[
https://issues.apache.org/jira/browse/MATH-1686?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18063493#comment-18063493
]
Gilles Sadowski commented on MATH-1686:
---------------------------------------
Thanks for the report. Please be careful of the formatting rules used by JIRA
(so that the "description" will render correctly).
> `Dfp.equals()` returns `false` for `NaN.equals(NaN)`, violating
> `Object.equals()` reflexivity
> ---------------------------------------------------------------------------------------------
>
> Key: MATH-1686
> URL: https://issues.apache.org/jira/browse/MATH-1686
> Project: Commons Math
> Issue Type: Bug
> Reporter: Shan Jiang
> Priority: Major
>
> # Summary
> `Dfp.equals()` returns `false` when both operands are NaN, violating the
> reflexivity requirement
> of `Object.equals()` (JDK 21 Javadoc):
> {noformat}
> "The `equals` method implements an equivalence relation on non-null object
> references: It is
> *reflexive*: for any non-null reference value `x`, `x.equals(x)` should
> return `true`."
> {noformat}
> # Root cause
> In [`Dfp.java` lines
> 895-907|https://github.com/apache/commons-math/blob/master/commons-math-legacy-core/src/main/java/org/apache/commons/math4/legacy/core/dfp/Dfp.java#L895-L907]:
> {code}
> @Override
> public boolean equals(final Object other) {
> if (other instanceof Dfp) {
> finalDfpx = (Dfp) other;
> if (isNaN() || x.isNaN() || field.getRadixDigits() !=
> x.field.getRadixDigits()) {
> returnfalse; // <-- NaN.equals(NaN) returns false
> }
> returncompare(this, x) == 0;
> }
> returnfalse;
> }
> {code}
> # Reproducer
> {code}
> Dfp nan = new DfpField(20).newDfp().newInstance((byte) 1, Dfp.QNAN);
> System.out.println(nan.equals(nan)); // false (should be true)
> {code}
> # Comparison with JDK
> The JDK's `Double.equals()` and `Float.equals()` deliberately return `true`
> for NaN:
> {noformat}
> `Double.equals()`: "If `d1` represents `+0.0` while `d2` represents `-0.0`,
> or vice versa,
> the `equal` test has the value `false`, even though `+0.0==-0.0` has the
> value `true`. [...]
> If `d1` and `d2` both represent `Double.NaN`, then the `equals` method
> returns `true`, even
> though `Double.NaN==Double.NaN` has the value `false`."
> {noformat}
> The JDK explicitly chose to break IEEE 754 NaN semantics in `equals()` to
> preserve the
> `Object.equals()` contract. `Dfp` should follow the same convention.
> # Impact
> Any `Dfp` NaN value placed in a `HashSet`, `HashMap`, or compared with
> `equals()` will behave
> incorrectly:
> ** `set.add(nan); set.contains(nan)` → may return `false`
> ** `map.put(nan, value); map.get(nan)` → may return `null`
> # How this was found
> Detected by an automated JDK conformance oracle
> (`ObjectOracles.checkEqualsReflexive`).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
