* Add new flag for tests, --no-preview-line-length-1, to be used for test cases known to not work in preview mode with line-length=1. Also split out the problematic cases in three cases to separate files. Removed now redundant file which explicitly tested preview annotations with line-length=1
* mode.preview -> preview_mode, mark pep_572_remove_parens as failing with ll1
- Broke tagging images together
- Saved only a few mins
- x86_64 build is fast, time is all spent on cross compile of arm64
- Also remove evil copy pasta ... which is nice
Was worth an attempt.
* [docker ci] Split up amd64 (x86_64) and arm64 builds
- Lets run them seperately to cut down total time
- Will also more clearly show if either arch has specific problems
- Kept amd64 (x86_64) using qemu actions so if GitHub ever offers arm64 boxes it could stay working too
Fixes#3971
* Add CHANGES entry
---------
Co-authored-by: Jelle Zijlstra <jelle.zijlstra@gmail.com>
- Add to run on MacOS + Windows too
- Do not install [d] dependecies as blackd is not actually run / checked
- Move to default GitHub action version - which is 3.12 today
* Make black[d] install + test run with 3.12
- With aiohttp >= 3.9.0 we can now install all dependencies with 3.12
- Add actions to run 3.12
- Lint still needs to be 3.11
Test:
- `python3.12 -m venv /tmp/tb --upgrade-deps`
- `/tmp/tb/bin/pip install tox`
- `/tmp/tb/bin/pip install .[d]`
- `/tmp/tb/bin/tox -e py312`
```
py312: OK (37.61=setup[3.98]+cmd[3.83,0.36,19.54,6.46,3.00,0.44] seconds)
congratulations :) (37.63 seconds)
```
* Move to pypy-3.9
---------
Co-authored-by: Jelle Zijlstra <jelle.zijlstra@gmail.com>
The second `if` cannot be true at its execution point, because it is
already covered by the first `if`. The condition
`comma.parent.type == syms.subscriptlist` always holds if
`closing.parent.type == syms.trailer` holds, because `subscriptlist`
only appears inside `trailer` in the grammar:
```
trailer: '(' [arglist] ')' | '[' subscriptlist ']' | '.' NAME
subscriptlist: (subscript|star_expr) (',' (subscript|star_expr))* [',']
```
Bracket depth is not an accurate indicator of standalone comment position inside more complex blocks because bracket depth can be virtual (in loops' and lambdas' parameter blocks) or from optional parens. Here we try to stop cumulating lines upon standalone comments in complex blocks, and try to make standalone comment processing more simple. The fundamental idea is, that if we have a standalone comment, it needs to go on its own line, so we always have to split.
This is not perfect, but at least a first step.
Python does not consider f-strings to be docstrings, so we probably
shouldn't be formatting them as such
Fixes#4018
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
