viirya opened a new issue, #6839:
URL: https://github.com/apache/datafusion-comet/issues/6839
### Describe the bug
The native string-to-floating-point and string-to-decimal casts accept some
strings that Spark rejects. Spark returns NULL for them (or raises
`CAST_INVALID_INPUT` under ANSI), while Comet returns a value. Found while
reviewing #6826.
**String to FLOAT/DOUBLE.** Spark parses with `Double.parseDouble` /
`Float.parseFloat` and falls back to
`Cast.processFloatingPointSpecialLiterals`. Java only allows a `d`/`D`/`f`/`F`
suffix after a decimal number, and accepts a signed `NaN` only in the exact
case `+NaN` / `-NaN`. Comet's `parse_string_to_float` strips the suffix first
and hands the rest to Rust's parser, which accepts `nan`, `inf` and `infinity`
in any case, so `'NaNF'`, `'NaND'`, `'InfinityD'`, `'infF'`, `'-nan'` and
`'+nan'` all cast to NaN or infinity.
**String to DECIMAL.** Spark parses with `new java.math.BigDecimal(str)`,
which rejects a scale (fractional digits minus exponent) outside the `int`
range. Comet's `parse_decimal_str` / `parse_string_to_decimal` compute the
scale in `i32`, which wraps in release builds (overflow checks are off), so
`'1e-2147483648'`, `'1.0e-2147483647'` and `'0e-2147483648'` cast to `0.00`.
Debug builds panic with `attempt to subtract with overflow` instead.
### Steps to reproduce
```sql
CREATE TABLE t(s string) USING parquet;
INSERT INTO t VALUES ('NaNF'), ('InfinityD'), ('-nan'), ('1e-2147483648');
SELECT s, try_cast(s AS DOUBLE), try_cast(s AS DECIMAL(10,2)) FROM t;
```
### Expected behavior
Spark returns NULL for every cast above. Under ANSI, `CAST(s AS DOUBLE)` and
`CAST(s AS DECIMAL(10,2))` raise `CAST_INVALID_INPUT`.
--
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]
