andygrove opened a new pull request, #5062:
URL: https://github.com/apache/datafusion-comet/pull/5062

   ## Which issue does this PR close?
   
   Closes #.
   
   ## Rationale for this change
   
   Several pages in the user guide and contributor guide link to issues that 
have since been closed, and in most of those cases the surrounding prose still 
describes the limitation the issue was filed for. A reader following the link 
lands on a closed issue and has no way to tell whether the documented behavior 
is still current.
   
   A separate problem is that issue and PR references are written four 
different ways across these pages: inline `[#NNNN](url)`, bare URLs, bare 
`#NNNN` with no link at all, and reference-style `[#NNNN]:` definitions. The 
bare forms are not clickable in the rendered docs.
   
   ## What changes are included in this PR?
   
   Every issue and PR reference in `docs/source/user-guide` and 
`docs/source/contributor-guide` was checked against its current state on 
GitHub. Where an issue was closed, the claim in the docs was re-verified 
against the serde and native code rather than taken from the issue state alone.
   
   Thirteen places described a limitation that no longer exists:
   
   | Page | Previous claim | Current behavior |
   | --- | --- | --- |
   | `collection_funcs.md` | `size()` over `MapType` falls back | `CometSize` 
returns `Compatible()` for `ArrayType` and `MapType` |
   | `math_funcs.md` | `CometRemainder` rejects `EvalMode.TRY`, so `try_mod` 
falls back | gates only on the left input type; all three eval modes serialize |
   | `conversion_funcs.md` | `CAST(map AS map)` falls back | runs natively via 
`cast_map_to_map`; support recurses into the key and value casts |
   | `conversion_funcs.md` | `spark.sql.legacy.castComplexTypesToString` is not 
honoured | detected in `CometCast` and reported `Unsupported` |
   | `string_funcs.md` | `replace` marks `Incompatible` on a literal empty 
search string | `Compatible` with a `NativeOptIn`; codegen dispatch by default |
   | `string_funcs.md` | `initcap` is unconditionally `Incompatible` | 
`Compatible` with a `NativeOptIn` |
   | `map_funcs.md` | `str_to_map` ignores 
`spark.sql.legacy.truncateForEmptyRegexSplit` | reads the config and reports 
`Incompatible`, routing to codegen dispatch |
   | `window_funcs.md` | `AVG(decimal)` over a window is dead on Spark 4.x | 
`CometWindowExec.convert` unwraps the `Cast(Divide(...))` shape |
   | `agg_funcs.md` | `Percentile` reports `Incompatible`, native path opt-in | 
plain `Compatible()`, native by default |
   | `pyarrow-udfs.md` | the row to Arrow round-trip is unchanged | 
`CometArrowPythonRunner` consumes `ColumnarBatch` directly |
   | `roadmap.md` | datafusion-spark migration tracked under #2084 | closed, 
superseded by #4150 |
   | `roadmap.md` | awslabs TPC-DS results tracked under #3799 | closed; the 
published results improved, so the section was trimmed |
   | `datatypes.md` | "Interval types fall back to Spark today" | all three 
types moved from planned to supported-with-caveats, with the per-operator gaps 
listed and tracked by #5061 |
   
   Links that pointed at a closed or unrelated tracker were dropped or 
repointed: the ten `tracking #4098` rows in `expressions.md` (that issue 
tracked Spark 4.1 CI failures and never covered `kurtosis`, `skewness`, or the 
Variant functions), `translate` (#4463, where the incompatibility is real but 
the issue is closed, so the reason is now stated inline), the `TimestampNTZ` 
scan note (#4219), the `from_utc_timestamp` timezone-parser note (#2013, whose 
linked issue was "Add from_utc_timestamp support"), and the `make_interval` / 
`try_make_interval` rows (#4540 to #5061).
   
   Seven closed-issue references were kept deliberately because they are 
historical or illustrative: the 0.9.1 jar-size incident in 
`release_process.md`, the hash-aggregate epic in `bug_triage.md` (now 
explicitly labelled as closed), the `to_json` code sample in 
`adding_a_new_expression.md` that mirrors live code in `structs.scala`, the two 
`query ignore(...)` syntax examples in `sql-file-tests.md`, and the two audit 
notes whose workarounds are still present in the code.
   
   For formatting, 52 bare URLs and 27 bare `#NNNN` mentions were converted to 
inline `[#NNNN](url)`. Reference-style definitions were left in place in 
`roadmap.md` and `iceberg-spark-tests.md`, which already used that style 
consistently. Cross-repo references use the `[apache/arrow-rs#10028](url)` 
form. 22 URLs were repointed from `/issues/` to `/pull/` where the number is a 
pull request. URLs inside fenced code blocks were left untouched, including the 
`issues/NNNN` placeholder in `adding_a_new_spark_version.md`.
   
   ## How are these changes tested?
   
   Documentation only, so there is no runtime behavior to test. Every claim 
that changed was checked against the code it describes (`CometSize`, 
`CometRemainder`, `CometCast`, `CometStringReplace`, `CometInitCap`, 
`CometStrToMap`, `CometWindowExec`, `CometPercentile`, 
`CometArrowPythonRunner`). Link formatting was verified with a script that 
finds any GitHub issue or PR reference outside a fenced code block that is not 
an inline link or a reference definition; it reports zero remaining. `prettier 
--check` passes on both guides.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to