AutoLock

[David Holmes]

Hi Heinz,

One slight glitch here is that, as I recall (and I'd like Joe to confirm 
as I can't find anything in writing) they updated the spec to disallow 
simple expressions, so you need to use something slightly more verbose 
that introduces a local variable to refer to the Lock.

Also as I understand it escape-analysis is only used by the server 
compiler, so we can not rely on having it to get rid of the object 
creation overhead. As indicated in the other threads on coin-dev there 
are ways to do this without having to introduce a new object as a 
wrapper - but they introduce their own overheads.

By the way, the "whopping 553 classes" that implement AuoCloseable comes 
about due to Closeable being retrofitted to also be AutoCloseable. No 
big surprise that these are primarily (all?) I/O related classes.

Cheers,
David Holmes


Dr Heinz M. Kabutz said the following on 02/28/11 05:30:
> Hi,
> 
> Brian told me about the discussion going on here on coin-dev.  I 
> published a newsletter today on the new Java 7 try-with-resource 
> construct and how we could use it for automatically unlocking.  By 
> combining it with static imports, we can write the following code:
> 
>     try (lock(lock)) {
>       printLockStatus();
>     }
> 
> Instead of the current more verbose
> 
>     lock.lock();
>     try {
>       printLockStatus();
>     } finally {
>       lock.unlock();
>     }
> 
> In my newsletter, I expressed the same concern as Neal, that 
> constructing objects, and in fact encouraging people to do this in 
> performance critical code, could be a killer for this idea.
> 
> However, initial tests indicate that escape analysis actually does a 
> nice job of eliminating the cost of the object construction.  These are 
> just completely basic initial tests and are quite possibly flawed.  
> However, with the default escape analysis turned on, my try (lock(lock)) 
> code did not cause any objects to be created, even though I called it 
> 1_000_000_000 times.
> 
> When I turned off escape analysis, it constructed about 30GB of objects.
> 
> As I said, I have not tested this properly, just an initial investigation.
> 
> Here is the link to my article: 
> http://www.javaspecialists.eu/archive/Issue190.html
> 
> Regards
> 
> Heinz