docs: changelog for v2.9.1, and correct the inference docs the fixes invalidated - #166
Conversation
…invalidated
Adds the v2.9.1 changelog section and moves the (Current) marker down to v2.9.0.
Two corrections the merged fixes made necessary:
- parameters.rst claimed ambiguous dates "default to US (month-first)
ordering, consistent with DuckDB conventions". That still holds for a
top-level column, whose format is settled by eliminating candidates across
all of its samples -- but it is no longer true inside a nested STRUCT field
or a repeated LIST element. No per-field format is recorded there, so an
ambiguous value is now declined rather than guessed (#159). Leaving the old
text would have documented a promise to return a wrong date.
- schema_inference.rst listed 1/0 as BOOLEAN detection examples. A column of
0/1 infers INTEGER, because integer detection runs first -- so that row was
already inaccurate before this release, independently of any change here.
Corrected, and the mixed boolean/numeric widening to VARCHAR (#160) is now
stated alongside it.
The changelog names the three behaviour changes first, since a patch version
does not otherwise signal them: SAX fragment output bytes change (#158),
malformed XML fails closed under SAX (#152), and a column mixing boolean and
numeric text now infers VARCHAR instead of failing the scan (#160).
check-docs unchanged at 105 examples, 0 broken (the one fixture case shifts
line number only). CI does not build the docs; RST structure verified by hand.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XoHU19qnngy1CJnXyX1csz
There was a problem hiding this comment.
🟡 Changes recommended
The updated BOOLEAN detection table is internally inconsistent (mentions on/off but omits them from the examples), so the docs should be corrected before merge.
Get a fresh assessment by requesting another Copilot review.
Pull request overview
Documentation-only updates preparing the v2.9.1 release notes and correcting schema/date inference documentation to match current behavior of read_xml/parse_xml inference and extraction.
Changes:
- Added a new v2.9.1 (Current) section to the changelog and moved the “Current” marker off v2.9.0.
- Updated
parameters.rstto clarify how ambiguous date auto-detection behaves at top level vs nested STRUCT/LIST extraction. - Updated
schema_inference.rstto correct boolean/numeric inference examples and to document nested temporal-format limitations and boolean/numeric widening behavior.
File summaries
| File | Description |
|---|---|
| docs/schema_inference.rst | Corrects BOOLEAN examples and adds clarifying text about temporal elimination and boolean/numeric inference outcomes. |
| docs/parameters.rst | Refines the datetime_format='auto' note to distinguish top-level column elimination vs nested extraction ambiguity handling. |
| docs/changelog.rst | Introduces the v2.9.1 changelog section and re-labels v2.9.0 accordingly. |
Review details
- Files reviewed: 3/3 changed files
- Comments generated: 1
- Review effort level: Lite
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
| - true/false, 1/0 | ||
| - ``true``, ``false``, ``1``, ``0`` | ||
| - true/false, yes/no, on/off | ||
| - ``true``, ``false``, ``yes``, ``no`` |
There was a problem hiding this comment.
Fair catch, and it was my edit — fixed in #167, merged as 3a658ea.
The examples column now matches the detection pattern:
* - BOOLEAN
- true/false, yes/no, on/off
- ``true``, ``false``, ``yes``, ``no``, ``on``, ``off``
Worth recording that the inconsistency shipped: this PR was merged into the v2.9.1 release, so the released docs carry the mismatched row and the correction rides in the next release. The fix landed after the v2.9.1 tag (cdd866f) deliberately — a docs-table inconsistency does not justify moving a published tag.
For the record, the other two comments on #165 were handled as well: the CDATA ]]> coverage gap was already fixed in 534127e, and I declined the internal-linkage suggestion with reasoning on the thread — the four sibling SAX callbacks are non-static and declared in xml_sax_reader.hpp:106-115, so marking only the two newest static would have made that set less consistent. #167 instead declares them alongside their siblings, which is what this file's convention calls for.
Pre-release documentation for v2.9.1. No source changes.
Changelog
Adds the
v2.9.1 (Current)section and moves the marker down tov2.9.0. It names the three behaviour changes first, because a patch version does not signal them on its own.Two doc corrections the merged fixes made necessary
parameters.rstclaimed ambiguous dates "default to US (month-first) ordering, consistent with DuckDB conventions". That still holds for a top-level column, whose format is settled by eliminating candidates across all of its samples. It is no longer true inside a nested STRUCT field or a repeated LIST element: no per-field format is recorded there, so after #159 an ambiguous value is declined rather than guessed. Leaving the old text would have documented a promise to return a wrong date.schema_inference.rstlisted1/0as BOOLEAN detection examples. A column of0/1infers INTEGER — integer detection runs first — so that row was already inaccurate before this release, independently of anything changed here. Corrected, and the mixed boolean/numeric widening to VARCHAR (#160) is now stated alongside it.Verification
make check-docs: 105 examples, 0 BROKEN — unchanged from before the edits; the single fixture case shifts line number only (216 → 280), exactly the offset the new changelog section adds..readthedocs.yaml— RTD does), so RST structure was checked by hand: thelist-tablerow keeps its 3 cells, and the.. note::block has uniform 3-space indentation with blank lines between paragraphs.🤖 Generated with Claude Code
https://claude.ai/code/session_01XoHU19qnngy1CJnXyX1csz