Use clang on windows?

[yetyetanotherusername]

Can I ask, is using msvc on windows a design choice or is it simply a convention thing?

In my experience, using clang over msvc would offer several benefits:

• Commonality: Linux and macOS both employ clang so it would reduce the number of used compilers to 1
• Compilation time: Clang usually compiles/links significantly faster compared to msvc
• Runtime performance: Clang has been shown to vectorize much more aggressively than msvc yielding better runtime performance of resulting binaries. I.e. large game studios usually use clang over msvc due to this.
• msvc sucks for c++ std compliance

I haven't looked into availability/feasability of using clang on windows in GH Actions, but if you're open to the possibility and there are no obvious blockers that I am missing, I'd be curious to give this a whirl.

[sudara]

Pamp did use clang at one point in time, it was via the same mechanism as linux: egor-tensin/setup-clang@v2

And I do love the idea of reducing compilers I have to support, that was a big benefit when all 3 platforms were on clang. However, people are opening pamplejuce locally inside of Visual Studio, so I also felt like it was smart to run CI on MSVC to some degree, as it will increase compatibility (catches warnings, errors) for that local use case.

I actually made the experience, at least with JUCE, that raw MSVC is fastest to compile. On top of that there's complexities with sccache to consider. That got really nasty on JUCE/MSVC, so I'm no longer sure what's working or not in clang-land. I believe there's some people on the JUCE forum who tried to get sccache working with clang on Windows...

Re: vectorization, I would love see relevant numbers for on-topic JUCE-y runtime things, otherwise my default assumption would be the differences in 2026 are likely not life changing. But since I have a very performance sensitive synth, for example, I'd personally LOVE to have some answers here.

In short: If we can demonstrate runtime benefits and sccache works, for me it's a no-brainer.

Next time I'm benchmarking on Windows I'll check it out my real world apps...