Error when compiling with GCC 7

[Diana-Dubov]

According to llvm documentation it shall [support GCC 7] (https://llvm.org/docs/GettingStarted.html#host-c-toolchain-both-compiler-and-standard-library).
This commit: [ADT] Add deduction guide for iterator_range breaks it.
Getting this error when compiling with GCC7.4:

     std::void_t<decltype(static_cast<To>(
                          ^~~~~~~~~~~~~~~~
         std::declval<std::add_rvalue_reference_t<From>>()))>> : std::true_type {
         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from external/llvm-project/mlir/include/mlir/Support/TypeID.h:20:0,
                 from external/llvm-project/mlir/include/mlir/IR/MLIRContext.h:13,
                 from external/llvm-project/mlir/include/mlir/IR/DialectRegistry.h:16,
                 from external/llvm-project/mlir/include/mlir/IR/Dialect.h:16,
                 from external/llvm-project/mlir/include/mlir/IR/OpDefinition.h:22,
                 from external/llvm-project/mlir/include/mlir/IR/Builders.h:12,
                 from external/llvm-project/mlir/include/mlir/IR/PatternMatch.h:12,
                 from external/llvm-project/mlir/lib/IR/PatternMatch.cpp:9:
external/llvm-project/llvm/include/llvm/ADT/STLExtras.h:399:3: note: candidate: llvm::mapped_iterator_base<DerivedT, ItTy, ReferenceTy>::mapped_iterator_base(ItTy) [with DerivedT = mlir::ValueTypeIterator<llvm::detail::indexed_accessor_range_base<mlir::ResultRange, mlir::detail::OpResultImpl*, mlir::OpResult, mlir::OpResult, mlir::OpResult>::iterator>; ItTy = llvm::detail::indexed_accessor_range_base<mlir::ResultRange, mlir::detail::OpResultImpl*, mlir::OpResult, mlir::OpResult, mlir::OpResult>::iterator; ReferenceTy = mlir::Type]
   mapped_iterator_base(ItTy U)

It is fixed in GCC8 and later, but the issue at GCC7 exists and breaks the promise to support this version.

Thank you

[llvmbot]

@llvm/issue-subscribers-mlir

[Diana-Dubov]

@steakhal can you please take a look?

[joker-eph]

@Diana-Dubov I left a comment in the revision, this is the most usual way to signal that something is broken.

[steakhal]

I'll look at this on the weekend.

[steakhal]

Here is a partial revert "fix" for the build, by removing the SFINAE constraint and putting back the TODO which was originally there.
https://reviews.llvm.org/D155441

[unterumarmung]

Why not just ifdef it on the version of the compiler?
I mean, if someone writes a code that doesn't satisfy SFINAE with GCC7, then there will be compiler-error in the CI

[steakhal]

Please reply to the patch. I don't usually monitor Github notifications.

Why not just ifdef it on the version of the compiler?
I mean, if someone writes a code that doesn't satisfy SFINAE with GCC7, then there will be compiler-error in the CI

Accommodated the patch according to this.

[MaskRay]

Not for this specific instance, but in general I suspect that 7.4 may not be good enough to build llvm-project, and we may require 7.5.

I created Require GCC>=7.5 as GCC 7.3 cannot build llvm.