groovy-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Paolo Di Tommaso <paolo.ditomm...@gmail.com>
Subject Re: Add a marker interface to bypass Collections and Maps formatting
Date Mon, 26 Feb 2018 08:58:47 GMT
I want to report an update on the status of this issue. In the conversation
in the GitHub pull request <https://github.com/apache/groovy/pull/566> I
was requested to perform a benchmark to evaluate  the impact of this
change.

The results are showed below:

Benchmark                        (size)   Mode  Cnt        Score
Error   Units
ListBench.equalsDefaultGroovy         1  thrpt   15  2320508.042 ±
84277.415  ops/ms
ListBench.equalsDefaultGroovy        10  thrpt   15  2329172.399 ±
69070.211  ops/ms
ListBench.equalsDefaultGroovy       100  thrpt   15  2354738.681 ±
64525.890  ops/ms
ListBench.equalsDefaultGroovy      1000  thrpt   15  2358917.549 ±
46171.873  ops/ms
ListBench.equalsDefaultGroovy   1000000  thrpt   15  2377102.274 ±
48227.814  ops/ms
ListBench.equalsWithAnnotation        1  thrpt   15  2213530.048 ±
93671.364  ops/ms
ListBench.equalsWithAnnotation       10  thrpt   15  2277729.772 ±
89214.824  ops/ms
ListBench.equalsWithAnnotation      100  thrpt   15  2256582.572 ±
85712.999  ops/ms
ListBench.equalsWithAnnotation     1000  thrpt   15  2266409.619 ±
45408.683  ops/ms
ListBench.equalsWithAnnotation  1000000  thrpt   15  2197481.973 ±
127202.807  ops/ms
ListBench.equalsWithInterface         1  thrpt   15  2155416.804 ±
80496.840  ops/ms
ListBench.equalsWithInterface        10  thrpt   15  2061873.464 ±
211658.982  ops/ms
ListBench.equalsWithInterface       100  thrpt   15  2106467.218 ±
89659.046  ops/ms
ListBench.equalsWithInterface      1000  thrpt   15  2219947.677 ±
87623.149  ops/ms
ListBench.equalsWithInterface   1000000  thrpt   15  2092834.055 ±
165819.668  ops/ms
ListBench.equalsDefaultGroovy         1   avgt   15       ≈ 10⁻⁶
    ms/op
ListBench.equalsDefaultGroovy        10   avgt   15       ≈ 10⁻⁶
    ms/op
ListBench.equalsDefaultGroovy       100   avgt   15       ≈ 10⁻⁶
    ms/op
ListBench.equalsDefaultGroovy      1000   avgt   15       ≈ 10⁻⁶
    ms/op
ListBench.equalsDefaultGroovy   1000000   avgt   15       ≈ 10⁻⁶
    ms/op
ListBench.equalsWithAnnotation        1   avgt   15       ≈ 10⁻⁶
    ms/op
ListBench.equalsWithAnnotation       10   avgt   15       ≈ 10⁻⁶
    ms/op
ListBench.equalsWithAnnotation      100   avgt   15       ≈ 10⁻⁶
    ms/op
ListBench.equalsWithAnnotation     1000   avgt   15       ≈ 10⁻⁶
    ms/op
ListBench.equalsWithAnnotation  1000000   avgt   15       ≈ 10⁻⁶
    ms/op
ListBench.equalsWithInterface         1   avgt   15       ≈ 10⁻⁶
    ms/op
ListBench.equalsWithInterface        10   avgt   15       ≈ 10⁻⁶
    ms/op
ListBench.equalsWithInterface       100   avgt   15       ≈ 10⁻⁶
    ms/op
ListBench.equalsWithInterface      1000   avgt   15       ≈ 10⁻⁶
    ms/op
ListBench.equalsWithInterface   1000000   avgt   15       ≈ 10⁻⁶
    ms/op


The benchmark code is available at this link
<https://github.com/apache/groovy/compare/master...pditommaso:jmh-list-equals-benchmark>.
These numbers suggest that checking the class annotation in the
`DefaultGroovyMethods.equals` method adds an overhead of ~5%, doing the
same with a marker interface ~10%.


What do you think ?


p



On Fri, Jan 26, 2018 at 3:21 PM, Paolo Di Tommaso <paolo.ditommaso@gmail.com
> wrote:

> I agree. The name @GroovyOverride sounds a good option.
>
>
>
> Cheers,
> Paolo
>
>
> On Thu, Jan 25, 2018 at 11:53 PM, MG <mgbiz@arscreat.com> wrote:
>
>> This looks like something that might be useful in certain scenarios. A
>> "perfect fix" would always be better, but since that might be some time off
>> (2019 being optimistic - some of the things I want Groovy to improve in
>> date back to at least 2006)...
>>
>> My only question would be, if it would perhaps make sense to introduce a
>> more generically named annotation (@AutomatismOverride, @GroovyOverride,
>> @Configuration,...), that would allow overriding/fine-tuning many of
>> Groovy's automatisms through different parameters, to avoid an
>> annotation-explosion over time ?
>>
>> mg
>>
>>
>> On 23.01.2018 09:25, Paolo Di Tommaso wrote:
>>
>> Hi all,
>>
>> I want to take the opportunity to renew my proposal and PR to add an
>> annotation that allows the override of the Groovy default formatting for
>> certain classes.
>>
>> https://github.com/apache/groovy/pull/566
>>
>>
>> To quickly remind you what the problem is, Groovy provides a nice default
>> formatting for some classes i.e. String, Map, and Collection data
>> structures which is good. But it makes impossible to override it by
>> sub-classes that implements their own toString method. The same problem for
>> the `equals` method. This makes difficult to handle some specific use
>> cases, leaving bytecode manipulation as the only alternative.
>>
>>
>> My proposal is to add an annotation named @IgnoreDefaultEqualsAndToString
>> (or maybe @OverrideEqualsAndToString) to bypass the Groovy formatting and
>> allow the invocation of sub-classes `toString` and `equals` methods.
>>
>> I agree that's a sub-optional solution, however no better solutions have
>> been proposed for the current and future releases.
>>
>>
>>
>> Cheers,
>> Paolo
>>
>>
>>
>> On Sun, Jun 25, 2017 at 8:35 PM, Paolo Di Tommaso <
>> paolo.ditommaso@gmail.com> wrote:
>>
>>> Dear all,
>>>
>>> Groovy still does not provide a mechanism to override the `toString` and
>>> `equals` methods for custom Collection and Map objects. This is a serious
>>> limitation in some use cases.
>>>
>>> I'm proposing with the following pull request to introduce a marker
>>> annotation that allows a custom object to use the `toString` and `equals`
>>> as expected.
>>>
>>> https://github.com/apache/groovy/pull/566
>>>
>>>
>>> Any comment or improvement is welcome.
>>>
>>>
>>> Cheers,
>>> Paolo
>>>
>>>
>>> On Thu, Jun 2, 2016 at 1:38 AM, Paul King <paulk@asert.com.au> wrote:
>>>
>>>> I am +1 on improving how we handle formatting for lists and maps. My
>>>> default position would be -1 on an implementation that smells like it
>>>> might be "yet another hack" that we have to maintain long term. The
>>>> main reason being that we are trying to streamline method selection
>>>> for our revised MOP (I know not much is happening in that space right
>>>> now) and it would be nicer if once that is done, the "inconsistent"
>>>> results you mention could be handled in an easy to understand way.
>>>> Having said that, if I get time to look into it further and can't
>>>> think of a better way to approach it long term, then I could easily be
>>>> moved to at least a -0.
>>>>
>>>> Cheers, Paul.
>>>>
>>>> On Tue, May 31, 2016 at 7:36 PM, Paolo Di Tommaso
>>>> <paolo.ditommaso@gmail.com> wrote:
>>>> > Hello guys,
>>>> >
>>>> > No feedback on this? Would you take in consideration a PR for this
>>>> proposal?
>>>> >
>>>> >
>>>> > Thanks,
>>>> > Paolo
>>>> >
>>>> >
>>>> > On Sat, May 28, 2016 at 6:26 PM, Paolo Di Tommaso
>>>> > <paolo.ditommaso@gmail.com> wrote:
>>>> >>
>>>> >> Hi all,
>>>> >>
>>>> >> Groovy implements a built-in formatting strategy for collection
and
>>>> map
>>>> >> objects that is surely nicer and more useful than the one provided
>>>> by the
>>>> >> default Java implementation for these classes.
>>>> >>
>>>> >> However there are use cases in which custom collection or map
>>>> classes need
>>>> >> to implement their own formatting rule.
>>>> >>
>>>> >> Currently in Groovy this is quite painful and may lead to
>>>> inconsistent
>>>> >> results. Take in consideration the following example:
>>>> >>
>>>> >> class MyList extends ArrayList {
>>>> >>  String toString() {
>>>> >>     this.join('-')
>>>> >>   }
>>>> >> }
>>>> >>
>>>> >> def x = new MyList()
>>>> >> x << 1 << 2 << 3
>>>> >>
>>>> >> println x.toString()
>>>> >> println x
>>>> >> println "$x"
>>>> >>
>>>> >> Which prints:
>>>> >>
>>>> >> 1-2-3
>>>> >> [1, 2, 3]
>>>> >> [1, 2, 3]
>>>> >>
>>>> >> Both the second and third `println` use the Groovy built-in
>>>> formatting
>>>> >> method and there's no easy way to override this behaviour. Also
>>>> there's not
>>>> >> a clear reason why the first and the second print return a different
>>>> output.
>>>> >>
>>>> >> The only options I've found is to define `MyList` with a @Delegate
>>>> without
>>>> >> implementing the `List` interface. But this leads to other weird
side
>>>> >> effects. The remaining possibility is to use some bytecode
>>>> manipulation to
>>>> >> bypass the default Groovy formatting, but it looks to me a really
>>>> >> overkilling solution for such problem.
>>>> >>
>>>> >> For this reason a would like to propose to introduce a mechanism
that
>>>> >> would allow custom collection and map classes to bypass the default
>>>> >> formatting method. This should not be too difficult. The current
>>>> Groovy
>>>> >> built-in formatting is implemented by formatList and formatMap
>>>> methods.
>>>> >>
>>>> >> It would be enough to add a marker interface (or an annotation)
that
>>>> when
>>>> >> applied to a class it would be used to by-pass the logic in the
>>>> formatList
>>>> >> and formatMap methods and simply return the string provided by the
>>>> object
>>>> >> `toString` method.
>>>> >>
>>>> >>
>>>> >> I could easily contribute this patch however I would know the
>>>> opinion of
>>>> >> the Groovy core committers. In particular:
>>>> >>
>>>> >> 1) What name should have this marker interface?
>>>> groovy.lagn.Something?
>>>> >> 2) Are formatList and formatMap methods the right place to add this
>>>> logic?
>>>> >> 3) A similar problem exists also when using the `equals` (and
>>>> hashCode?)
>>>> >> method for collections and maps. Should this mechanism be extended
>>>> also to
>>>> >> this case?
>>>> >>
>>>> >>
>>>> >>
>>>> >> Best,
>>>> >> Paolo
>>>> >>
>>>> >>
>>>> >
>>>>
>>>
>>>
>>
>>
>

Mime
View raw message