Update to Clang 0.19

[benegee]

Also removed Glob and switched to using detect_headers as suggested here: https://github.com/JuliaInterop/Clang.jl?tab=readme-ov-file#quick-start

Overall this causes lots of changes in the generated Libt8.jl. The diff is far too huge. Line count went from ~17k to ~21k.

Fixes I had to add:

• manually added const P4EST_QMAXLEVEL = 29, which is defined in some p4est header
• ignored t8code t8_dtri_to_dtet.h header, which, for reasons I do not know precisely, redefines a lot of things, resulting in broken bindings

[benegee]

In principle this should be non-breaking as the underlying t8code_jll, used to generate the bindings, has not changed.
But, if we release this as v0.9.3, Trixi.jl will pick it up immediately, right? @JoshuaLampert

[JoshuaLampert]

But, if we release this as v0.9.3, Trixi.jl will pick it up immediately, right? @JoshuaLampert

Yes.

[JoshuaLampert]

Overall this causes lots of changes in the generated Libt8.jl. The diff is far too huge. Line count went from ~17k to ~21k.

Do you know why?

[benegee]

Unfortunately not!

Clang.jl v0.18.3 dates back to April, 2024. I can only speculate lots of things have changed since then under the hood, and somehow more code is now processed and transformed.

[benegee]

Should we consider a breaking release, just as a safety measure?

[JoshuaLampert]

I don't know. Can you check if there is a difference in the generated file between Clang.jl v0.19.0 and v0.19.3 or if the changes only come from v0.18.3->v0.19.0? Then we could try to understand, which of these PRs is responsible for the changes https://github.com/JuliaInterop/Clang.jl/releases/tag/v0.19.0.

[benegee]

Actually, I checked this yesterday. Same situation with v0.19.0, so it comes with the major release.

[JoshuaLampert]

Ok, can you recognize any specific pattern in the diff when looking at it locally (it is too big to be shown in the PR on GitHub)?