Inconsistent equals semantics for BigDecimal between = and .equals

Description

Numbers.equiv for BigDecimal uses compareTo to test equality instead of equals. This means that = and .equals have different results when the scale of the two BigDecimals are different. For example:
=> (= 1.0M 1.00M)
true
=> (.equals 1.0M 1.00M)
false

I see that another JIRA (http://dev.clojure.org/jira/browse/CLJ-1118) changed this behavior and asserts it in unit tests, but this seems like very incorrect behavior.

The motivation for this issue is a unit test using clojure.data/diff to verify that a test message is the same as a message generated with my platform's code. Our downstream customers care about the scale of the decimal, so Clojure's = operator saying two decimals with a different scale are equal caused a difficult-to-detect bug.

For reference, the line causing the issue is here: https://github.com/clojure/clojure/blob/master/src/jvm/clojure/lang/Numbers.java#L964
And a test asserting this behavior is here: https://github.com/clojure/clojure/blob/master/test/clojure/test_clojure/numbers.clj#L75
Javadoc for BigDecimal: https://docs.oracle.com/javase/7/docs/api/java/math/BigDecimal.html#equals(java.lang.Object)

Environment

RHEL5, VirtualBox

Activity

Show:
Andy Fingerhut
June 23, 2015, 5:25 AM

Just to set your expectations, unless the Clojure core team considers a bug or enhancement critical, it is not unusual for a ticket to go for months with no changes or comments.

clojure.core/= and Java .equals are different for numeric arguments in many cases, by design. For example, clojure.core/= is true for numerically equal Byte, Short, Integer, Long, and BigInteger arguments, whereas Java .equals returns false if the types are different. I know this is not the issue you are raising in this ticket – it is just an example of one of many ways in which these things are different from each other.

Have you considered creating a modified version of clojure.data/diff that compares BigDecimal values in the way you prefer?

Alex Miller
June 23, 2015, 5:29 AM

Greg, I haven't managed to get definitive feedback from Rich on this so I have not been able to update it either way. Certainly CLJ-1118 went through his review prior to being committed.

import
June 23, 2015, 6:42 AM

Comment made by: greg.mitchell

Andy - thanks, I didn't realize that. I haven't submitted a Jira to Clojure before. You have a good point, = does more type-coercion work for numerics. I believe that this is qualitatively different because the scale has important informational content in financial and scientific computing. == exists for cases where the type is less relevant, and the linked Jira seems like a reasonable solution for == specifically. As mentioned in my comment, we have tried a couple of ways to re-implement clojure.data/diff, but none of the easy or obvious solutions work. It relies quite deeply on the value-semantics for collections as implemented by = that short-circuits logic to handle BigDecimals specially.

Alex - Understood, thank you for the update.

Andy Fingerhut
June 23, 2015, 7:23 AM

Greg, agreed that there are cases where clojure.core/= and clojure.core/== differ today, e.g. (== 1 1.0) is true but (= 1 1.0) is false.

If you are arguing that Clojure should change so that (= 1.0M 1.00M) is false, but (== 1.0M 1.00M) is still true, that seems reasonable, and perhaps CLJ-1118 went too far by making them not only == but also = (my bad, if so).

Alex Miller
June 23, 2015, 8:19 AM

Rich confirmed that the current behavior is desired and we do not plan to make the suggested change.

Declined

Assignee

Unassigned

Reporter

import

Labels

Approval

None

Patch

None

Affects versions

Priority

Major