On 23 July 2012 01:54, Jeevitha Muthusamy <
mail2jeevithamk@xxxxxxxxx> wrote:
> May I know what is Bytecode Caching in AspectJ1.7 and its benefits??
https://bugs.eclipse.org/bugs/show_bug.cgi?id=367673
It is a per classloader cache of woven bytecode that avoids re-weaving
on subsequent restarts.
> And when caching is enabled, what exactly is cached??
Per the readme and that bug, the caching is enabled when you turn on
the system property:
-Daj.weaving.cache.enabled=true
The woven bytecode is cached on a per classloader basis. On a
subsequent JVM restart
the cached woven bytecode will be used instead of repeating the weaving process.
> What is the life of Caching? Is it automatically refreshed or is Static?
It is not a super intelligent cache. If the incoming classes change
(in length) then the
cache will not be used (as it looks like this class needs reweaving),
but changing the
set of aspects will not cause reweaving to happen (IIRC). So this is
for a system
that is pretty static in terms of aspects but you want faster startup
on secondary starts.
> Is there any Memory limit for Caching? ie. If only to certain extent
> Caching is done or is unlimited??
disk space is the limiting factor I believe.
> Is there any constraints(Where this feature is most applicable? ) to use
> this feature?
If you have a system where the set of aspects are fixed, this can work
well for you.
The bug report contains some performance metrics showing you the benefits.
> It will be more helpful, if any use-cases provided or any documents
> regarding this feature!!
Unfortunately the only real docs are in the bugzilla, I haven't had the time
to integrate anything into the main docs.
There are further changes coming into this area on two fronts:
- 'asynchronous' caching. It has become apparent that storing entries
on the disk in the first start is a bottle-neck and slowing down that
initial start. aynchronous caching will do the storing in a thread.
- assume common aspects across all classloaders to give a further
boost. If your system meets this requirement then the cache will
actually be consulted during the initial start if it sees two
classloaders loading the same class (only one will suffer the weaving
cost). This can make the initial start faster than without caching
(as well as speeding up the latter restarts).
Andy
_______________________________________________
aspectj-users mailing list
aspectj-users@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/aspectj-users
