> On May 6, 2014, at 10:51 PM, Duncan Jones <dun...@wortharead.com> wrote: > >> On 7 May 2014 01:52, "Paul Benedict" <pbened...@apache.org> wrote: >> >> When you dereference a null pointer, you get an NPE. We can agree to that. >> We can also agree it's not inherently wrong to throw IAE on a null > argument >> check, but this discussion has never been about that. The discussion has >> always centered on what the trend setters are doing -- and they say go > with >> NPE. >> >> Oracle/Sun throws NPE in its method: > http://docs.oracle.com/javase/7/docs/api/java/util/Objects.html#requireNonNull%28T%29 >> >> Google Guava throws NPE in its method: > http://docs.guava-libraries.googlecode.com/git-history/release/javadoc/com/google/common/base/Preconditions.html#checkNotNull%28T%29 >> >> It's pretty clear where industry is going and not using NPE is not >> expected. We shouldn't try to resist where all the thought leaders in our >> industry are going. It doesn't make any sense. No matter what personal >> affinity/preference you have towards IAE, it's a losing battle because the >> march is going the other direction. > > This is a very compelling point. We don't want to alienate our user base > because a handful of us believe the Java community is going in the wrong > direction. > > I could be swayed by that argument alone. We hardly need more reasons for > people to consider Lang to be outdated compared with Guava etc.
Interesting how I see that argument as exactly the opposite - ie if the only "argument" in support of a technical decision is "that's where things are going" I see that as a good reason not to capitulate. The best thing about OSS (at least how we have traditionally done it here @apache) is we make technical decisions based on technical arguments and the people who actually contribute to the code get listened to. So instead of vague references to "thought leaders" or what others are doing, what matters is what arguments are presented here. In this thread, I have yet to see a good argument for tossing npe when an API advertises parm checking. Do you have one? Does anyone else? Phil > >> >> Paul >> >> >> Cheers, >> Paul >> >> >>> On Tue, May 6, 2014 at 6:27 PM, sebb <seb...@gmail.com> wrote: >>> >>>> On 6 May 2014 22:54, Paul Benedict <pbened...@apache.org> wrote: >>>> This is not a matter of law. If Oracle/Sun set a direction on how to > use >>>> NPE, it's a really good idea to follow them -- even if you don't > agree, >>> do >>>> it for the sake of consistency. I don't think using IAE is somehow >>> "better" >>>> Java than what they are doing. And I give weight to what Joshua said >>>> because he's a former architect of that company. Lang3 was designed to >>>> throw NPE on invalid null arguments because that's what the gurus, > like >>> he, >>>> in our industry who publish "best practices" say it should. If your >>> opinion >>>> bears greater weight than those set forth the best practices, then you >>> win, >>>> but I don't advocate going back to IAE for nulls for the reasons > stated. >>> >>> The problem is still that NPE can be thrown by the JVM for code bugs. >>> If you Google/Bing for "what does NPE mean?" most of the postings say >>> that this is due to a bug in the code that throws it rather than a bug >>> in the code that calls it. >>> >>> There is nothing inherently wrong with using IAE for reporting a null >>> argument. >>> I think it was a mistake to suggest using NPE for that. >>> One might as well throw ArithmeticException for a zero argument that >>> is going to be used as a divisor. >>> Neither is as helpful as IAE. >>> >>> The problem is that NPE is ambiguous. IAE is not. >>> >>>> >>>> >>>> >>>> Cheers, >>>> Paul >>>> >>>> >>>>> On Tue, May 6, 2014 at 4:40 PM, Duncan Jones <dun...@wortharead.com> >>>> wrote: >>>> >>>>>> On 6 May 2014 22:27, "Michael Osipov" <1983-01...@gmx.net> wrote: >>>>>> >>>>>> Am 2014-05-06 15:27, schrieb Benedikt Ritter: >>>>>> >>>>>>> Hi Thiago, >>>>>>> >>>>>>> >>>>>>> 2014-05-06 14:53 GMT+02:00 Thiago Andrade <thia...@gmail.com>: >>>>>>> >>>>>>>> Hello people, >>>>>>>> >>>>>>>> Analizing the JIRA issue >>>>> https://issues.apache.org/jira/browse/LANG-1008the >>>>>>>> contributors noticed that NumberUtils.max/min methods all have > the >>> same >>>>>>>> problem: >>>>>>>> They all throw an IllegalArgumentException when according to the >>>>> official >>>>>>>> documentation (Oracle|Sun) says that a NullPointerException must > be >>>>> thrown >>>>>>>> when an argument must not be null. >>>>>>> >>>>>>> This is not a problem imho. It is a question of API design. I > don't >>> now >>>>> an >>>>>>> offical documentation that say when IAE or NPE _must_ be thrown. >>>>> Sun/Oracle >>>>>>> at some point decided to throw NPE when ever a null reference is >>> passed >>>>> to >>>>>>> a method that doesn't accept null inputs. I don't feel this is > right, >>>>> since >>>>>>> a null input is also an illegal argument. Why make a > differenciation? >>>>> IMHO >>>>>>> NPE should be reserved to the JVM, when a method is called on a > null >>>>>>> reference, but that's just my opinion. >>>>>> >>>>>> >>>>>> It *is* a problem because NullPointerException and >>>>> IllegalArgumentException have concrete semantics layed out in the > JDK's >>>>> Javadocs. If you see how both are used in the JDK, you see that NPE > and >>> IAE >>>>> are used properly and there is no such restriction to the JDK only. > If >>> you >>>>> aread Effective Java, you'll see that you *have to* use NPE if a null >>>>> argument is passed. One might remember the NullArgumentException > back in >>>>> Lang 2, it was removed because it is imperative to use NPE instead. >>>>> >>>>> Effective Java is a great book, but don't confuse Joshua's advice > with >>> law. >>>>> >>>>>> >>>>>> Moreover, the Lang 3 package includes a great class, Validate, > which >>> does >>>>> things right and now I can ask, why the hell is that not used > throughout >>>>> the entire library? >>>>> >>>>> +1 to this. We should update all of lang to use Validate once we've >>> nailed >>>>> this issue. >>>>> >>>>> Duncan >>>>> >>>>>> >>>>>> >>>>>>>> However according to Apache Commons Lang Developer Guide, these >>> methods >>>>> are >>>>>>>> all correct. This guide says that "When throwing an exception to >>>>> indicate a >>>>>>>> bad argument, always try to throw IllegalArgumentException, even > if >>> the >>>>>>>> argument was null. Do not throw NullPointerException.". >>>>>> >>>>>> >>>>>> Correct to the dev guide only -- not Java. >>>>>> >>>>>> >>>>>>> Since [lang] is currently designed this way, I'd rather deal with >>> this >>>>>>> issue for 4.0. We can then revisit our initial decision to only > throw >>>>> IAE >>>>>>> an maybe align it to what the JDK now does. If you want to file an >>>>> issue, >>>>>>> my opinion is, that it should be fix version 4.0. Changing the >>>>> exceptions >>>>>>> that are thrown now may break clients (although I think there are >>> very >>>>> few >>>>>>> use cases where one should catch IAE or NPE). >>>>>> >>>>>> >>>>>> 4.0 has to use Validate throughout the entire package. NPE and IAE >>>>> indicate a programming error in the client not adhering to the > contract >>>>> depicted by the Javadocs, so it is the client's problem to deal with >>> them. >>>>> With proper programming, you should not have to catch those > exception at >>>>> all. >>>>>> >>>>>> >>>>>>>> This mail was sent in order to discuss around and make decisions > to >>>>> solve >>>>>>>> this dilemma where the Java official specification says X and the >>>>> Apache >>>>>>>> official specification says Y. >>>>>>> >>>>>>> Can you please provide a lnk to the official specification you're >>>>> refering >>>>>>> to? ;-) >>>>>> >>>>>> >>>>>> Read Effective Java on exceptions. Thiago provided a URL in the > JIRA >>>>> issue. >>>>>> >>>>>> Further good resources: >>>>>> >>>>>> 1. > http://docs.oracle.com/javase/7/docs/api/java/lang/NullPointerException.html >>>>>> 2. >>> http://docs.oracle.com/javase/tutorial/essential/exceptions/runtime.html > : >>>>> "One case where it is common practice to throw a RuntimeException is >>> when >>>>> the user calls a method incorrectly. For example, a method can check > if >>> one >>>>> of its arguments is incorrectly null. If an argument is null, the > method >>>>> might throw a NullPointerException, which is an unchecked exception." >>>>>> >>>>>> Michael > --------------------------------------------------------------------- >>>>>> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org >>>>>> For additional commands, e-mail: dev-h...@commons.apache.org >>> >>> --------------------------------------------------------------------- >>> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org >>> For additional commands, e-mail: dev-h...@commons.apache.org >>> >>> --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org For additional commands, e-mail: dev-h...@commons.apache.org
