[ https://issues.apache.org/jira/browse/MATH-259?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12699014#action_12699014
]
Phil Steitz commented on MATH-259:
----------------------------------
I am OK with adding a check and throwing illegalArgumentExeption if an object that does not
implement Comparable is supplied to these methods (as indicated in the javadoc), but not keen
on introducing the compatibility issue.
> Bugs in Frequency API
> ---------------------
>
> Key: MATH-259
> URL: https://issues.apache.org/jira/browse/MATH-259
> Project: Commons Math
> Issue Type: Bug
> Reporter: Sebb
>
> I think the existing Frequency API has some bugs in it.
> The addValue(Object v) method allows one to add a plain Object, but one cannot add anything
further to the instance, as the second add fails with IllegalArgumentException.
> In fact, the problem is with the first call to addValue(Object) which should not allow
a plain Object to be added - it should only allow Comparable objects.
> This could be fixed by checking that the object is Comparable.
> Similar considerations apply to the getCumFreq(Object) and getCumPct(Object) methods
- they will only work with objects that implement Comparable.
> The getCount(Object) and getPct(Object) methods don't fail when given a non-Comparable
object (because the class cast exception is caught), however they just return 0 as if the
object was not present:
> {code}
> final Object OBJ = new Object();
> f.addValue(OBJ); // This ought to fail, but doesn't, causing the unexpected behaviour
below
> System.out.println(f.getCount(OBJ)); // 0
> System.out.println(f.getPct(OBJ)); // 0.0
> {code}
> Rather than adding extra checks for Comparable, it seems to me that the API would be
much improved by using Comparable instead of Object.
> Also, it should make it easier to implement generics.
> However, this would cause compilation failures for some programs that pass Object rather
than Comparable to the class.
> These would need recoding, but I think they would continue to run OK against the new
API.
> It would also affect the run-time behaviour slightly, as the first attempt to add a non-Comparable
object would fail, rather than the second add of a possibly valid object.
> But is that a viable program? It can only add one object, and any attempt to get statistics
will either return 0 or an Exception, and applying the instanceof fix would also cause it
to fail.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
|