Opaque Pointers Integration (#730)

* replaced llvm-14 with llvm-15 in relevant files

* annotation fix + test fix

* removed bitcasts and fixed tests

* Fix Annotation.cpp + TaintConfigTest

* Quick-fix LLVMTypeHierarchyTest

* fix one IIA test

* Fix StringtestCpp for generalized LCA

* first half of tests fixed

* fixed first half of tests

* fixed all but one test in IDEInstInteractionAnalysisTest

* fixed all but two tests

* Fix IIAFlowFact equality

* fixed some newly failed tests

* trade soundness for precision in LLVMAliasSet

* ci.yml update

* Revert "fixed some newly failed tests"

* pre-commit hook

* Two Tests + xtaint09 test fix for pipeline

* Basic Opaque Pointer Impl, bugged

* switching to DebugInfoFinder

* re-add the quick-fix for LLVMTypeHierarchy

* OpaquePtr type mapping, missing subroutines

* Introducing a pass to save ptr types

* Revert "Introducing a pass to save ptr types"

This reverts commit 2a91b6d.

* moving phasar to DIBasedTypeHierarchy

* full switch to DIBasedTypeHierarchy + Test fixes

* fixed PathTracingTest

* dtaresolver deprecated and test fixes

* Fixed OTFTest

* trimmed trailing whitespace

* minor fixes

* readded TypeToDIType map for RTAResolver

* pre-commit clang-format fix

* pre-commit clang-format llvmbasedicfg.cpp

* moved RTAResolver to DITypes

* implemented review suggestions

* Log error if trying to instantiate DTAResolver + minor

* Add breaking changes

* Also compare gep type in IIA EqualGEPDescriptor

* debugging CI

* debugging

* DITH fix attempt for CI

* fix build

* Start adding more sophisticated type extraction (WIP)

* Handle function calls in getVarTypeFromIR

* better fallback handling for getDebugLocation, etc

* cleanup getVarTypeFromIR

* Add libzstd-dev as apt dependency; otherwise the docker build fails on arm

---------

Co-authored-by: mxHuber <huber.maximilian.leo@gmail.com>
Co-authored-by: Fabian Schiebel <fabianbs@mail.uni-paderborn.de>

Author: 3 people

BreakingChanges.md

@@ -2,6 +2,8 @@
 
 ## development HEAD
 
+- The `DTAResolver` and the cli option `--call-graph-analysis=dta` do not work anymore (due to opaque pointers) and will be removed for the next release. Please use the `OTF` or `RTA` resolver instead.
+- The default type-hierarchy implementation has been changed from `LLVMTypeHierarchy` to `DIBasedTypeHierarchy`. This also requires all affected analyses to be performed on LLVM IR that contains debug information.
 - Removed the phasar-library `phasar_controller`. It is now part of the tool `phasar-cli`.
 - The API of the `TypeHierarchy` interface (and thus the `LLVMTypeHierarchy` and `DIBasedTypeHierarchy` as well) has changed:
   - No handling of the super-type relation (only sub-types)

CMakeLists.txt

@@ -312,7 +312,7 @@ option(USE_LLVM_FAT_LIB  "Link against libLLVM.so instead of the individual LLVM
 
 # LLVM
 if (NOT PHASAR_LLVM_VERSION)
-  set(PHASAR_LLVM_VERSION 14)
+  set(PHASAR_LLVM_VERSION 15)
 endif()
 include(add_llvm)
 add_llvm()

bootstrap.sh

@@ -6,10 +6,10 @@ source ./utils/safeCommandsSet.sh
 
 readonly PHASAR_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null 2>&1 && pwd )"
 PHASAR_INSTALL_DIR="/usr/local/phasar"
-LLVM_INSTALL_DIR="/usr/local/llvm-14"
+LLVM_INSTALL_DIR="/usr/local/llvm-15"
 
 NUM_THREADS=$(nproc)
-LLVM_RELEASE=llvmorg-14.0.6
+LLVM_RELEASE=llvmorg-15.0.7
 DO_UNIT_TEST=true
 DO_INSTALL=false
 BUILD_TYPE=Release

cmake/phasar_macros.cmake

@@ -163,17 +163,6 @@ function(generate_ll_file)
   set(GEN_C_FLAGS -fno-discard-value-names -emit-llvm -S -w)
   set(GEN_CMD_COMMENT "[LL]")
 
-  string(REGEX MATCH "clang\\+*-?[0-9]*$" compiler "${CMAKE_CXX_COMPILER}")
-  if (compiler)
-    if(PHASAR_LLVM_VERSION GREATER_EQUAL 15)
-      list(APPEND GEN_CXX_FLAGS -Xclang -no-opaque-pointers)
-    endif()
-
-    if(PHASAR_LLVM_VERSION GREATER_EQUAL 15)
-      list(APPEND GEN_C_FLAGS -Xclang -no-opaque-pointers)
-    endif()
-  endif()
-
   if(GEN_LL_MEM2REG)
     list(APPEND GEN_CXX_FLAGS -Xclang -disable-O0-optnone)
     list(APPEND GEN_C_FLAGS -Xclang -disable-O0-optnone)
@@ -219,7 +208,7 @@ function(generate_ll_file)
       add_custom_command(
       OUTPUT ${test_code_ll_file}
       COMMAND ${GEN_CMD} ${test_code_file_path} -o ${test_code_ll_file}
-      COMMAND ${CMAKE_CXX_COMPILER_LAUNCHER} ${opt} -mem2reg -S -opaque-pointers=0 ${test_code_ll_file} -o ${test_code_ll_file}
+      COMMAND ${CMAKE_CXX_COMPILER_LAUNCHER} ${opt} -mem2reg -S ${test_code_ll_file} -o ${test_code_ll_file}
       COMMENT ${GEN_CMD_COMMENT}
       DEPENDS ${GEN_LL_FILE}
       VERBATIM

include/phasar/DataFlow/IfdsIde/Solver/IDESolver.h

@@ -121,7 +121,7 @@ class IDESolver
             llvm::StringRef(NToString(Cells[I].getRowKey())).trim().str();
 
         std::string NodeStr =
-            ICF->getFunctionName(ICF->getFunctionOf(Curr)) + "::" + NStr;
+            ICF->getFunctionName(ICF->getFunctionOf(Curr)).str() + "::" + NStr;
         J[DataFlowID][NodeStr];
         std::string FactStr =
             llvm::StringRef(DToString(Cells[I].getColumnKey())).trim().str();

include/phasar/PhasarLLVM/ControlFlow/LLVMBasedCallGraphBuilder.h

@@ -17,14 +17,14 @@
 namespace psr {
 class LLVMProjectIRDB;
 enum class CallGraphAnalysisType;
-class LLVMTypeHierarchy;
+class DIBasedTypeHierarchy;
 class LLVMVFTableProvider;
 class Resolver;
 
 [[nodiscard]] LLVMBasedCallGraph
 buildLLVMBasedCallGraph(LLVMProjectIRDB &IRDB, CallGraphAnalysisType CGType,
                         llvm::ArrayRef<const llvm::Function *> EntryPoints,
-                        LLVMTypeHierarchy &TH, LLVMVFTableProvider &VTP,
+                        DIBasedTypeHierarchy &TH, LLVMVFTableProvider &VTP,
                         LLVMAliasInfoRef PT = nullptr,
                         Soundness S = Soundness::Soundy);
 
@@ -36,7 +36,7 @@ buildLLVMBasedCallGraph(const LLVMProjectIRDB &IRDB, Resolver &CGResolver,
 [[nodiscard]] LLVMBasedCallGraph
 buildLLVMBasedCallGraph(LLVMProjectIRDB &IRDB, CallGraphAnalysisType CGType,
                         llvm::ArrayRef<std::string> EntryPoints,
-                        LLVMTypeHierarchy &TH, LLVMVFTableProvider &VTP,
+                        DIBasedTypeHierarchy &TH, LLVMVFTableProvider &VTP,
                         LLVMAliasInfoRef PT = nullptr,
                         Soundness S = Soundness::Soundy);
 

include/phasar/PhasarLLVM/ControlFlow/LLVMBasedICFG.h

@@ -36,10 +36,8 @@
 #include "llvm/IR/Value.h"
 #include "llvm/Support/raw_ostream.h"
 
-#include <memory>
-
 namespace psr {
-class LLVMTypeHierarchy;
+class DIBasedTypeHierarchy;
 class LLVMProjectIRDB;
 class Resolver;
 
@@ -73,7 +71,7 @@ class LLVMBasedICFG : public LLVMBasedCFG, public ICFGBase<LLVMBasedICFG> {
   /// IRDB. True by default
   explicit LLVMBasedICFG(LLVMProjectIRDB *IRDB, CallGraphAnalysisType CGType,
                          llvm::ArrayRef<std::string> EntryPoints = {},
-                         LLVMTypeHierarchy *TH = nullptr,
+                         DIBasedTypeHierarchy *TH = nullptr,
                          LLVMAliasInfoRef PT = nullptr,
                          Soundness S = Soundness::Soundy,
                          bool IncludeGlobals = true);
@@ -93,7 +91,7 @@ class LLVMBasedICFG : public LLVMBasedCFG, public ICFGBase<LLVMBasedICFG> {
   explicit LLVMBasedICFG(const LLVMProjectIRDB *IRDB,
                          const CallGraphData &SerializedCG);
 
-  // Deleter of LLVMTypeHierarchy may be unknown here...
+  // Deleter of DIBasedTypeHierarchy may be unknown here...
   ~LLVMBasedICFG();
 
   LLVMBasedICFG(const LLVMBasedICFG &) = delete;

include/phasar/PhasarLLVM/ControlFlow/LLVMVFTableProvider.h

@@ -16,7 +16,7 @@
 
 namespace llvm {
 class Module;
-class StructType;
+class DIType;
 class GlobalVariable;
 } // namespace llvm
 
@@ -28,12 +28,12 @@ class LLVMVFTableProvider {
   explicit LLVMVFTableProvider(const llvm::Module &Mod);
   explicit LLVMVFTableProvider(const LLVMProjectIRDB &IRDB);
 
-  [[nodiscard]] bool hasVFTable(const llvm::StructType *Type) const;
+  [[nodiscard]] bool hasVFTable(const llvm::DIType *Type) const;
   [[nodiscard]] const LLVMVFTable *
-  getVFTableOrNull(const llvm::StructType *Type) const;
+  getVFTableOrNull(const llvm::DIType *Type) const;
 
 private:
-  std::unordered_map<const llvm::StructType *, LLVMVFTable> TypeVFTMap;
+  std::unordered_map<const llvm::DIType *, LLVMVFTable> TypeVFTMap;
 };
 } // namespace psr
 

include/phasar/PhasarLLVM/ControlFlow/Resolver/CHAResolver.h

@@ -25,11 +25,11 @@ class CallBase;
 } // namespace llvm
 
 namespace psr {
-class LLVMTypeHierarchy;
+class DIBasedTypeHierarchy;
 class CHAResolver : public Resolver {
 public:
   CHAResolver(const LLVMProjectIRDB *IRDB, const LLVMVFTableProvider *VTP,
-              const LLVMTypeHierarchy *TH);
+              const DIBasedTypeHierarchy *TH);
 
   // Deleting an incomplete type (LLVMTypeHierarchy) is UB, so instantiate the
   // dtor in CHAResolver.cpp
@@ -45,7 +45,7 @@ class CHAResolver : public Resolver {
   }
 
 protected:
-  MaybeUniquePtr<const LLVMTypeHierarchy, true> TH;
+  MaybeUniquePtr<const DIBasedTypeHierarchy, true> TH;
 };
 } // namespace psr
 

include/phasar/PhasarLLVM/ControlFlow/Resolver/DTAResolver.h

@@ -19,6 +19,7 @@
 
 #include "phasar/PhasarLLVM/ControlFlow/Resolver/CHAResolver.h"
 #include "phasar/PhasarLLVM/Pointer/TypeGraphs/CachedTypeGraph.h"
+#include "phasar/PhasarLLVM/TypeHierarchy/DIBasedTypeHierarchy.h"
 // To switch the TypeGraph
 // #include "phasar/PhasarLLVM/Pointer/TypeGraphs/LazyTypeGraph.h"
 
@@ -33,12 +34,13 @@ class BitCastInst;
 
 namespace psr {
 
-class DTAResolver : public CHAResolver {
+class [[deprecated("Does not work with opaque pointers anymore")]] DTAResolver
+    : public CHAResolver {
 public:
   using TypeGraph_t = CachedTypeGraph;
 
   DTAResolver(const LLVMProjectIRDB *IRDB, const LLVMVFTableProvider *VTP,
-              const LLVMTypeHierarchy *TH);
+              const DIBasedTypeHierarchy *TH);
 
   ~DTAResolver() override = default;
 
@@ -48,8 +50,8 @@ class DTAResolver : public CHAResolver {
 
   [[nodiscard]] std::string str() const override;
 
-  [[nodiscard]] bool
-  mutatesHelperAnalysisInformation() const noexcept override {
+  [[nodiscard]] bool mutatesHelperAnalysisInformation()
+      const noexcept override {
     return false;
   }
 
@@ -60,13 +62,14 @@ class DTAResolver : public CHAResolver {
    * An heuristic that return true if the bitcast instruction is interesting to
    * take into the DTA relational graph
    */
-  static bool
-  heuristicAntiConstructorThisType(const llvm::BitCastInst *BitCast);
+  static bool heuristicAntiConstructorThisType(
+      const llvm::BitCastInst *BitCast);
 
   /**
    * Another heuristic that return true if the bitcast instruction is
    * interesting to take into the DTA relational graph (use the presence or not
    * of vtable)
+
    */
   bool heuristicAntiConstructorVtablePos(const llvm::BitCastInst *BitCast);
 };