Martin Storsjö

> > Back on topic, since the config triple hack was rejected, I'm not sure how hard it would be to fix this on Rust or LLVM side. While the...

> > > 14 % for that sounds like a lot - that's not what I'm seeing, so I would think there's another factor playing in here too. > >...

> I modified this workflow to add a fuchsia-clang benchmark, and got similar results to "Clang hosted, LTO, nodylib". But my local machine is indeed very close to 14% (fuchsia...

> Forgot to mention that I used the `--host-clang` option. Thanks, that's a pretty vital clue here. I don't have any installs of openSUSE readily available for testing and looking...

There's a combination of two issues here. With the mingw headers, `off_t` defaults to a 32 bit size, unless building with `-D_FILE_OFFSET_BITS=64`. Secondly, the code that uses `off_t` and `ftello`...

I remembered there's another thing that we could do on the libcxx side, though - we can configure libcxx with `-DLIBCXX_EXTRA_SITE_DEFINES="_FILE_OFFSET_BITS=64"`, which adds that define into libcxx's `__config_site` which all...

> `time_t` is 64bit in MSVC now but `off_t` is still 32bit, so it doesn't seem too wrong what we have now when trying to reduce the difference there. I'd...

> > That being said, whether libc++ defines `_FILE_OFFSET_BITS=64` is questionable; but despite that, it really should call `ftello64()` and `feeko64()` explicitly. > > That's indeed right; if libcxx wants...

> Hello I use clang 19.1.7 from MSYS2 clang64 Technically, this repo is for the llvm-mingw packaging, and for upstream Clang issues the relevant place is https://github.com/llvm/llvm-project/issues - but I...

Thanks; I've reproduced the issue. I've further reduced the issue down to this: ```c++ struct tlsdata { int val; ~tlsdata() {} }; inline thread_local struct tlsdata tlsvar; int main(int argc,...