Use one line between top-level items if formatting a stub file

[tjkuson]

Summary

When formatting stub files, use only one line between top-level items. Use zero lines between class definitions of the same type that have an empty body (nothing but ...). Closes #5821.

This PR doesn't implement moving the ellipsis to the same line as the class definition signature. I have never touched the formatter code before, so was looking around about how to do it and found #5822. Seeing as it has its own issue, I might just leave it out of this PR.

Test Plan

cargo test

[github-actions]

PR Check Results

Benchmark

Linux

group                                      main                                   pr
-----                                      ----                                   --
formatter/large/dataset.py                 1.00      3.9±0.02ms    10.4 MB/sec    1.00      3.9±0.03ms    10.4 MB/sec
formatter/numpy/ctypeslib.py               1.00    780.1±3.25µs    21.3 MB/sec    1.00    780.8±3.16µs    21.3 MB/sec
formatter/numpy/globals.py                 1.00     80.1±0.37µs    36.8 MB/sec    1.01     80.8±0.69µs    36.5 MB/sec
formatter/pydantic/types.py                1.00   1563.9±6.25µs    16.3 MB/sec    1.00   1562.4±3.42µs    16.3 MB/sec
linter/all-rules/large/dataset.py          1.00     10.3±0.02ms     4.0 MB/sec    1.01     10.4±0.01ms     3.9 MB/sec
linter/all-rules/numpy/ctypeslib.py        1.00      2.8±0.01ms     6.0 MB/sec    1.01      2.8±0.00ms     6.0 MB/sec
linter/all-rules/numpy/globals.py          1.00    394.8±0.48µs     7.5 MB/sec    1.00    395.4±0.59µs     7.5 MB/sec
linter/all-rules/pydantic/types.py         1.00      5.3±0.01ms     4.8 MB/sec    1.01      5.4±0.01ms     4.7 MB/sec
linter/default-rules/large/dataset.py      1.00      5.5±0.01ms     7.5 MB/sec    1.01      5.5±0.02ms     7.4 MB/sec
linter/default-rules/numpy/ctypeslib.py    1.00   1206.4±1.11µs    13.8 MB/sec    1.00   1207.8±5.27µs    13.8 MB/sec
linter/default-rules/numpy/globals.py      1.00    140.1±0.25µs    21.1 MB/sec    1.01    141.2±1.11µs    20.9 MB/sec
linter/default-rules/pydantic/types.py     1.00      2.5±0.00ms    10.3 MB/sec    1.00      2.5±0.01ms    10.3 MB/sec

Windows

group                                      main                                   pr
-----                                      ----                                   --
formatter/large/dataset.py                 1.00      4.3±0.06ms     9.5 MB/sec    1.00      4.3±0.06ms     9.5 MB/sec
formatter/numpy/ctypeslib.py               1.00   823.3±11.00µs    20.2 MB/sec    1.00   827.4±10.63µs    20.1 MB/sec
formatter/numpy/globals.py                 1.00     82.5±1.26µs    35.8 MB/sec    1.01     83.5±1.87µs    35.3 MB/sec
formatter/pydantic/types.py                1.00  1695.2±29.73µs    15.0 MB/sec    1.01  1710.9±48.78µs    14.9 MB/sec
linter/all-rules/large/dataset.py          1.00     12.5±0.15ms     3.3 MB/sec    1.02     12.8±0.19ms     3.2 MB/sec
linter/all-rules/numpy/ctypeslib.py        1.00      3.4±0.03ms     4.9 MB/sec    1.02      3.5±0.02ms     4.8 MB/sec
linter/all-rules/numpy/globals.py          1.00    424.6±7.44µs     6.9 MB/sec    1.03    438.7±5.69µs     6.7 MB/sec
linter/all-rules/pydantic/types.py         1.00      6.4±0.06ms     4.0 MB/sec    1.03      6.6±0.09ms     3.8 MB/sec
linter/default-rules/large/dataset.py      1.00      6.8±0.07ms     5.9 MB/sec    1.06      7.2±0.20ms     5.6 MB/sec
linter/default-rules/numpy/ctypeslib.py    1.00  1437.1±14.35µs    11.6 MB/sec    1.04  1487.7±20.53µs    11.2 MB/sec
linter/default-rules/numpy/globals.py      1.00    166.6±5.67µs    17.7 MB/sec    1.04    172.8±3.69µs    17.1 MB/sec
linter/default-rules/pydantic/types.py     1.00      3.1±0.05ms     8.3 MB/sec    1.03      3.1±0.03ms     8.1 MB/sec

[MichaReiser]

This PR doesn't implement moving the ellipsis to the same line as the class definition signature. I have never touched the formatter code before, so was looking around about how to do it and found #5822. Seeing as it has its own issue, I might just leave it out of this PR.

That sounds reasonable. We'll need to change all call sites (because they currently always wrap the body in a block_indent) and/or introduce a FormatBody that we use in all places where a compound statement (not module level) renders a sequence of statements. It can render the ellipsis flat if they have no comments and it is the only statement.

[tjkuson]

Turns out this doesn't deal with leading comments. For example, it formats to

class Load(expr_context):
    ...
# Some comment.
class Other(expr_context):
    ...

I assumed this wouldn't happen because it deals with leading comments in non-stub files. For example,

class Load(expr_context):
    ...


# Some comment.
class Other(expr_context):
    ...

[MichaReiser]

I'll have a closer look at this after merging #6477 and #6452 that both touch the same logic in the suite statement formatting to avoid having to rebase two PRs. I volunteer to rebase your PR.

[MichaReiser]

This change improves the typeshed similarity index from 0.74233 to 0.74261

[MichaReiser]

Thank you! Do you want to look into the body formatting of when it is a single ... next?

[tjkuson]

Thank you! Do you want to look into the body formatting of when it is a single ... next?

I'll take a look at it.