RFR 8200696 : Optimal initial capacity of java.lang.Class.enumConstantDirectory

[Ivan Gerasimov]

Hi David!


On 4/3/18 4:49 PM, David Holmes wrote:
> Hi Ivan,
>
> On 4/04/2018 9:22 AM, Ivan Gerasimov wrote:
>> Hello!
>>
>> Yet another occurrence of not-optimally pre-sized HashMap.
>>
>> When java.lang.Class.enumConstantDirectory is created, the initial 
>> capacity is set to be (2 * universe.length), which is more than 
>> necessary in some cases.
>>
>> Choosing the capacity optimally will allow us to save a few bytes 
>> with some enum classes.
>
> How are you defining optimal? 
By optimal I meant the minimum value that will not make the internal 
storage to be reallocated during insertion.

Here's a quotation from the documentation [1]:
"If the initial capacity is greater than the maximum number of entries 
divided by the load factor, no rehash operations will ever occur."

Given the default load factor 0.75, the formula (expectedSize / 0.75 + 
1) gives us the optimal initial capacity.

The testlibrary class testlibrary/OptimalCapacity.java is used to make 
sure that the chosen initial capacity is neither too low (so that no 
reallocations happen) nor too high (no memory is wasted).

With kind regards,
Ivan

[1] https://docs.oracle.com/javase/10/docs/api/java/util/HashMap.html


> The bug report only gives one example where the current calculation 
> produces a too large value. If the HashMap itself rounds up the 
> initial capacity to a power of 2, how much extra space do we need 
> compared to universe.length? Do we just need to pass 
> (universe.length+1) to get the smallest power of 2 greater than 
> universe.length?
>
> Thanks,
> David
>
>> Would you please help review this trivial fix?
>>
>> BUGURL: https://bugs.openjdk.java.net/browse/JDK-8200696
>> WEBREV: http://cr.openjdk.java.net/~igerasim/8200696/00/webrev/
>>
>

-- 
With kind regards,
Ivan Gerasimov