andygrove opened a new issue, #286:
URL: https://github.com/apache/datafusion-spark/issues/286

   Part of #271 and apache/datafusion-comet#7300.
   
   ### What the function does
   
   `xpath_long(xml, path)` evaluates an XPath 1.0 expression, converts the 
result with XPath `number()` into a
   Java `Double`, and narrows it with `Double.longValue()`, giving a `BIGINT`. 
Both arguments are strings
   (collation-aware from 4.0), and `path` must be foldable. The result is 
always nullable. Available since Spark
   2.0.0.
   
   The `number()` conversion is in the `xpath_double` issue. Parsing, DTD and 
entity handling, namespaces, NULL
   handling and errors are in the `xpath` issue. This issue covers only the 
double-to-long step, which is Java's
   `(long) d`:
   
   - `NaN` becomes `0`;
   - values are truncated toward zero;
   - out-of-range values saturate at `Long.MAX_VALUE` / `Long.MIN_VALUE`.
   
   This matches Rust's `d as i64`.
   
   ### Examples
   
   ```sql
   SELECT xpath_long('<a><b>1</b><b>2</b></a>', 'sum(a/b)');
   -- 3
   SELECT xpath_long('<a>this is not a number</a>', 'a');
   -- 0
   SELECT xpath_long('<a><b>2000000000</b><c>40000000000</c></a>', 'a/b * a/c');
   -- 9223372036854775807
   SELECT xpath_long('<a>try a boolean</a>', 'a = 10');
   -- 0
   SELECT xpath_long('<a><b class="odd">1</b><b class="even">2</b><b 
class="odd">4</b><c>8</c></a>', 'a/b');
   -- 1
   SELECT xpath_long('<a><b class="odd">1</b><b class="even">2</b><b 
class="odd">4</b><c>8</c></a>', 'sum(a/*)');
   -- 15
   SELECT xpath_long('<a><b class="odd">9000000000</b><b class="even">2</b><b 
class="odd">4</b><c>8</c></a>', 'sum(a/b[@class="odd"])');
   -- 9000000004
   ```
   
   These come from the `ExpressionDescription`, the Hive `udf_xpath_long` 
golden files run by
   `HiveCompatibilitySuite`, and `XPathExpressionSuite`.
   
   ### Edge cases to test
   
   - Precision. The value always goes through a double, so integers above 2^53 
lose precision before the cast.
     `<a>9007199254740993</a>` with `'a'` gives `9007199254740992`, by code 
reading (the nearest double). Spark
     does not read the text directly as a long.
   - Saturation: `8.0E19` gives `9223372036854775807` (above). By the same 
rule, `-1 div 0` gives
     `-9223372036854775808` and `1 div 0` gives `9223372036854775807`.
   - Truncation toward zero (`<a>-2.7</a>` gives `-2`), and `NaN` gives `0`. 
This covers non-numeric text,
     `0 div 0`, and the empty node-set, which is `NaN` in XPath 1.0.
   - `NULL` or `''` arguments give NULL. A non-literal path gives 
`DATATYPE_MISMATCH.NON_FOLDABLE_INPUT`.
     Malformed XML and an invalid path raise runtime errors (see the `xpath` 
issue).
   
   ### Differences across Spark versions
   
   None found beyond the shared ones in the `xpath` issue.
   
   ### Implementation notes
   
   - Spark: `XPathLong` in 
`sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/xml/xpath.scala`.
     It calls `XPathLongEvaluator` (`ret.longValue()`) in 
`.../xml/XmlExpressionEvalUtils.scala`.
   - DataFusion core has no equivalent.
   - Reuse the shared `number()` evaluator from the `xpath` and `xpath_double` 
issues in
     `crates/spark/src/function/xml/`, and narrow with `as i64`.
   - Tests: add `crates/sqllogictest/test_files/xml/xpath_long.slt` and 
document the function with `#[user_doc]`.
   
   ---
   _This issue was generated by an LLM (Claude) from the Apache Spark source 
code and tests. Examples come from Spark's `ExpressionDescription` examples, 
test suites and `.sql.out` golden files, but please verify them against Spark 
before relying on them._


-- 
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