>> > +
>> > +  template<typename T, typename ...Args>
>> > +  std::pair<Node*, bool> getOrCreateNode(Args &&...As) {
>> > +    llvm::FoldingSetNodeID ID;
>> > +    profileCtor(ID, NodeKind<T>::Kind, As...);
>> > +
>> > +    void *InsertPos;
>> > +    if (NodeHeader *Existing = Nodes.FindNodeOrInsertPos(ID,
>> > InsertPos))
>> > +      return {static_cast<T*>(Existing->getNode()), false};
>> > +
>> > +    static_assert(alignof(T) <= alignof(NodeHeader),
>> > +                  "underaligned node header for specific node kind");
>> > +    void *Storage =
>> > +        RawAlloc.Allocate(sizeof(NodeHeader) + sizeof(T),
>> > alignof(NodeHeader));
>> > +    NodeHeader *New = new (Storage) NodeHeader;
>> > +    T *Result = new (New->getNode()) T(std::forward<Args>(As)...);
>> > +    Nodes.InsertNode(New, InsertPos);
>> > +    return {Result, true};
>> > +  }
>> > +
>> > +  template<typename T, typename... Args>
>> > +  Node *makeNode(Args &&...As) {
>> > +    return getOrCreateNode<T>(std::forward<Args>(As)...).first;
>> > +  }
>> > +
>> > +  void *allocateNodeArray(size_t sz) {
>> > +    return RawAlloc.Allocate(sizeof(Node *) * sz, alignof(Node *));
>> > +  }
>> > +};
>> > +
>> > +// FIXME: Don't canonicalize forward template references for now,
>> > because they
>> > +// contain state (the resolved template node) that's not known at
>> > their point
>> > +// of creation.
>> > +template<>
>> > +std::pair<Node *, bool>
>> > +FoldingNodeAllocator::getOrCreateNode<ForwardTemplateReference>(size_t
>> > &Index) {
>> > +  return {new (RawAlloc.Allocate(sizeof(ForwardTemplateReference),
>> > +                                 alignof(ForwardTemplateReference)))
>> > +              ForwardTemplateReference(Index),
>> > +          true};
>> > +}
>> > +
>> > +class CanonicalizerAllocator : public FoldingNodeAllocator {
>> > +  Node *MostRecentlyCreated = nullptr;
>> > +  Node *TrackedNode = nullptr;
>> > +  bool TrackedNodeIsUsed = false;
>> > +  llvm::SmallDenseMap<Node*, Node*, 32> Remappings;
>> > +
>> > +  template<typename T, typename ...Args> Node *makeNodeSimple(Args
>> > &&...As) {
>> > +    std::pair<Node *, bool> Result =
>> > +        getOrCreateNode<T>(std::forward<Args>(As)...);
>> > +    if (Result.second) {
>> > +      // Node is new. Make a note of that.
>> > +      MostRecentlyCreated = Result.first;
>> > +    } else {
>> > +      // Node is pre-existing; check if it's in our remapping table.
>> > +      if (auto *N = Remappings.lookup(Result.first)) {
>> > +        Result.first = N;
>> > +        assert(Remappings.find(Result.first) == Remappings.end() &&
>> > +               "should never need multiple remap steps");
>> > +      }
>> > +      if (Result.first == TrackedNode)
>> > +        TrackedNodeIsUsed = true;
>> > +    }
>> > +    return Result.first;
>> > +  }
>> > +
>> > +  /// Helper to allow makeNode to be partially-specialized on T.
>> > +  template<typename T> struct MakeNodeImpl {
>> > +    CanonicalizerAllocator &Self;
>> > +    template<typename ...Args> Node *make(Args &&...As) {
>> > +      return Self.makeNodeSimple<T>(std::forward<Args>(As)...);
>> > +    }
>> > +  };
>> > +
>> > +public:
>> > +  template<typename T, typename ...Args> Node *makeNode(Args &&...As)
>> > {
>> > +    return MakeNodeImpl<T>{*this}.make(std::forward<Args>(As)...);
>> > +  }
>> > +
>> > +  void reset() { MostRecentlyCreated = nullptr; }
>> > +
>> > +  void addRemapping(Node *A, Node *B) {
>> > +    // Note, we don't need to check whether B is also remapped,
>> > because if it
>> > +    // was we would have already remapped it when building it.
>> > +    Remappings.insert(std::make_pair(A, B));
>> > +  }
>> > +
>> > +  bool isMostRecentlyCreated(Node *N) const { return
>> > MostRecentlyCreated == N; }
>> > +
>> > +  void trackUsesOf(Node *N) {
>> > +    TrackedNode = N;
>> > +    TrackedNodeIsUsed = false;
>> > +  }
>> > +  bool trackedNodeIsUsed() const { return TrackedNodeIsUsed; }
>> > +};
>> > +
>> > +/// Convert St3foo to NSt3fooE so that equivalences naming one also
>> > affect the
>> > +/// other.
>> > +template<>
>> > +struct CanonicalizerAllocator::MakeNodeImpl<
>> > +           itanium_demangle::StdQualifiedName> {
>> > +  CanonicalizerAllocator &Self;
>> > +  Node *make(Node *Child) {
>> > +    Node *StdNamespace =
>> > Self.makeNode<itanium_demangle::NameType>("std");
>> > +    if (!StdNamespace)
>> > +      return nullptr;
>> > +    return Self.makeNode<itanium_demangle::NestedName>(StdNamespace,
>> > Child);
>> > +  }
>> > +};
>> > +
>> > +// FIXME: Also expand built-in substitutions?
>> > +
>> > +using CanonicalizingDemangler =
>> > itanium_demangle::Db<CanonicalizerAllocator>;
>> > +}
>> > +
>> > +struct ItaniumManglingCanonicalizer::Impl {
>> > +  CanonicalizingDemangler Demangler = {nullptr, nullptr};
>> > +};
>> > +
>> > +ItaniumManglingCanonicalizer::ItaniumManglingCanonicalizer() : P(new
>> > Impl) {}
>> > +ItaniumManglingCanonicalizer::~ItaniumManglingCanonicalizer() { delete
>> > P; }
>> > +
>> > +ItaniumManglingCanonicalizer::EquivalenceError
>> > +ItaniumManglingCanonicalizer::addEquivalence(FragmentKind Kind,
>> > StringRef First,
>> > +                                             StringRef Second) {
>> > +  auto &Alloc = P->Demangler.ASTAllocator;
>> > +
>> > +  auto Parse = [&](StringRef Str) {
>> > +    P->Demangler.reset(Str.begin(), Str.end());
>> > +    Node *N = nullptr;
>> > +    switch (Kind) {
>> > +      // A <name>, with minor extensions to allow arbitrary namespace
>> > and
>> > +      // template names that can't easily be written as <name>s.
>> > +    case FragmentKind::Name:
>> > +      // Very special case: allow "St" as a shorthand for "3std". It's
>> > not
>> > +      // valid as a <name> mangling, but is nonetheless the most
>> > natural
>> > +      // way to name the 'std' namespace.
>> > +      if (Str.size() == 2 && P->Demangler.consumeIf("St"))
>> > +        N = P->Demangler.make<itanium_demangle::NameType>("std");
>> > +      // We permit substitutions to name templates without their
>> > template
>> > +      // arguments. This mostly just falls out, as almost all template
>> > names
>> > +      // are valid as <name>s, but we also want to parse
>> > <substitution>s as
>> > +      // <name>s, even though they're not.
>> > +      else if (Str.startswith("S"))
>> > +        // Parse the substitution and optional following template
>> > arguments.
>> > +        N = P->Demangler.parseType();
>> > +      else
>> > +        N = P->Demangler.parseName();
>> > +      break;
>> > +
>> > +      // A <type>.
>> > +    case FragmentKind::Type:
>> > +      N = P->Demangler.parseType();
>> > +      break;
>> > +
>> > +      // An <encoding>.
>> > +    case FragmentKind::Encoding:
>> > +      N = P->Demangler.parseEncoding();
>> > +      break;
>> > +    }
>> > +
>> > +    // If we have trailing junk, the mangling is invalid.
>> > +    if (P->Demangler.numLeft() != 0)
>> > +      N = nullptr;
>> > +
>> > +    // If any node was created after N, then we cannot safely remap it
>> > because
>> > +    // it might already be in use by another node.
>> > +    return std::make_pair(N, Alloc.isMostRecentlyCreated(N));
>> > +  };
>> > +
>> > +  Node *FirstNode, *SecondNode;
>> > +  bool FirstIsNew, SecondIsNew;
>> > +
>> > +  std::tie(FirstNode, FirstIsNew) = Parse(First);
>> > +  if (!FirstNode)
>> > +    return EquivalenceError::InvalidFirstMangling;
>> > +
>> > +  Alloc.trackUsesOf(FirstNode);
>> > +  std::tie(SecondNode, SecondIsNew) = Parse(Second);
>> > +  if (!SecondNode)
>> > +    return EquivalenceError::InvalidSecondMangling;
>> > +
>> > +  // If they're already equivalent, there's nothing to do.
>> > +  if (FirstNode == SecondNode)
>> > +    return EquivalenceError::Success;
>> > +
>> > +  if (FirstIsNew && !Alloc.trackedNodeIsUsed())
>> > +    Alloc.addRemapping(FirstNode, SecondNode);
>> > +  else if (SecondIsNew)
>> > +    Alloc.addRemapping(SecondNode, FirstNode);
>> > +  else
>> > +    return EquivalenceError::ManglingAlreadyUsed;
>> > +
>> > +  return EquivalenceError::Success;
>> > +}
>> > +
>> > +ItaniumManglingCanonicalizer::Key
>> > +ItaniumManglingCanonicalizer::canonicalize(StringRef Mangling) {
>> > +  P->Demangler.reset(Mangling.begin(), Mangling.end());
>> > +  return reinterpret_cast<Key>(P->Demangler.parse());
>> > +}
>> >
>> > Modified: llvm/trunk/unittests/Support/CMakeLists.txt
>> > URL: http://llvm.org/viewvc/llvm-
>> > project/llvm/trunk/unittests/Support/CMakeLists.txt?rev=340663&r1=34066
>> > 2&r2=340663&view=diff
>> > =======================================================================
>> > =======
>> > --- llvm/trunk/unittests/Support/CMakeLists.txt (original)
>> > +++ llvm/trunk/unittests/Support/CMakeLists.txt Fri Aug 24 15:31:51
>> > 2018
>> > @@ -31,6 +31,7 @@ add_llvm_unittest(SupportTests
>> >    FormatVariadicTest.cpp
>> >    GlobPatternTest.cpp
>> >    Host.cpp
>> > +  ItaniumManglingCanonicalizerTest.cpp
>> >    JSONTest.cpp
>> >    LEB128Test.cpp
>> >    LineIteratorTest.cpp
>> >
>> > Added:
>> > llvm/trunk/unittests/Support/ItaniumManglingCanonicalizerTest.cpp
>> > URL: http://llvm.org/viewvc/llvm-
>> > project/llvm/trunk/unittests/Support/ItaniumManglingCanonicalizerTest.c
>> > pp?rev=340663&view=auto
>> > =======================================================================
>> > =======
>> > --- llvm/trunk/unittests/Support/ItaniumManglingCanonicalizerTest.cpp
>> > (added)
>> > +++ llvm/trunk/unittests/Support/ItaniumManglingCanonicalizerTest.cpp
>> > Fri Aug 24 15:31:51 2018
>> > @@ -0,0 +1,315 @@
>> > +//===-------------- ItaniumManglingCanonicalizerTest.cpp -------------
>> > -----===//
>> > +//
>> > +//                     The LLVM Compiler Infrastructure
>> > +//
>> > +// This file is dual licensed under the MIT and the University of
>> > Illinois Open
>> > +// Source Licenses. See LICENSE.TXT for details.
>> > +//
>> > +//===-----------------------------------------------------------------
>> > -----===//
>> > +
>> > +#include <cstdlib>
>> > +#include "llvm/Support/ItaniumManglingCanonicalizer.h"
>> > +#include "gtest/gtest.h"
>> > +
>> > +using EquivalenceError =
>> > llvm::ItaniumManglingCanonicalizer::EquivalenceError;
>> > +using FragmentKind = llvm::ItaniumManglingCanonicalizer::FragmentKind;
>> > +
>> > +struct Equivalence {
>> > +  FragmentKind Kind;
>> > +  llvm::StringRef First;
>> > +  llvm::StringRef Second;
>> > +};
>> > +
>> > +// A set of manglings that should all be considered equivalent.
>> > +using EquivalenceClass = std::initializer_list<llvm::StringRef>;
>> > +
>> > +struct Testcase {
>> > +  // A set of equivalences to register.
>> > +  std::initializer_list<Equivalence> Equivalences;
>> > +  // A set of distinct equivalence classes created by registering the
>> > +  // equivalences.
>> > +  std::initializer_list<EquivalenceClass> Classes;
>> > +};
>> > +
>> > +static std::initializer_list<Testcase> Testcases = {
>> > +  // Three different manglings for std::string (old libstdc++, new
>> > libstdc++,
>> > +  // libc++).
>> > +  {
>> > +    {
>> > +      {FragmentKind::Type, "Ss",
>> > +
>> > "NSt3__112basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEE"},
>> > +      {FragmentKind::Type, "Ss",
>> > +       "NSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE"},
>> > +    },
>> > +    {
>> > +      {"_Z1fv"},
>> > +      {"_Z1fSs",
>> > +
>> > "_Z1fNSt3__112basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEE",
>> > +       "_Z1fNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE"},
>> > +      {"_ZNKSs4sizeEv",
>> > +
>> > "_ZNKSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEE4sizeEv",
>> > +
>> > "_ZNKSt3__112basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEE4sizeE
>> > v"},
>> > +    }
>> > +  },
>> > +
>> > +  // Check that substitutions are properly handled.
>> > +  {
>> > +    {
>> > +      // ::X <-> ::N::X<int>
>> > +      {FragmentKind::Type, "1X", "N1N1XIiEE"},
>> > +      // ::T<T<int, int>, T<int, int>> <-> T<int>
>> > +      {FragmentKind::Type, "1TIS_IiiES0_E", "1TIiE"},
>> > +      // A::B::foo <-> AB::foo
>> > +      {FragmentKind::Name, "N1A1B3fooE", "N2AB3fooE"},
>> > +    },
>> > +    {
>> > +      {"_Z1f1XPS_RS_", "_Z1fN1N1XIiEEPS1_RS1_"},
>> > +      {"_ZN1A1B3fooE1TIS1_IiiES2_EPS3_RS3_",
>> > "_ZN2AB3fooE1TIiEPS1_RS1_"},
>> > +    }
>> > +  },
>> > +
>> > +  // Check that nested equivalences are properly handled.
>> > +  {
>> > +    {
>> > +      // std::__1::char_traits == std::__cxx11::char_traits
>> > +      // (Note that this is unused and should make no difference,
>> > +      // but it should not cause us to fail to match up the cases
>> > +      // below.)
>> > +      {FragmentKind::Name,
>> > +       "NSt3__111char_traitsE",
>> > +       "NSt7__cxx1111char_traitsE"},
>> > +      // std::__1::allocator == std::allocator
>> > +      {FragmentKind::Name,
>> > +       "NSt3__19allocatorE",
>> > +       "Sa"}, // "Sa" is not strictly a <name> but we accept it as
>> > one.
>> > +      // std::__1::vector == std::vector
>> > +      {FragmentKind::Name,
>> > +       "St6vector",
>> > +       "NSt3__16vectorE"},
>> > +      // std::__1::basic_string<
>> > +      //   char
>> > +      //   std::__1::char_traits<char>,
>> > +      //   std::__1::allocator<char>> ==
>> > +      // std::__cxx11::basic_string<
>> > +      //   char,
>> > +      //   std::char_traits<char>,
>> > +      //   std::allocator<char>>
>> > +      {FragmentKind::Type,
>> > +
>> > "NSt3__112basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEE",
>> > +       "NSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE"},
>> > +      // X<A> <-> X<B>
>> > +      {FragmentKind::Type, "1XI1AE", "1XI1BE"},
>> > +      // X <-> Y
>> > +      {FragmentKind::Name, "1X", "1Y"},
>> > +    },
>> > +    {
>> > +      // f(std::string)
>> > +
>> > {"_Z1fNSt3__112basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEE",
>> > +       "_Z1fNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE"},
>> > +      // f(std::vector<int>)
>> > +      {"_Z1fSt6vectorIiSaIiEE",
>> > "_Z1fNSt3__16vectorIiNS_9allocatorIiEEEE"},
>> > +      // f(X<A>), f(X<B>), f(Y<A>), f(Y<B>)
>> > +      {"_Z1f1XI1AE", "_Z1f1XI1BE", "_Z1f1YI1AE", "_Z1f1YI1BE"},
>> > +      // f(X<C>), f(Y<C>)
>> > +      {"_Z1f1XI1CE", "_Z1f1YI1CE"},
>> > +    }
>> > +  },
>> > +
>> > +  // Check namespace equivalences.
>> > +  {
>> > +    {
>> > +      // std::__1 == std::__cxx11
>> > +      {FragmentKind::Name, "St3__1", "St7__cxx11"},
>> > +      // std::__1::allocator == std::allocator
>> > +      {FragmentKind::Name, "NSt3__19allocatorE", "Sa"},
>> > +      // std::vector == std::__1::vector
>> > +      {FragmentKind::Name, "St6vector", "NSt3__16vectorE"},
>> > +      // std::__cxx11::char_traits == std::char_traits
>> > +      // (This indirectly means that std::__1::char_traits ==
>> > std::char_traits,
>> > +      // due to the std::__cxx11 == std::__1 equivalence, which is
>> > what we rely
>> > +      // on below.)
>> > +      {FragmentKind::Name, "NSt7__cxx1111char_traitsE",
>> > "St11char_traits"},
>> > +    },
>> > +    {
>> > +      // f(std::foo)
>> > +      {"_Z1fNSt7__cxx113fooE",
>> > +       "_Z1fNSt3__13fooE"},
>> > +      // f(std::string)
>> > +      {"_Z1fNSt7__cxx1111char_traitsIcEE",
>> > +       "_Z1fNSt3__111char_traitsIcEE",
>> > +       "_Z1fSt11char_traitsIcE"},
>> > +      // f(std::string)
>> > +
>> > {"_Z1fNSt3__112basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEE",
>> > +       "_Z1fNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE"},
>> > +      // f(std::vector<int>)
>> > +      {"_Z1fSt6vectorIiSaIiEE",
>> > "_Z1fNSt3__16vectorIiNS_9allocatorIiEEEE"},
>> > +    }
>> > +  },
>> > +
>> > +  // Check namespace equivalences for namespace 'std'. We support
>> > using 'St'
>> > +  // for this, despite it not technically being a <name>.
>> > +  {
>> > +    {
>> > +      // std::__1 == std
>> > +      {FragmentKind::Name, "St3__1", "St"},
>> > +      // std::__1 == std::__cxx11
>> > +      {FragmentKind::Name, "St3__1", "St7__cxx11"},
>> > +      // FIXME: Should a 'std' equivalence also cover the predefined
>> > +      // substitutions?
>> > +      // std::__1::allocator == std::allocator
>> > +      {FragmentKind::Name, "NSt3__19allocatorE", "Sa"},
>> > +    },
>> > +    {
>> > +      {"_Z1fSt3foo", "_Z1fNSt3__13fooE", "_Z1fNSt7__cxx113fooE"},
>> > +      {"_Z1fNSt3bar3bazE", "_Z1fNSt3__13bar3bazE"},
>> > +      // f(std::string)
>> > +
>> > {"_Z1fNSt3__112basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEE",
>> > +       "_Z1fNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE"},
>> > +      // f(std::vector<int>)
>> > +      {"_Z1fSt6vectorIiSaIiEE",
>> > "_Z1fNSt3__16vectorIiNS_9allocatorIiEEEE"},
>> > +    }
>> > +  },
>> > +
>> > +  // Check mutually-recursive equivalences.
>> > +  {
>> > +    {
>> > +      {FragmentKind::Type, "1A", "1B"},
>> > +      {FragmentKind::Type, "1A", "1C"},
>> > +      {FragmentKind::Type, "1D", "1B"},
>> > +      {FragmentKind::Type, "1C", "1E"},
>> > +    },
>> > +    {
>> > +      {"_Z1f1A", "_Z1f1B", "_Z1f1C", "_Z1f1D", "_Z1f1E"},
>> > +      {"_Z1f1F"},
>> > +    }
>> > +  },
>> > +
>> > +  // Check <encoding>s.
>> > +  {
>> > +    {
>> > +      {FragmentKind::Encoding, "1fv", "1gv"},
>> > +    },
>> > +    {
>> > +      // f(void) -> g(void)
>> > +      {"_Z1fv", "_Z1gv"},
>> > +      // static local 'n' in f(void) -> static local 'n' in g(void)
>> > +      {"_ZZ1fvE1n", "_ZZ1gvE1n"},
>> > +    }
>> > +  },
>> > +
>> > +  // Corner case: the substitution can appear within its own
>> > expansion.
>> > +  {
>> > +    {
>> > +      // X <-> Y<X>
>> > +      {FragmentKind::Type, "1X", "1YI1XE"},
>> > +      // A<B> <-> B
>> > +      {FragmentKind::Type, "1AI1BE", "1B"},
>> > +    },
>> > +    {
>> > +      // f(X) == f(Y<X>) == f(Y<Y<X>>) == f(Y<Y<Y<X>>>)
>> > +      {"_Z1f1X", "_Z1f1YI1XE", "_Z1f1YIS_I1XEE",
>> > "_Z1f1YIS_IS_I1XEEE"},
>> > +      // f(B) == f(A<B>) == f(A<A<B>>) == f(A<A<A<B>>>)
>> > +      {"_Z1f1B", "_Z1f1AI1BE", "_Z1f1AIS_I1BEE",
>> > "_Z1f1AIS_IS_I1BEEE"},
>> > +    }
>> > +  },
>> > +
>> > +  // Redundant equivalences are accepted (and have no effect).
>> > +  {
>> > +    {
>> > +      {FragmentKind::Name, "3std", "St"},
>> > +      {FragmentKind::Name, "1X", "1Y"},
>> > +      {FragmentKind::Name, "N1X1ZE", "N1Y1ZE"},
>> > +    },
>> > +    {}
>> > +  },
>> > +
>> > +  // ForwardTemplateReference does not support canonicalization.
>> > +  // FIXME: We should consider ways of fixing this, perhaps by
>> > eliminating
>> > +  // the ForwardTemplateReference node with a tree transformation.
>> > +  {
>> > +    {
>> > +      // X::operator T() <with T = A> == Y::operator T() <with T = A>
>> > +      {FragmentKind::Encoding, "N1XcvT_I1AEEv", "N1YcvT_I1AEEv"},
>> > +      // A == B
>> > +      {FragmentKind::Name, "1A", "1B"},
>> > +    },
>> > +    {
>> > +      // All combinations result in unique equivalence classes.
>> > +      {"_ZN1XcvT_I1AEEv"},
>> > +      {"_ZN1XcvT_I1BEEv"},
>> > +      {"_ZN1YcvT_I1AEEv"},
>> > +      {"_ZN1YcvT_I1BEEv"},
>> > +      // Even giving the same string twice gives a new class.
>> > +      {"_ZN1XcvT_I1AEEv"},
>> > +    }
>> > +  },
>> > +};
>> > +
>> > +TEST(ItaniumManglingCanonicalizerTest, TestTestcases) {
>> > +  for (const auto &Testcase : Testcases) {
>> > +    llvm::ItaniumManglingCanonicalizer Canonicalizer;
>> > +    for (const auto &Equiv : Testcase.Equivalences) {
>> > +      auto Result =
>> > +          Canonicalizer.addEquivalence(Equiv.Kind, Equiv.First,
>> > Equiv.Second);
>> > +      EXPECT_EQ(Result, EquivalenceError::Success)
>> > +          << "couldn't add equivalence between " << Equiv.First << "
>> > and "
>> > +          << Equiv.Second;
>> > +    }
>> > +
>> > +    using CanonKey = llvm::ItaniumManglingCanonicalizer::Key;
>> > +    std::map<CanonKey, llvm::StringRef> Found;
>> > +    for (const auto &Class : Testcase.Classes) {
>> > +      CanonKey ClassKey = {};
>> > +      for (llvm::StringRef Str : Class) {
>> > +        CanonKey ThisKey = Canonicalizer.canonicalize(Str);
>> > +        EXPECT_NE(ThisKey, CanonKey()) << "couldn't canonicalize " <<
>> > Str;
>> > +        if (ClassKey) {
>> > +          EXPECT_EQ(ThisKey, ClassKey)
>> > +              << Str << " not in the same class as " <<
>> > *Class.begin();
>> > +        } else {
>> > +          ClassKey = ThisKey;
>> > +        }
>> > +      }
>> > +      EXPECT_TRUE(Found.insert({ClassKey, *Class.begin()}).second)
>> > +          << *Class.begin() << " is in the same class as " <<
>> > Found[ClassKey];
>> > +    }
>> > +  }
>> > +}
>> > +
>> > +TEST(ItaniumManglingCanonicalizerTest, TestInvalidManglings) {
>> > +  llvm::ItaniumManglingCanonicalizer Canonicalizer;
>> > +  EXPECT_EQ(Canonicalizer.addEquivalence(FragmentKind::Type, "",
>> > "1X"),
>> > +            EquivalenceError::InvalidFirstMangling);
>> > +  EXPECT_EQ(Canonicalizer.addEquivalence(FragmentKind::Type, "1X",
>> > "1ab"),
>> > +            EquivalenceError::InvalidSecondMangling);
>> > +  EXPECT_EQ(Canonicalizer.canonicalize("_Z3fooE"),
>> > +            llvm::ItaniumManglingCanonicalizer::Key());
>> > +  EXPECT_EQ(Canonicalizer.canonicalize("foo"),
>> > +            llvm::ItaniumManglingCanonicalizer::Key());
>> > +
>> > +  // A reference to a template parameter ('T_' etc) cannot appear in a
>> > <name>,
>> > +  // because we don't have template arguments to bind to it. (The
>> > arguments in
>> > +  // an 'I ... E' construct in the <name> aren't registered as
>> > +  // backreferenceable arguments in this sense, because they're not
>> > part of
>> > +  // the template argument list of an <encoding>.
>> > +  EXPECT_EQ(Canonicalizer.addEquivalence(FragmentKind::Name,
>> > "N1XcvT_I1AEE",
>> > +                                         "1f"),
>> > +            EquivalenceError::InvalidFirstMangling);
>> > +}
>> > +
>> > +TEST(ItaniumManglingCanonicalizerTest, TestBadEquivalenceOrder) {
>> > +  llvm::ItaniumManglingCanonicalizer Canonicalizer;
>> > +  EXPECT_EQ(Canonicalizer.addEquivalence(FragmentKind::Type, "N1P1XE",
>> > "N1Q1XE"),
>> > +            EquivalenceError::Success);
>> > +  EXPECT_EQ(Canonicalizer.addEquivalence(FragmentKind::Type, "1P",
>> > "1Q"),
>> > +            EquivalenceError::ManglingAlreadyUsed);
>> > +
>> > +  EXPECT_EQ(Canonicalizer.addEquivalence(FragmentKind::Type, "N1C1XE",
>> > "N1A1YE"),
>> > +            EquivalenceError::Success);
>> > +  EXPECT_EQ(Canonicalizer.addEquivalence(FragmentKind::Type, "1A",
>> > "1B"),
>> > +            EquivalenceError::Success);
>> > +  EXPECT_EQ(Canonicalizer.addEquivalence(FragmentKind::Type, "1C",
>> > "1D"),
>> > +            EquivalenceError::Success);
>> > +  EXPECT_EQ(Canonicalizer.addEquivalence(FragmentKind::Type, "1B",
>> > "1D"),
>> > +            EquivalenceError::ManglingAlreadyUsed);
>> > +}
>> >
>> >
>> > _______________________________________________
>> > llvm-commits mailing list
>> > llvm-commits at lists.llvm.org
>> > http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at lists.llvm.org
>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180826/f5019e73/attachment.html>


More information about the llvm-commits mailing list