It seems these compilers are detected on Windows:

packages:
  llvm:
    externals:
    - spec: [email protected]+clang+lld~lldb
      prefix: C:\Program Files\LLVM
      extra_attributes:
        compilers:
          c: C:\Program Files\LLVM\bin\clang.exe
          cxx: C:\Program Files\LLVM\bin\clang++.exe
          ld: C:\Program Files\LLVM\bin\ld.lld.exe
  gcc:
    externals:
    - spec: [email protected] languages='c,c++,fortran'
      prefix: C:\Strawberry\c
      extra_attributes:
        compilers:
          c: C:\Strawberry\c\bin\gcc.exe
          cxx: C:\Strawberry\c\bin\g++.exe
          fortran: C:\Strawberry\c\bin\gfortran.exe
    - spec: [email protected] languages='c,c++,fortran'
      prefix: C:\mingw64
      extra_attributes:
        compilers:
          c: C:\mingw64\bin\gcc.exe
          cxx: C:\mingw64\bin\g++.exe
          fortran: C:\mingw64\bin\gfortran.exe

Apparently one needs to source:

. "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\Common7\Tools\Launch-VsDevShell.ps1"

to get compilers in PATH on windows-latest.

Why is ld: C:\Program Files\LLVM\bin\ld.lld.exe added?

Why is ld: C:\Program Files\LLVM\bin\ld.lld.exe added?

Not part of this PR. It's the external search logic in LLVM that finds it.

I have some questions.

How will mixed toolchains be accounted for?

Say I have a compiler that only supplies languages=[c,cxx] like apple-clang. Is the expectation to manually add a fortran entry when needed and then still use foo%apple-clang if foo depends_on("fortran") or would I do something likefoo%apple-clang.c%apple-clang.cxx%gcc.fortran? Also is double % syntactically correct here? I'm already forgetting some of these nuances.

Seems like the former will work for many cases and meets some of my requests for a concise way to represent a tool chain. However, it feels really hacky and also seems super limited. The latter is really verbose, but also very clear. Anything in the middle seems scary like foo%apple-clang%gcc. I suppose foo%apple-clang%gcc.fortran would be a little better if the concretizer were smart enough to know and enforce that apple-clang.[c,cxx] need to take precedent over gcc.[c,cxx].

Anyways what's the correct syntax here?

@psakievich Syntax changes are discussed in #44379 (I see a 👍 from you there 😉 ) There's nothing like %gcc.cxx or similar syntax, so not sure where this is coming from. Using a dot is ambiguous: gcc.fortran right now parses as the fortran package from the gcc repository.

The % gets translated to "direct build dependency".

To you question on mixed toolchain, the example we have in the issue reads:

foo %apple-clang %[virtuals=fortran] gcc

@psakievich Syntax changes are discussed in #44379 (I see a 👍 from you there 😉 )

Sorry. Too many sources of input to keep track of. I'll review and move my comments there.

Superseded by #45784