See my comments in your mail body. ----- Original Message ----- We should strive to define our classes well. Actually that point refers to the design stage rather than to the implementation. On other hand, if we put the objects (that include all members in their equals) in a HashSet in order to keep unique instances, we might have a bug there. Those classes are used all over engine code and even in the UI code, so changing the implementation in the existing code is very risky. Having DAO covered isn't enough IMHO. As you probably know hashCode implementation could not be (conceptually) different from the equals one.

See my comments in your mail body. ----- Original Message ----- Gilad, thank you for opinion even though it's not very relevant to the thread.

I'm also new to oVirt, so I can just add mine opinion to strengthen what you've said. What you've described is a recommended way how to form "identity" in hibernate, that's true. Use 'candidate key' instead of just comparing upon PK — some business/natural key of that entity, not to be dependent on entity persistent state. The only problem is, and I encounter it countless times, that there can be a lot of situations, where I cannot find such natural key, or I don't want to persist properties of real-world object, forming that natural key. Lets say I want to persist comments to an article, then natural key could be precise timestamp & identification of author. In app without authentization I don't want to store such data along with comments. (If you know solution to this, let me know.) In this case, most common thing I saw is to just compare ID & do not compare unpersisted entities, or comparing every property, which is failproof I think. To your question: ad 1. I totally agree with your approach, where applicable. ad 2. I don't see any benefit in sticking to "everywhere in same way" approach, it goes against evolution, and just brings stability in suboptimal position. Almost every convention (except for code-style) is wrong. Every convention requiring something in uniform, duplicate manner lets say "name all queries with suffix '_query'" should be expressed programmatically to avoid errors. In our case if we want 'all members equals/hashcode for business objects' then parent for all business objects with reflection final equals/hashcode should be introduced to eliminate possible errors. (note: this is rather problematic with gwt, but if I checked correctly, maybe doable). ad 3. I cannot see any benefit in batch changing working code, which does not influence performance is not required. If product owner is ok with this to change, is ok to change it (after lot of tests), when some business objects is 'touched' by some of your changes. Even if you potentially break it(and you can never completely avoid it), it's just one thing that got broken. m. ----- Original Message ----- From: "Yair Zaslavsky" <yzaslavs(a)redhat.com&gt; To: "Yevgeny Zaspitsky" <yzaspits(a)redhat.com&gt; Cc: devel(a)ovirt.org Sent: Sunday, May 25, 2014 3:09:11 PM Subject: Re: [ovirt-devel] Implementing equals & hashCode methods ----- Original Message ----- I assume your idea came (at least from ) the way identify is defined at Hibernate. The advantages are obvious - a. shorter code b. let's time to compute However, I fear this is may lead to some bugs if identities are not defined well. Regarding 3 - might be dangerous as we don't have enough unit test coverage, probably. Actually, IMHO the situation with DAO tests which uses equals is quite good. For hashCode - I assume you have a point _______________________________________________ Devel mailing list Devel(a)ovirt.org http://lists.ovirt.org/mailman/listinfo/devel