[libc-commits] [libc] [reland][libc] Remove unnecessary `FPBits` functions and properties (PR #79128)

[Guillaume Chatelet via libc-commits]

gchatelet wrote:

> > Have a test covering the fact that we do indeed ship the C functions,
> 
> So, some `.c` files under libc/test/ rather than just `.cpp` files? I assume GTEST (or our variant) works with C and not just C++? Was C vs C++ one of the related issues in this case?

Our GTEST uses C++, I don't think we can use it with C. In that case the issue is that our integration tests do refer to the C++ implementation but they should really refer to the C functions. I've started changing them but I'm facing a build problem that I'm trying to address with @lntue in #79319. They should be covered soon.


> > Have most of the build bots as part of the precommit check in github,
> 
> Yeah this is a good idea. The tradeoff between "should this be in presubmit vs post submit" is usually latency. Even if we just did builds in presubmit and then ran the tests in post submit, that might be the right balance. I think the presubmit bots configurations are somewhere in https://github.com/llvm/llvm-zorg; we can play with moving more checks into presubmit.

I agree it's hard to strike a good balance. For small patches it's probably going to be annoying but for large refactoring it definitely helps catch bugs earlier and prevents going through the painful revert / reland cycle. Also it reduces downstream users impact. 

> > Have more build bots covering more architectures (GPU, Fuchsia)
> 
> So we do have the Fuchsia build bot spot issues, though IDK how to find it on lab.llvm.org...is it just https://lab.llvm.org/buildbot/#/builders/98 (@petrhosek )? Maybe we should look into adding a dedicated build of just llvm-libc just for Fuchsia?

To make sure I don't miss a bot I usually use a [filtered view](https://lab.llvm.org/buildbot/#/grid?tag=libc) with the `libc` tag. Can we add the `libc` tag to all build bots that depend on libc?

> > Having faster cross-compilation would definitely help on the latency side of things.
> 
> Yeah, with enable runtimes, I bet I can BUILD arm32 faster on an x86_64 host than on the arm32 build bot itself.
> For running the unit tests, we need to explore qemu. I'd also bet that running the unit tests in an arm32 vm in qemu on an x86_64 host is also faster than running the tests natively on the arm32 build bot.

No doubt about it 😁, that would be fantastic.

---

Last but not least, the `API Test` from [build and test](https://libc.llvm.org/build_and_test.html) page is broken. Once it's fixed we should add it to the build bots.

https://github.com/llvm/llvm-project/pull/79128