Add i686-pc-windows-gnullvm target

[jeremyd2019]

Adds i686-pc-windows-gnullvm target.

Fixes: #2960

[jeremyd2019]

On my fork, the "cross" test fails now. Oddly, I modified the "cross" workflow first, and it actually succeeded before I started adding i686_gnullvm elsewhere. I guess it's testing a mix between the current git crates/libs/targets and older published crates/targets/..., where there is no such i686_gnullvm version? That was the only failure I was seeing though.

[jeremyd2019]

On my fork, the "cross" test fails now.

I'm an idiot, the build.rs I based on x86_64_gnullvm, and changed arch != "x86_64" to arch != "i686", but looking at i686_gnu it should have been x86. That was the cause of the failure.

[jeremyd2019]

Should i686_gnu's build.rs be modified also, to add || !target.ends_with("-gnu") as is present in x86_64_gnu? I only just noticed that.

[kennykerr]

I think that came from #2774 courtesy of @seritools.

[kennykerr]

Tests pass, so I'm going to merge this PR. If you have a repro do let us know. Thanks for the contribution!

[jeremyd2019]

If you have a repro do let us know.

I now believe that the "cross" test I added failed to fail as I expected because of #2774. Specifically,

windows-rs/crates/targets/i686_gnu/build.rs

Line 5 in a08ab56

if family != "windows" || arch != "x86" || env != "gnu" {

does not explicitly exclude -gnullvm as is done in

windows-rs/crates/targets/x86_64_gnu/build.rs

Lines 5 to 6 in a08ab56

	let target = std::env::var("TARGET").unwrap();
	if family != "windows" || arch != "x86_64" || env != "gnu" || !target.ends_with("-gnu") {

[kennykerr]

Some failure tests would be great, although I'm not sure how exactly it would fail since windows-targets would/should still exclude it?

[jeremyd2019]

I'm talking about the status quo before this PR was merged - the cfg expression in

windows-rs/crates/libs/targets/Cargo.toml

Lines 25 to 26 in 158a528

	[target.'cfg(all(target_arch = "x86", target_env = "gnu", not(windows_raw_dylib)))'.dependencies]
	windows_i686_gnu = { path = "../../targets/i686_gnu", version = "0.52" }

does not check target_abi, so both i686-pc-windows-gnu and -gnullvm would match, pulling in windows_i686_gnu. The build.rs of i686_windows_gnu before #2774 would have returned before adding the lib directory because the target was not exactly i686-pc-windows-gnu or i686-uwp-windows-gnu. After #2774, it would not return early, because the env is gnu for gnullvm, and it does not do the target.ends_with check that would exclude gnullvm.

[kennykerr]

Do you have a repro? I don't see that here:

D:\git\windows-rs>rustup default
nightly-i686-pc-windows-gnu (default)

D:\git\windows-rs>cargo check -p windows-targets
   Compiling windows_i686_gnu v0.52.4 (D:\git\windows-rs\crates\targets\i686_gnu)
    Checking windows-targets v0.52.4 (D:\git\windows-rs\crates\libs\targets)
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.53s

[jeremyd2019]

C:\>curl -LO https://github.com/mstorsjo/llvm-mingw/releases/download/20231128/llvm-mingw-20231128-ucrt-x86_64.zip
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
  0     0    0     0    0     0      0      0 --:--:-- --:--:-- --:--:--     0
100  128M  100  128M    0     0  26.8M      0  0:00:04  0:00:04 --:--:-- 34.9M

C:\>tar -xf llvm-mingw-20231128-ucrt-x86_64.zip

C:\>git clone --depth 1 --single-branch -b 0.52.0 https://github.com/microsoft/windows-rs

C:\windows-rs>path c:\llvm-mingw-20231128-ucrt-x86_64\bin;%PATH%

C:\windows-rs>rustup default
nightly-i686-pc-windows-gnu (default)

C:\windows-rs>rustup component add rust-src
info: downloading component 'rust-src'
info: installing component 'rust-src'
C:\windows-rs>del \llvm-mingw-20231128-ucrt-x86_64\bin\*gcc.exe

C:\windows-rs>del \llvm-mingw-20231128-ucrt-x86_64\bin\*g++.exe

C:\windows-rs>cargo test -p test_win32 --target i686-pc-windows-gnullvm -Zbuild-std

...
  = note: clang-17: warning: argument unused during compilation: '-nolibc' [-Wunused-command-line-argument]
          clang-17: warning: argument unused during compilation: '-no-pie' [-Wunused-command-line-argument]
          lld: error: unable to find library -lwindows.0.52.0
          lld: error: unable to find library -lwindows.0.52.0
          clang-17: error: linker command failed with exit code 1 (use -v to see invocation)


error: could not compile `test_win32` (test "winsock") due to 1 previous error
warning: `test_win32` (test "win32") generated 3 warnings
error: could not compile `test_win32` (test "win32") due to 1 previous error; 3 warnings emitted

C:\windows-rs>cd ..

C:\>rmdir /s windows-rs
windows-rs, Are you sure (Y/N)? y

C:\>git clone --depth 1 --single-branch https://github.com/microsoft/windows-rs

C:\>cd windows-rs

C:\windows-rs>cargo test -p test_win32 --target i686-pc-windows-gnullvm -Zbuild-std

...
     Running unittests src\lib.rs (target\i686-pc-windows-gnullvm\debug\deps\test_win32-a6badaa2d7e7e7e5.exe)

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

     Running tests\hresult.rs (target\i686-pc-windows-gnullvm\debug\deps\hresult-3b4ba2f37ba503d2.exe)

running 1 test
test test_message ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s

     Running tests\win32.rs (target\i686-pc-windows-gnullvm\debug\deps\win32-e90a6c871585c2a0.exe)

running 14 tests
test callback ... ok
test size32 ... ok
test bool_as_error ... ok
test constant ... ok
test dxgi_mode_desc ... ok
test function ... ok
test rect ... ok
test com ... ok
test signed_enum32 ... ok
test unsigned_enum32 ... ok
test empty_struct ... ok
test interface ... ok
test com_inheritance ... ok
test onecore_imports ... ok

test result: ok. 14 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.13s

     Running tests\winsock.rs (target\i686-pc-windows-gnullvm\debug\deps\winsock-e6e3a30783a5f0b9.exe)

running 6 tests
test in6_addr ... ok
test sockaddr_inet6 ... ok
test sockaddr_inet4 ... ok
test sockaddr_in ... ok
test sockaddr_in6 ... ok
test in_addr ... ok

test result: ok. 6 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.01s

Is that what you're looking for? BTW, as expected, 0.53.0 through 0.55.0 succeed due to the build.rs change referenced above. I hadn't noticed that before, I was dealing with breakages of 0.52.0 and 0.48.x.

[jeremyd2019]

Also, once rust-lang/rust#121712 goes somewhere, it should be possible to get more test coverage of gnullvm (without having to use nightly so that -Zbuild-std can be used)

[riverar]

@jeremyd2019 Is there any remaining work to be done here? I can't quite tell by reading the latest messages here.

[jeremyd2019]

I created #2970 for completeness' sake. The only other thing is I want to try to add more testing of the -gnullvm targets.