RFR 8209387: Follow ups to JDK-8195100 Use a low latency hashtable for SymbolTable

[coleen.phillimore at oracle.com]

This looks good to me.  I agree with Gerard's comment about 
assert(sym->refcount() == 0).  It seems to be where it's important that 
the refcount is zero, ie before we free the node.

Thanks,
Coleen

On 12/1/18 12:27 PM, Gerard Ziemski wrote:
> Hi all,
>
> Please review these couple small “cleanup” tasks as a followup to JDK-8195100 (Use a low latency hashtable for SymbolTable)
>
> The first 2 of these affect both Symbol and String hashtables:
>
> #1 We replace macros with constant constructs.
>
> #2 We use “size_t" for counters, instead of “int”, to reduce the need of castings.
>
> #3 I expanded on the comment for “mark_item_clean_count"
>
> Please notice that:
>
> A) “_alt_hash” is handled separately by https://bugs.openjdk.java.net/browse/JDK-8214449
>
> B) I don’t personally agree with the 2nd point raised in this issue (i.e. placement of "assert that sym->refcount() == 0” in “delete_symbol”), so I’d like to leave the code as is.
>
> Issue:   https://bugs.openjdk.java.net/browse/JDK-8209387
> Webrev:  http://cr.openjdk.java.net/~gziemski/8209387_rev1
> Testing: Mach5 hs_tier1,2,3,4,5
>
>
> Cheers