Well that's essentially what --fast mode does. It turns on NOFRAME
everywhere it's safe, and yes, it has a big impact on performance,
especially for small methods. The NOFRAME pragma came from a time where
--fast wasn't as smart as it is now about determining when it's safe.
I still plan to make --fast fully functional for JRuby 1.3, though it
runs a lot of stuff already.
Joseph Athman wrote:
Well I would think if JRuby can figure out on its own if the NOFRAME
optimization is safe or not then would I really need to tell it myself?
Maybe this is an optimization that will be enabled for everything
eventually? I would think these pragmas would only be used if there was
no way for the compiler to decide if an optimization was safe or not.
This is really interesting stuff, I wonder how much of a speed boost
could be made with little hints like this?
Joe
On Mon, Apr 13, 2009 at 10:51 PM, Charles Oliver Nutter
<[email protected] <mailto:[email protected]>> wrote:
Joseph Athman wrote:
This looks pretty good to me. So basically if a __NOFRAME__
local variable is found inside a method then JRuby will make its
enhancements?
That's right, though currently it won't make enhancements that its
own static inspection detects are unsafe. In other words, if you say
__NOFRAME__ = true, but then proceed to call eval or use $~, we'll
save you from yourself. Would it be better for it to be a "hard"
pragma that lets you blow up if you want to?
Have you talked with the Rubinius or IronRuby guys about these
ideas at all? I'm not really sure if these optimizations are
implementation specific or not, but maybe clues could be added
that more than just JRuby would understand.
I've talked a few times with Evan about this. Rubinius already has
one enhancement I'd like to add, an "asm { }" block where you can
put raw RBX ASM. In our case, it would essentially be BiteScript
within that block, and you'd be emitting JVM bytecode. Given that,
you could implement almost any hack you wanted, since you'd
essentially be adding raw JVM bytecode into the compile sequence.
Outside of this, we haven't talked much about pragmas.
Could you explain this a little more? A compile plugin sounds a
little scary to me.
The basic idea is that you'd provide plugins to be loaded when
running "jrubyc" and they'd modify compilation by hooking into the
process. I haven't given it a whole lot of thought, but I'd like a
way to be able to modify the compile phase without directly
modifying the compiler.
- Charlie
---------------------------------------------------------------------
To unsubscribe from this list, please visit:
http://xircles.codehaus.org/manage_email
---------------------------------------------------------------------
To unsubscribe from this list, please visit:
http://xircles.codehaus.org/manage_email
