RFR(S) 8210388 Use hash table to store archived subgraph_info records

[Ioi Lam]

On 09/20/2018 03:07 PM, Jiangli Zhou wrote:
> Hi Ioi,
>
>
> On 9/20/18 11:28 AM, Ioi Lam wrote:
>> http://cr.openjdk.java.net/~iklam/jdk12/8210388-hashtable-subgraph-info.v01/ 
>>
>> https://bugs.openjdk.java.net/browse/JDK-8210388
>>
>> Here's another small step towards CDS support for Lambda (JDK-8198698).
>>
>> Replace the linear search of ArchivedKlassSubGraphInfoRecord with
>> a hashtable. Currently we have just 6 such records, but with Lambda 
>> support
>> that number is going to increase substiantially.
>>
>>      Shared subgraphs table stats
>>      Number of entries       :         6
>>      Total bytes used        :        72
>>      Average bytes per entry :    12.000
>>      Average bucket size     :     6.000
>>      Variance of bucket size :     0.000
>>      Std. dev. of bucket size:     0.000
>>      Empty buckets           :         0
>>      Value_Only buckets      :         0
>>      Other buckets           :         1
>>
>> I also took the chance to clean up the use of CompactHashtableWriter in
>> the symbol/string tables. In doing so, I fixed a bug where
>> SymbolTable::copy_shared_symbol_table was skipping symbols with zero 
>> hashcodes.
>> This should be done for the string table only.
>>
>> Jiangli, could you help me with this comments?
>>
>>     unsigned int hash = java_lang_String::hash_code(s);
>>     if (hash == 0) {
>>       // We do not archive Strings with a 0 hashcode because ......
>>       return true;
>>     }
> The intent was to prevent possible write into the closed archive 
> region when identity_hash was called on an empty String. The empty 
> String is archived in the open archive heap region. This may not be 
> necessary as we always pre-compute identity hash for archived object. 
> If you want to remove this restriction, please be careful to verify 
> all possible cases.
>

Hi Jiangli,

Thanks for the explanation. I am not going to change the logic in the 
code quoted above. I just wanted to add a comment there to explain why 
we are doing that. Do we have an existing comment for this else where?

I removed the (hash == 0) check in the symbol table dumping code, since 
that's a bug.

Thanks
- Ioi


> Thanks,
> Jiangli
>>
>>
>> Thanks
>> - Ioi
>