LuciferYang opened a new issue, #860: URL: https://github.com/apache/iceberg-cpp/issues/860
**Summary** `Literal::operator<=>` orders a negative NaN as *greater than* a positive NaN, the opposite of the total ordering the code documents and tests elsewhere. The helper `CompareFloat` (`src/iceberg/expression/literal.cc`) returns `lhs_is_negative <=> rhs_is_negative` for the both-NaN case, so `-NaN <=> +NaN` evaluates to `true <=> false` = `greater`. The adjacent comment states "-NAN < NAN", and `FloatSpecialValuesComparison` / `DoubleSpecialValuesComparison` assert the full ordering `-NaN < -Infinity < ... < +Infinity < +NaN`. The both-NaN sign branch contradicts both. **Root Cause** ```cpp bool lhs_is_negative = std::signbit(lhs); bool rhs_is_negative = std::signbit(rhs); return lhs_is_negative <=> rhs_is_negative; ``` For `lhs = -NaN` (`signbit = true`) and `rhs = +NaN` (`signbit = false`), this is `true <=> false` = `std::strong_ordering::greater`, i.e. it reports `-NaN > +NaN`. A negative sign bit should sort *below* a positive one, so the operands must be swapped: `rhs_is_negative <=> lhs_is_negative`. The existing `FloatNaNComparison` / `DoubleNaNComparison` tests only cover *same-sign* NaN pairs (qNaN vs sNaN, which are equivalent), so the mixed-sign case was never exercised. **Reproduction** ```cpp auto neg_nan = Literal::Float(-std::numeric_limits<float>::quiet_NaN()); auto pos_nan = Literal::Float(std::numeric_limits<float>::quiet_NaN()); // Expected per the documented ordering: less // Actual: greater (neg_nan <=> pos_nan); ``` **Proposed Fix** Swap the operands so a negative sign bit compares as less, consistent with the `-NaN < +NaN` comment and the special-values ordering tests: ```cpp return rhs_is_negative <=> lhs_is_negative; ``` I have the fix and a mixed-sign NaN regression test (float and double) ready and will open a PR. -- 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]
