[build] SR-11880 skip building compiler-rt with --xcode

[ixrevo]

After merging #28341 ./utils/build-script build the compiler-rt even with --xcode parameter, which breaks a build. This PR resolves this issue by skipping compiler-rt in cases when it is not supported (build with Xcode generator and/or cross-compiling).

Resolves https://bugs.swift.org/browse/SR-11880.

@shahmishal

@swift-ci Please test

[ixrevo]

@swift-ci Please smoke test

[CodaFi]

This seems like an answer to fix the xcode build, but I’m more curious as to a root cause for VALID_ARCH not containing Haswell on the compiler-rt side. I think @danliew-apple was fighting the CMake there last. Any ideas?

[CodaFi]

@swift-ci please smoke test

[ixrevo]

Hi, @CodaFi! Unfortunately, I'm not competent enough to answer your question, but a quick search in the repo gave me this https://github.com/apple/llvm-project/blob/5b4cbf67ff36f82ab95551e8a5e51b8b0b7bfce2/compiler-rt/cmake/Modules/CompilerRTDarwinUtils.cmake#L136

  # The simple program will build for x86_64h on the simulator because it is 
  # compatible with x86_64 libraries (mostly), but since x86_64h isn't actually
  # a valid or useful architecture for the iOS simulator we should drop it.
  if(${os} MATCHES "^(iossim|tvossim|watchossim)$")
    list(REMOVE_ITEM archs "x86_64h")
  endif()

[CodaFi]

Hm, makes sense.

Either way, this LGTM. Thank you so so much for unblocking the Xcode build. It’s a huge gateway to getting more external contributors.

⛵️

[CodaFi]

Please be sure to update the SR.

[danliew-apple]

I think @danliew-apple was fighting the CMake there last. Any ideas?

Last time I looked the Xcode build of Compiler-rt was very broken. It seemed like a bug in CMake where the generated build rules didn't work correctly. I gave up in the end.