clang-format 14 is extremely slow

[manxorist]

Formatting https://github.com/OpenMPT/openmpt/blob/072d4ccc83d842cfa34601dddf2c69a4cc889c74/src/mpt/arch/x86_amd64.hpp (~50kB) using https://github.com/OpenMPT/openmpt/blob/072d4ccc83d842cfa34601dddf2c69a4cc889c74/src/mpt/.clang-format with clang-format 14 takes about 14 minutes for me (CPU is a 3.4GHz AMD Piledriver).
I am using clang-format version 14.0.5 (as shipped with Visual Studio 2022 17.3.0) or Ubuntu clang-format version 14.0.6-++20220622053131+f28c006a5895-1~exp1~20220622173215.157 (from deb http://apt.llvm.org/focal/ llvm-toolchain-focal-14 main).

I do not see any excessive memory consumption, it's only CPU time.

The performance was fine with clang-format 13.

I could also provide other source files where performance is noticeably slow with clang-format 14 if required, however the given file is the worst offender.

[hansfn]

I can "confirm" this problem. I just upgrade Qt Creator from 6.0.0 to 8.0.1, and suddenly it ate up all my CPUs - 99,7 % load when idle... Disabling the ClangFormat plugin dropped the CPU load to less than 1 % when idle.

Maybe related to https://bugs.llvm.org/show_bug.cgi?id=49519

[llvmbot]

@llvm/issue-subscribers-clang-format

[codido]

@manxorist Could you please run this with QualifierAlignment set to Leave instead of Custom and remove QualifierOrder in your .clang-format?

[manxorist]

manx@quadratus:~/projects/openmpt/trunk-1$ time clang-format-14 -i src/mpt/arch/x86_amd64.hpp
real    0m2.576s
user    0m2.555s
sys     0m0.020s

@codido So yes, changing QualifierAlignment to Leave makes it run in seconds and avoids the performance problem.

I'll probably use that as a work-around for now.

[mydeveloperday]

Qualifier Ordering along with a file with lots of #ifdef will likely be the combined cause

[tru]

@brad0 since you added it to the 15.x milestone and it seems to be pretty high impact issue - I will move this from triage to needs fix. Next rc is released the 23rd of august if you want to get a fix in that one.