Replace cmake-format with gersemi

[Copilot]

Replaces cmake-format (cmakelang, pinned at 0.6.11) with gersemi as the CMake file formatter.

New files

• .gersemirc — formatter config: 2-space indent, 80-char line length, favour-inlining
• cmake/gersemi_definitions.cmake — stub definitions for custom CCF commands (add_ccf_app, add_e2e_test, add_picobench, etc.) and third-party Corrosion commands so gersemi understands their keyword argument structure

Modified scripts

• scripts/check-cmake-format.sh — swaps cmake-format for gersemi --check / gersemi -i, replaces the per-file loop with a single batch invocation, adds set -o pipefail
• scripts/setup-dev.sh — pip install cmakelang → pip install gersemi
• scripts/python-format-checks.sh — removes .cmake-format.py from black's file list

Removed

• .cmake-format.py — 250-line cmake-format config, superseded by .gersemirc

Reformatted

All CMake files under cmake/, samples/, src/, tests/, and root CMakeLists.txt have been reformatted by gersemi. Changes are whitespace/layout only.

One benign warning remains: _message is flagged as unknown because it's CMake's internal save of the original message() builtin when overridden in the root CMakeLists.txt. This is expected and does not affect formatting.

[achamayou]

@copilot is there a way to tweak gersemi settings so as to minimise the reformatting of our cmake?

[Copilot]

@copilot is there a way to tweak gersemi settings so as to minimise the reformatting of our cmake?

I tested all available gersemi settings against the original cmake-format output. The current config (indent: 2, line_length: 80, list_expansion: favour-inlining) already produces the smallest diff:

Configuration	Changed lines	Files
indent=2, len=80, favour-inlining (current)	1120	16
indent=2, len=90, favour-inlining	1157	16
indent=2, len=100, favour-inlining	1254	16
indent=2, len=120, favour-inlining	1295	16
indent=2, len=80, favour-expansion	1799	19
indent=4, len=80, favour-inlining	3034	18

The bulk of the diff (~714 of 1120 lines) is in the root CMakeLists.txt. The changes are structural differences between the two formatters' algorithms — gersemi consistently puts each keyword argument on its own line when wrapping and places the first argument on a new line after the opening parenthesis, while cmake-format tried to pack more onto the same line. These are style differences that no gersemi setting can eliminate.

The definitions and warn_about_unknown_commands settings don't affect the diff size — they only control whether gersemi emits warnings about unrecognised custom commands.

[achamayou]

@copilot how does the run time compare between gersemi and cmake-format on the repo?

[Copilot]

@copilot how does the run time compare between gersemi and cmake-format on the repo?

Benchmarked both on the repo's 22 CMake files (5 runs each, median wall-clock):

Tool	Invocation style	Time
gersemi	single batch (as in new script)	~0.12s
cmake-format	per-file loop (as in old script)	~3.14s
cmake-format	single batch	~1.85s

gersemi is ~15–25× faster, largely because it processes all files in one Python process while the old script spawned a new cmake-format process per file. Even comparing single-batch invocations, gersemi is ~15× faster.

[achamayou]

@copilot how does the run time compare between gersemi and cmake-format on the repo?

Benchmarked both on the repo's 22 CMake files (5 runs each, median wall-clock):

Tool Invocation style Time
gersemi single batch (as in new script) ~0.12s
cmake-format per-file loop (as in old script) ~3.14s
cmake-format single batch ~1.85s
gersemi is ~15–25× faster, largely because it processes all files in one Python process while the old script spawned a new cmake-format process per file. Even comparing single-batch invocations, gersemi is ~15× faster.

That's quite a bit quicker, arguably real time. What do you think @eddyashton?