Jens Elkner <[email protected]> writes:

> On Tue, Oct 30, 2012 at 09:52:02AM +0000, [email protected] wrote:
>> Author: Knut Anders Hatlen <[email protected]>
>> Repository: /hg/opengrok/trunk
>> Branch: default
>> Latest revision: d91dcc85adc3927332d46d0c311e7e7cbe9d81fc
>> Total changesets: 1
>> Log message:
>> Don't use String.intern() in the fix for bug #809.
>> 
>> The String.intern() method may exhaust permgen space, so it's better
>> to use the Interner class which places the seen symbols on the heap.
>
> Wondering, whether their is any benefit in "interning" the match/signature 
> as well. My guts say no ...

It currently does intern them, but I think you're right that there's
less duplication in that field, so it's probably not necessary. It
shouldn't cause much harm, though.

> BTW: Anyone already tried to rise the IntegerCache Limit like
> -XX:AutoBoxCacheMax=43561|32639|21716|16255|10793|8063|5332|3967|2601|1919|1236|895|553|383
> ?
>
> Since OG uses a lot of autoboxing / int <-> Integer mapping, often
> outside of -128..127, it may save a little bit mem as well (on 64-bit 
> 24 byte/Integer IIRC) and may give some speed improvements (less obj
> create, less work for the GC ...).

Haven't tested it. Would be interesting to see if it had any measurable
impact on the performance if someone ran an experiment. Would be a cheap
way to improve the performance if we could just add that option to the
OpenGrok script. I'd expect the big bottlenecks to be somewhere else,
but who knows...

-- 
Knut Anders
_______________________________________________
opengrok-dev mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/opengrok-dev

Reply via email to