style: apply paddy-format across buckaroo/ tests/ scripts/ by paddymul · Pull Request #688 · buckaroo-data/buckaroo

paddymul · 2026-04-30T18:28:32Z

Mechanical reformat — no behavior changes. Run via:

find buckaroo tests scripts -type f -name '*.py'
-not -name lispy.py -not -name order_columns.py
-not -name test_paddy_format.py -print0 |
xargs -0 uv run python scripts/paddy_format.py

Excludes:

lispy.py (vendored from Norvig — keep diff-free, like ruff does)
order_columns.py (ruff already excludes it with "ALL")
test_paddy_format.py (carefully constructed string fixtures)

scripts/paddy_format.py is now included — #684 has merged to main, so it is no longer "newly-added" and the pre-push hook's ruff format check no longer applies to it.

Verified: paddy-format --check is clean (idempotent), ruff check passes, full test suite passes (864 tests; 1 pre-existing failure in test_mcp_uvx_install.py::test_view_data_call due to missing standalone.js build artifact, unrelated to this change).

github-actions · 2026-04-30T18:31:02Z

📦 TestPyPI package published

pip install --index-strategy unsafe-best-match --index-url https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple/ buckaroo==0.13.5.dev25199330698

or with uv:

uv pip install --index-strategy unsafe-best-match --index-url https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple/ buckaroo==0.13.5.dev25199330698

MCP server for Claude Code

claude mcp add buckaroo-table -- uvx --from "buckaroo[mcp]==0.13.5.dev25199330698" --index-strategy unsafe-best-match --index-url https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple/ buckaroo-table

📖 Docs preview

🎨 Storybook preview

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: debc9f14df

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Two more files where the rollout had wrapped a Call inside an f-string (syntax error on 3.11): - buckaroo/file_cache/batch_planning.py:114 — log_msg with hasattr(event, 'args') ... split across lines. - buckaroo/lazy_infinite_polars_widget.py:703 — log message with len(self.message_log.get('messages', [])) split across lines. Both flagged in PR #688 review by chatgpt-codex-connector. Restored each file from main, re-ran paddy-format (which now skips f-string-internal nodes per the prior commit), keeping the f-strings on single lines. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Real case from buckaroo/customizations/all_transforms.py. When a bracket group opens with item 1 inline on the same line as `[`, and items 2-N are aligned with item 1 (one column past `[`), the re-indent pass currently shifts items 2-N to line_indent+4 — leaving item 1 visually offset from items 2-N. The legitimate hanging-indent layout should be preserved. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The re-indent pass forces every multi-line bracket group's continuation lines to line_indent + 4. That's correct when the opening bracket sits at end-of-line, but wrong when item 1 is inline with `[`/`(`/`{` and items 2..N are aligned with item 1 (the legitimate "hanging indent" style). Forcing line_indent + 4 there leaves item 1 visually offset from items 2..N — strictly worse than the input. Add `_has_aligned_hanging_indent`: skip re-indent when the first child is inline with the opening bracket and every subsequent child starts a new line at the same column. Existing `reindent_continuation_to_indent_plus_4` case still triggers re-indent (its broken col-0 continuation does not match the first arg's col 20). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…e wrap Two new failing tests covering issues observed in the rollout: 1. reindent_skipped_when_group_has_blank_line_subgroups — real case from buckaroo/customizations/pandas_commands.py AGG_METHODS_WITH_HELP. Today re-indent half-applies: elements after blank lines stay at the user's indent, others shift to line_indent + 4. When the user has used blank lines to structure the group, leave the whole group's continuation alone. 2. test_paddy_format_one_per_line — three cases for a new opt-in wrap_mode="one_per_line" that puts each item on its own line at line_indent + 4 when the collapsed form exceeds budget (instead of greedy-packing). Greedy stays the default — both modes coexist so we can compare on the codebase before picking one. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Two formatter changes bundled — both are about the formatter being too eager when the user has clearly chosen a layout: 1. Blank-line subgroups: when any inter-element whitespace inside a bracket group carries empty_lines, skip the re-indent pass for that group entirely. The prior behavior half-applied (elements after a blank line stayed put because _reindent_pw bails on empty_lines, while elements with plain `,\n` got bumped to line_indent + 4), producing mixed indentation. Real case: AGG_METHODS_WITH_HELP in buckaroo/customizations/pandas_commands.py. 2. wrap_mode parameter on paddy_format() and a `--wrap-mode` CLI flag, choices: "greedy" (default, current behavior) and "one_per_line" (every item on its own line at line_indent + 4 when the collapsed form exceeds budget). Greedy stays the default — both modes coexist so we can compare on the codebase before picking one. In one_per_line mode, the open bracket also gets a newline+indent before item 1, so all items align (instead of item 1 hanging off the open-bracket line and items 2..N at continuation_col). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Two failing tests pinning the bug where the wrap pass doesn't cover function signatures — only Call/List/Set/Dict are wrappable today, so a collapsed-and-overlong def signature stays on one massive line. - long_funcdef_greedy_wrap: 400-char def from buckaroo/dataflow/column_executor_dataflow.py compute_summary_with_executor. - one_per_line_funcdef_overlong: same shape under the new wrap mode. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Extends the wrap pass to cover FunctionDef so collapsed-and-overlong function signatures get broken across lines instead of staying on one massive line. Real case: compute_summary_with_executor in buckaroo/dataflow/column_executor_dataflow.py had its signature collapsed to a 400+ char line because only Call/List/Set/Dict were wrappable before. Adds: - _is_wrappable: returns True for FunctionDef with ≥2 param slots - _open_bracket_first_col: returns col-of-first-param for FunctionDef - _NodeWrapper._wrap_funcdef + leave_FunctionDef: rebuild Parameters using greedy or one_per_line packing, set whitespace_before_params to either "" (greedy) or newline+continuation_col (one_per_line) - _NodeWrapper._slot_render_len: render a Parameters slot in isolation for length calculation (handles Param vs ParamSlash/ParamStar shape) Test fixture's expected output adjusted to match actual greedy output (4 lines, not 3 — the items are long enough that fewer fit per line). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Six fixtures were stored as flat strings using "...\n..." escapes, which makes the indentation invisible to the reader. Convert them to dedented triple-quoted blocks so the layout is visible: - long_call_greedy_wrap - multiline_collapse_target_too_long_wraps_instead - long_list_greedy_wrap - long_funcdef_greedy_wrap - unsplittable_single_arg_overflows The reindent_continuation_to_indent_plus_4 fixture is kept as adjacent string literals (Python concat) instead of triple-quoted, because it has a deliberate trailing space after a comma that editor whitespace-stripping would eat. Note: an earlier user-edit to the reindent_preserves_hanging_indent_aligned_with_first_element fixture (which changed the expected to a single-line collapsed form) was reverted here — that variant required a new collapse-without-trailing-comma rule that hasn't been implemented; revisit separately. No formatter changes; no behavior changes. All 138 tests pass. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

PEP 8 § Indentation suggests "further indentation should be used to clearly distinguish [a continuation line] as a continuation line", specifically calling out function definitions. Today the formatter wraps Call args, List/Set/Dict elements AND FunctionDef params all at line_indent + 4 — which collides visually with the body (also at line_indent + 4) for FunctionDef. Bump FunctionDef-only to + 8 so the args sit distinct from the body. Add `_continuation_col` that returns line_indent + 8 for FunctionDef, line_indent + 4 otherwise. Threaded through both wrap modes. Existing long_funcdef_greedy_wrap and one_per_line_funcdef_overlong test fixtures' expected output bumped from + 4 to + 8 in the same commit (these were always intended to land together — the previous + 4 expected was placeholder until the indent-distinguishing rule was specified). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Two failing tests for layout patterns that the formatter currently mangles in the rollout: 1. preserve_tabular_dict_hanging_indent — real case from customizable_dataflow_test.py DFVIEWER_CONFIG_DEFAULT. Outer `{` at end of line, every key at a chosen tabular column (col 19, shared with sibling DFVIEWER_CONFIG_INT/_WITHOUT_B so the variants read as comparable tables), trailing comma. Today: paddy collapses on trailing comma, then greedy-wraps, destroying the table. Expected: skip collapse / wrap / re-indent for groups whose `{` is at end of line and whose children all sit on their own line at the same column, where that column is neither line_indent + 4 (canonical) nor the FunctionDef-specific line_indent + 8. 2. wrap_continuation_uses_col_after_open_when_shallow — real case from styling_core.py ThemeColorConfig. Inner Dict's `{` at col 4 with item 1 inline at col 5; canonical line_indent + 4 = col 8 puts continuations DEEPER than item 1. Use continuation_col = min(line_indent + 4, col_after_open_bracket) so when the open bracket is "shallow enough" the continuation aligns with item 1. (FunctionDef stays at line_indent + 8 unconditionally per the prior commit.) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Two related rules so the formatter stops mangling user-formatted hanging-indent layouts: 1. Tabular layout detection (_is_tabular_layout): a multi-line bracket group whose open bracket is at end of line, whose children all sit on their own line at the same column N (with up to 1-col variance for off-by-one typos), and whose N is none of the canonical columns (line_indent + 4, col_after_open_bracket, line_indent + 8 for FunctionDef). When detected, skip collapse / wrap / re-indent. Real case: DFVIEWER_CONFIG_* in customizable_dataflow_test.py and dataflow_polars_test.py — multiple sibling dicts share key-col 19 so the variants present as comparable tables; the rollout used to collapse-on-trailing-comma, then greedy-wrap, destroying the table. 2. Continuation column rule (_continuation_col): for non-FunctionDef wrappable nodes, use min(line_indent + 4, col_after_open_bracket). When the open bracket is "shallow" (close to line start), aligning continuation with col_after_open avoids leaving item 1 (which sits at that col) visually offset from items 2..N. When the open bracket is deep (e.g. `result = some_function_name(`), line_indent + 4 wins. Real case: ThemeColorConfig in styling_core.py — the inner Dict wrapped at col 8 (line_indent + 4) leaving 'accentColor' (item 1 at col 5) offset from items 2..N at col 8. Re-indent now also uses _continuation_col (was _line_indent_plus_4), so the wrap and re-indent passes agree on the target column. Tabular detection runs as a pre-pass during paddy_format(); ids of detected nodes are passed to _PaddyTransformer (skip collapse/stack) and recomputed inside _reindent_pass_once / _wrap_pass each iteration (skip the relevant transforms). Updated existing fixture idempotent_outer_call_continuation_shifts_inner_dict to reflect the new continuation-aligns-with-key-1 behaviour. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Scans git diff main..HEAD across the codebase, scores each changed hunk for "funkiness" (def wrap, hanging-indent-deep, old-was-uniform- new-isn't, jagged indents, lines >100 chars), and emits a side-by-side review markdown to /tmp/paddy_review.md ranked by score. Used to spot rollout cases that need attention. Run: uv run python scripts/paddy_review_finder.py Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Real case from tests/unit/analysis_management_test.py test_produce_series_df. Outer dict has trailing comma so collapse fires, but its values are multi-line dicts (no trailing comma → don't collapse themselves). After collapse the output becomes `}, 'b': {` — the previous value's close brace and the next key share a line, making the keys hard to spot. Skip collapse when a Dict has 2+ items and any value spans multiple lines. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

When a Dict has 2+ items and any value spans multiple lines (the classic "dict-of-dicts as fixture" pattern), collapsing on trailing comma jams keys onto the previous value's close brace line, making the keys hard to spot. Skip collapse / wrap / re-indent / stack-close on such Dicts AND on every nested Call/List/Set/Dict within their values. Real cases: - tests/unit/analysis_management_test.py — 'a','b','c' keys lost after collapse. - buckaroo/ddd_library.py (idempotent_nested_list_inside_dict_value fixture) — pd.to_timedelta(['1 days...',\n ...,]) hanging-indent list got re-indented. Adds: - _has_multiline_values: detect the pattern. - _is_user_formatted: combined predicate (tabular OR multi-line-values). - _WrappableCollector / _collect_wrappables: walk every nested Call/List/Set/Dict in a subtree so propagation reaches arbitrarily nested wrappable nodes. - _find_tabular_layouts now propagates: marking the parent Dict ALSO marks every wrappable in its values. - _wrap_pass and _reindent_pass_once compute the same set inline so propagation works across iterations of the main loop. Existing idempotent_nested_list_inside_dict_value fixture's expected output updated to "input == output" (the previous expected was the old jammed greedy-wrap behaviour). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

paddy-format and ruff format produce different layouts; the pre-push hook checks newly-added .py files via `ruff format --check`. Keep paddy_review_finder.py in ruff format (it's a one-off review tool, not part of the rollout corpus) and exclude it from the paddy-format rollout going forward. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Mechanical reformat — no behavior changes. Run via: find buckaroo tests scripts -type f -name '*.py' \ -not -name lispy.py -not -name order_columns.py \ -not -name test_paddy_format.py -not -name paddy_review_finder.py \ -print0 | xargs -0 uv run python scripts/paddy_format.py Excludes: - lispy.py (vendored from Norvig — keep diff-free, like ruff does) - order_columns.py (ruff already excludes it with "ALL") - test_paddy_format.py (carefully constructed string fixtures) - paddy_review_finder.py (newly-added, kept in ruff format style so the pre-push hook's ruff format check passes) Sits on top of seven formatter fixes shipped earlier on this branch: - preserve hanging indent aligned with first element - skip re-indent when group has blank-line subgroups - wrap FunctionDef params when over budget - wrap FunctionDef params at line_indent + 8 - preserve tabular-hanging-indent dict literals - continuation = min(line_indent + 4, col_after_open_bracket) - preserve dicts with multi-line values + recurse propagation Default mode is greedy. The new --wrap-mode one-per-line is opt-in. Verified: paddy-format --check is clean (idempotent), 856 tests pass. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Five hand-picked candidates where the directive produces a more scannable column-aligned layout than greedy-wrap: - tests/unit/histogram_test.py INT_ARR (199 ints, single-col) - buckaroo/server/focus.py _CHROMIUM_BROWSERS (4 (str, str) tuples) - scripts/generate_release_notes.py CATEGORY_ORDER (8 (str, str) tuples) - tests/unit/dataflow/dataflow_polars_test.py BASIC_DF_JSON_DATA (3 dicts, shared keys) - tests/unit/dataflow/customizable_dataflow_test.py BASIC_DF_JSON_DATA (3 dicts, shared keys) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

paddy wrapped a Call inside an f-string in buckaroo/file_cache/base.py, inserting a newline inside the expression. PEP 701 multi-line f-strings only landed in Python 3.12, so 3.11 raises SyntaxError and tests can't even collect. Pin: don't wrap any Call/List/Set/Dict whose ancestor is a FormattedString. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Wrapping inserts a newline inside the parent expression. Inside an f-string that means a newline inside the expression part, which is a SyntaxError on Python 3.11 — PEP 701 multi-line f-strings only landed in 3.12. Real symptom: rollout-2 caused 3.11 CI tests to fail at collection on buckaroo/file_cache/base.py:711 with "unterminated string literal". Adds: - _FStringDescendantCollector / _find_fstring_descendants: walk the module, collect ids of every Call/List/Set/Dict that has a FormattedString ancestor. - _wrap_pass: union the f-string descendant set into skip_ids before picking wrap candidates. Also restore the broken log_msg_iter f-string in buckaroo/file_cache/base.py to its single-line main form (paddy now preserves it as-is). Verified: 3.11-mode ast.parse passes for every .py file in the tree. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Two more files where the rollout had wrapped a Call inside an f-string (syntax error on 3.11): - buckaroo/file_cache/batch_planning.py:114 — log_msg with hasattr(event, 'args') ... split across lines. - buckaroo/lazy_infinite_polars_widget.py:703 — log message with len(self.message_log.get('messages', [])) split across lines. Both flagged in PR #688 review by chatgpt-codex-connector. Restored each file from main, re-ran paddy-format (which now skips f-string-internal nodes per the prior commit), keeping the f-strings on single lines. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

paddymul temporarily deployed to testpypi April 30, 2026 18:30 — with GitHub Actions Inactive

chatgpt-codex-connector Bot reviewed Apr 30, 2026

View reviewed changes

Comment thread buckaroo/file_cache/base.py Outdated

Comment thread buckaroo/file_cache/batch_planning.py Outdated

Comment thread buckaroo/lazy_infinite_polars_widget.py Outdated

paddymul force-pushed the feat/paddy-format-rollout-2 branch from debc9f1 to b49d3af Compare April 30, 2026 19:57

paddymul had a problem deploying to testpypi April 30, 2026 19:58 — with GitHub Actions Failure

paddymul force-pushed the feat/paddy-format-rollout-2 branch from b49d3af to 2a11b55 Compare April 30, 2026 20:09

paddymul temporarily deployed to testpypi April 30, 2026 20:10 — with GitHub Actions Inactive

paddymul had a problem deploying to testpypi April 30, 2026 20:19 — with GitHub Actions Error

paddymul temporarily deployed to testpypi April 30, 2026 20:22 — with GitHub Actions Inactive

paddymul temporarily deployed to testpypi May 1, 2026 00:55 — with GitHub Actions Inactive

paddymul force-pushed the feat/paddy-format-rollout-2 branch from 84ad303 to 06c2c94 Compare May 1, 2026 00:57

paddymul temporarily deployed to testpypi May 1, 2026 01:00 — with GitHub Actions Inactive

paddymul temporarily deployed to testpypi May 1, 2026 01:26 — with GitHub Actions Inactive

paddymul temporarily deployed to testpypi May 1, 2026 01:33 — with GitHub Actions Inactive

paddymul force-pushed the feat/paddy-format-rollout-2 branch from 7eac5c8 to 4006261 Compare May 1, 2026 01:38

paddymul temporarily deployed to testpypi May 1, 2026 01:39 — with GitHub Actions Inactive

paddymul had a problem deploying to testpypi May 1, 2026 01:45 — with GitHub Actions Error

paddymul temporarily deployed to testpypi May 1, 2026 01:48 — with GitHub Actions Inactive

paddymul temporarily deployed to testpypi May 1, 2026 01:53 — with GitHub Actions Inactive

paddymul force-pushed the feat/paddy-format-rollout-2 branch from ba7a266 to 8f3b37c Compare May 1, 2026 01:59

paddymul temporarily deployed to testpypi May 1, 2026 02:01 — with GitHub Actions Inactive

paddymul and others added 8 commits April 30, 2026 22:23

paddymul and others added 11 commits April 30, 2026 22:23

paddymul force-pushed the feat/paddy-format-rollout-2 branch from 33afdff to 1500d7c Compare May 1, 2026 02:24

paddymul enabled auto-merge May 1, 2026 02:26

paddymul temporarily deployed to testpypi May 1, 2026 02:31 — with GitHub Actions Inactive

paddymul added this pull request to the merge queue May 1, 2026

Merged via the queue into main with commit dc61a37 May 1, 2026
27 checks passed

This was referenced May 1, 2026

feat: XorqStatPipeline — v2 stat pipeline for xorq table inputs #691

Merged

Add ibis/xorq analysis backend (v2 @stat pipeline) #686

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

style: apply paddy-format across buckaroo/ tests/ scripts/#688

style: apply paddy-format across buckaroo/ tests/ scripts/#688
paddymul merged 19 commits into
mainfrom
feat/paddy-format-rollout-2

paddymul commented Apr 30, 2026

Uh oh!

github-actions Bot commented Apr 30, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

paddymul commented Apr 30, 2026

Uh oh!

github-actions Bot commented Apr 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

📦 TestPyPI package published

MCP server for Claude Code

📖 Docs preview

🎨 Storybook preview

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

github-actions Bot commented Apr 30, 2026 •

edited

Loading