[PROPOSAL] So what do we do about HashCodeBuilder parameters ?

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

[PROPOSAL] So what do we do about HashCodeBuilder parameters ?

Thomas Mortagne
Administrator
Hi devs,

Most Object#hashCode() are implemented using HashCodeBuilder.
HashCodeBuilder provide an empty constructor and also allow setting
"random" numbers.

http://dev.xwiki.org/xwiki/bin/view/Community/JavaCodeStyle#HEquals2FHashCodeandToStringimplementations
does not really say anything about what do do and when currently so I
would like to have a clear rule about it.

On my side when I don't plan to use different types of objects in a
hash key I usually simply use the empty HashCodeBuilder constructor.
The only point of setting different numbers is when you might end up
comparing different object types (and even in that case you have the
#equals() safeguard anyway so it's mostly about performances in
practice).

So here is a proposal: it's OK to use empty HashCodeBuilder constructor except:
* when overwriting the hachCode of a super class you should use
different numbers than the super class (since there is big chance to
have that super class used as a hash key)
* as a best effect in general if you know that you are going to have
several different types as hash key, try to make sure they use
different random numbers, your proc will thank you :)

WDYT ?

--
Thomas Mortagne
Reply | Threaded
Open this post in threaded view
|

Re: [PROPOSAL] So what do we do about HashCodeBuilder parameters ?

vmassol
Administrator
Hi,

> On 8 Dec 2017, at 11:12, Thomas Mortagne <[hidden email]> wrote:
>
> Hi devs,
>
> Most Object#hashCode() are implemented using HashCodeBuilder.
> HashCodeBuilder provide an empty constructor and also allow setting
> "random" numbers.
>
> http://dev.xwiki.org/xwiki/bin/view/Community/JavaCodeStyle#HEquals2FHashCodeandToStringimplementations
> does not really say anything about what do do and when currently so I
> would like to have a clear rule about it.
>
> On my side when I don't plan to use different types of objects in a
> hash key I usually simply use the empty HashCodeBuilder constructor.
> The only point of setting different numbers is when you might end up
> comparing different object types (and even in that case you have the
> #equals() safeguard anyway so it's mostly about performances in
> practice).
>
> So here is a proposal: it's OK to use empty HashCodeBuilder constructor except:
> * when overwriting the hachCode of a super class you should use
> different numbers than the super class (since there is big chance to
> have that super class used as a hash key)
> * as a best effect in general if you know that you are going to have
> several different types as hash key, try to make sure they use
> different random numbers, your proc will thank you :)
>
> WDYT ?

I don’t fully master the topic but I’m trusting you so +0 from me. Only comment is that it means there’s not a single way of writing our code and thus is not that easy to check.

Thanks
-Vincent


>
> --
> Thomas Mortagne

Reply | Threaded
Open this post in threaded view
|

Re: [PROPOSAL] So what do we do about HashCodeBuilder parameters ?

Marius Dumitru Florea
In reply to this post by Thomas Mortagne
+0

Thanks,
Marius

On Fri, Dec 8, 2017 at 12:12 PM, Thomas Mortagne <[hidden email]>
wrote:

> Hi devs,
>
> Most Object#hashCode() are implemented using HashCodeBuilder.
> HashCodeBuilder provide an empty constructor and also allow setting
> "random" numbers.
>
> http://dev.xwiki.org/xwiki/bin/view/Community/JavaCodeStyle#
> HEquals2FHashCodeandToStringimplementations
> does not really say anything about what do do and when currently so I
> would like to have a clear rule about it.
>
> On my side when I don't plan to use different types of objects in a
> hash key I usually simply use the empty HashCodeBuilder constructor.
> The only point of setting different numbers is when you might end up
> comparing different object types (and even in that case you have the
> #equals() safeguard anyway so it's mostly about performances in
> practice).
>
> So here is a proposal: it's OK to use empty HashCodeBuilder constructor
> except:
> * when overwriting the hachCode of a super class you should use
> different numbers than the super class (since there is big chance to
> have that super class used as a hash key)
> * as a best effect in general if you know that you are going to have
> several different types as hash key, try to make sure they use
> different random numbers, your proc will thank you :)
>
> WDYT ?
>
> --
> Thomas Mortagne
>
Reply | Threaded
Open this post in threaded view
|

Re: [PROPOSAL] So what do we do about HashCodeBuilder parameters ?

Thomas Mortagne
Administrator
In reply to this post by vmassol
On Fri, Dec 8, 2017 at 11:33 AM, Vincent Massol <[hidden email]> wrote:

> Hi,
>
>> On 8 Dec 2017, at 11:12, Thomas Mortagne <[hidden email]> wrote:
>>
>> Hi devs,
>>
>> Most Object#hashCode() are implemented using HashCodeBuilder.
>> HashCodeBuilder provide an empty constructor and also allow setting
>> "random" numbers.
>>
>> http://dev.xwiki.org/xwiki/bin/view/Community/JavaCodeStyle#HEquals2FHashCodeandToStringimplementations
>> does not really say anything about what do do and when currently so I
>> would like to have a clear rule about it.
>>
>> On my side when I don't plan to use different types of objects in a
>> hash key I usually simply use the empty HashCodeBuilder constructor.
>> The only point of setting different numbers is when you might end up
>> comparing different object types (and even in that case you have the
>> #equals() safeguard anyway so it's mostly about performances in
>> practice).
>>
>> So here is a proposal: it's OK to use empty HashCodeBuilder constructor except:
>> * when overwriting the hachCode of a super class you should use
>> different numbers than the super class (since there is big chance to
>> have that super class used as a hash key)
>> * as a best effect in general if you know that you are going to have
>> several different types as hash key, try to make sure they use
>> different random numbers, your proc will thank you :)
>>
>> WDYT ?
>
> I don’t fully master the topic but I’m trusting you so +0 from me. Only comment is that it means there’s not a single way of writing our code and thus is not that easy to check.

The problem is that It's not easy to have a single way to write our
code here, if you want one it's going to be "it's OK to use empty
constructor of HashCodeBuilder" and the rest will move somewhere like
http://platform.xwiki.org/xwiki/bin/view/AdminGuide/Performances#HMemory
:)
In practice those are more performances related advices because saying
something like "all classes must use different numbers" is simply
impossible to enforce/check in practice because it's way to hard to
know if you are using unique numbers or not.

>
> Thanks
> -Vincent
>
>
>>
>> --
>> Thomas Mortagne
>



--
Thomas Mortagne