GCC Compatibility #787
GCC Compatibility #787
Conversation
…related to json ser/de with LLVMAliasSet and LLVMBasedICFG that were not detected untio now, because of nlohmann's implicit conversions feature
CMakeLists.txt
Outdated
| @@ -82,6 +82,11 @@ option(PHASAR_BUILD_MODULES "Build C++20 modules for phasar" OFF) | |||
|
|
|||
| include(CheckCXXCompilerFlag) | |||
|
|
|||
| if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU") | |||
| # Match clang's behavior, when handling naming collisions, such as "using l_t = l_t;" | |||
| string(APPEND CMAKE_CXX_FLAGS " -fpermissive") | |||
There was a problem hiding this comment.
I would highly advice against adding permissive here. Permissive turns many actual errors into warnings/nothing. It is not recommended to be turned on.
Either we should fine another fix for the collisions or, even better, just fix them.
There was a problem hiding this comment.
Yes, I agree.
Fixed it
| static constexpr auto isLLVMZeroValue = [](const llvm::Value *V) noexcept { | ||
| return V == getInstance(); | ||
| return isZeroValueHelper(V); |
There was a problem hiding this comment.
What's the underlying reason to have a lambda here instead of a normal static member?
| @@ -36,6 +36,10 @@ class LLVMZeroValue : public llvm::GlobalVariable { | |||
| LLVMZeroValue(llvm::Module &Mod); | |||
|
|
|||
| static constexpr llvm::StringLiteral LLVMZeroValueInternalName = "zero_value"; | |||
| static bool isZeroValueHelper(const llvm::Value *V) noexcept { | |||
| // Need this helper function to make gcc happy | |||
There was a problem hiding this comment.
What's the problem here for gcc?
There was a problem hiding this comment.
For gcc, the type LLVMZeroValue is incomplete in the lambda, so it cannot perform the implicit conversion to pointer-to-base for the comparison V == getInstance()
| @@ -942,6 +942,14 @@ class IDEInstInteractionAnalysisT | |||
|
|
|||
| inline l_t join(l_t Lhs, l_t Rhs) override { return joinImpl(Lhs, Rhs); } | |||
|
|
|||
| struct IIAAKillOrReplaceEF; | |||
| struct IIAAAddLabelsEF; | |||
| // These friend declarations are needed to make gcc happy | |||
There was a problem hiding this comment.
Could you maybe add the problem here in the comment, it's a bit hard to figure out why you needed this here.
| struct IIAAKillOrReplaceEF; | ||
| struct IIAAAddLabelsEF; | ||
| // These friend declarations are needed to make gcc happy | ||
| friend llvm::raw_ostream &operator<<(llvm::raw_ostream &OS, |
There was a problem hiding this comment.
I assume making them hidden friends was not an option because you would need the full def. for the implementation?
There was a problem hiding this comment.
Actually, we can simplify this, as we only need the friend on the IIA analysis class, not on the edge-functions.
Thanks for pointing that out!
Changed it.
Make phasar compile with gcc (g++-11 and g++-13) and uncover (and fix) two bugs related to json ser/de with LLVMAliasSet and LLVMBasedICFG that were not detected until now, because of nlohmann's implicit conversions feature.