On 11/17/25 08:52, Miklos Vajna wrote:
Yes, I agree, at some point the old clang-format will just misformat newer C++ and we'll have to stop using that old version. Which gives us (I think) at least two options:
We're already past that point, see e.g. the commit message of <https://git.libreoffice.org/core/+/56e6d683dba66d4f2f80145064d2bda2ea4b27b1%5E%21> "double operator < is not a strict weak ordering due to NaN": "And then our venerable clang-format 5.0.0 would have broken the token `<=>` into `<= >`, so exclude include/o3tl/compare.hxx from its mis-treatment."
1) The whole clang-format story started by trying to avoid reformats -- by enforcing one reformat before git commit. So in case a clang-format upgrade would result in a reformat, then clang-format is useless for our needs and this can be solved by not enforcing clang-format anymore.
That's still my preferred way out. In hindsight, I consider our experiment of using clang-format broken-by-design, and the original rationale flawed.
