[experiment] Enable link-time thread local support on MinGW

[joboet]

I read somewhere that GCC supports link-time TLS on Windows. That would simplify the TLS handling code in std by a bit and would also hopefully improve performance.

Uses the CI runners for testing as I don't have a Windows machine.

[rust-highfive]

r? @Mark-Simulacrum

(rust-highfive has picked a reviewer for you, use r? to override)

[rust-highfive]

⚠️ Warning ⚠️

• These commits modify compiler targets. (See the Target Tier Policy.)

[ChrisDenton]

cc @mati865, see also #100917

Note that i686 windows-msvc also does not have thread-local enabled because the last attempt failed (see #95429). This might or might not be related to TLS issues with the Rust compiler.

[mati865]

I've tried togging it in the past but EmuTLS used by GCC is just too messed up and causes problems for all languages that use TLS.
I don't think it's possible to workaround this problem in Rust but FWIW windows-gnullvm targets that I added not so long ago are working perfectly fine with it (although they are not tested on CI): https://github.com/rust-lang/rust/pull/94872/files#diff-5b82e76d5e73e79c281214a612797bd5ffadd853cc83f1c4e37eef19a1bba35dR48 This is because LLVM uses native Windows TLS by default which is not implemented in GCC.

[rust-log-analyzer]

The job i686-mingw-2 failed! Check out the build log: (web) (plain)

Click to see the possible cause of the failure (guessed by this bot)

Updating files:  98% (37295/38056)
Updating files:  99% (37676/38056)
Updating files: 100% (38056/38056)
Updating files: 100% (38056/38056), done.
Note: switching to 'refs/remotes/pull/102243/merge'.
You are in 'detached HEAD' state. You can look around, make experimental
changes and commit them, and you can discard any commits you make in this
state without impacting any branches by switching back to a branch.

---
HEAD is now at bae2e926 Merge 618b0546ec6cc5395d4819374110db672aae52f5 into 6580010551063718462f9dfe41c9490d92994d0e
##[endgroup]
[command]"C:\Program Files\Git\bin\git.exe" log -1 --format='%H'
'bae2e926c18de2e5195281bb538ace567fbed892'
##[group]Run echo "[CI_PR_NUMBER=$num]"
echo "[CI_PR_NUMBER=$num]"
env:
  CI_JOB_NAME: i686-mingw-2
  SCCACHE_BUCKET: rust-lang-ci-sccache2
  TOOLSTATE_REPO: https://github.com/rust-lang-nursery/rust-toolstate
---
   Compiling unic-langid-macros-impl v0.9.0
error: could not compile `unic-langid-macros-impl`

Caused by:
  process didn't exit successfully: `D:\a\rust\rust\build\bootstrap\debug\rustc --crate-name unic_langid_macros_impl --edition=2018 C:\Users\runneradmin\.cargo\registry\src\git.luolix.top-1285ae84e5963aae\unic-langid-macros-impl-0.9.0\src\lib.rs --error-format=json --json=diagnostic-rendered-ansi,artifacts,future-incompat --crate-type proc-macro --emit=dep-info,link -C prefer-dynamic -C embed-bitcode=no -C debuginfo=0 -Zunstable-options --check-cfg values(feature) --check-cfg names() --check-cfg values() -C metadata=99cdadb944cbfccc -C extra-filename=-99cdadb944cbfccc --out-dir D:\a\rust\rust\build\i686-pc-windows-gnu\stage1-rustc\release\deps -L dependency=D:\a\rust\rust\build\i686-pc-windows-gnu\stage1-rustc\release\deps --extern proc_macro_hack=D:\a\rust\rust\build\i686-pc-windows-gnu\stage1-rustc\release\deps\proc_macro_hack-4b1e6edf13150f97.dll --extern quote=D:\a\rust\rust\build\i686-pc-windows-gnu\stage1-rustc\release\deps\libquote-86fa47d5a22c8465.rlib --extern syn=D:\a\rust\rust\build\i686-pc-windows-gnu\stage1-rustc\release\deps\libsyn-513900d7f78bc0a5.rlib --extern unic_langid_impl=D:\a\rust\rust\build\i686-pc-windows-gnu\stage1-rustc\release\deps\libunic_langid_impl-4f3e471d6f782a3b.rlib --extern proc_macro --cap-lints allow -Z binary-dep-depinfo` (exit code: 0xc0000005, STATUS_ACCESS_VIOLATION)
Build completed unsuccessfully in 0:16:55
Build completed unsuccessfully in 0:16:55
make: *** [Makefile:85: ci-mingw-subset-2] Error 1

[mati865]

FYI there are multiple links regarding bugs in GCC's TLS for Windows here: msys2/MINGW-packages#2519

[bjorn3]

Does MinGW use COFF TLS the same way as MSVC? If not, this will break cg_clif for MinGW.

[mati865]

MinGW is only the CRT (a bit like glibc or musl on Linux).
TLS is provided by the compiler, in case of GCC it's emuTLS and for Clang it's native Windows TLS (so COFF TLS I suppose?).

[bjorn3]

Is there any documentation on emuTLS for implementing it in Cranelift?

[mati865]

I have no idea, I'm not familiar with GCC code.

[Mark-Simulacrum]

r? compiler

I don't think I have enough context on this platform right now to properly review this.

[fee1-dead]

r? compiler

[eholk]

Since this is still a draft I'm going to mark it as waiting on the author.

@rustbot author

[bors]

☔ The latest upstream changes (presumably #106087) made this pull request unmergeable. Please resolve the merge conflicts.

[Dylan-DPC]

closing this as it was an experiment