[ruby-core:79830] [Ruby trunk Feature#12901] Anonymous functions without scope lookup overhead

Issue #12901 has been updated by Shyouhei Urabe.


Benoit Daloze wrote:
> Shyouhei Urabe wrote:
> > I think it's impossible to optimize lambdas without breaking backwards compatibility.
> 
> It is possible with deoptimization though.
> An optimizer might assume that a block does not call binding and in case it does, deoptimize.

Let me ask a (possibly stupid) question.  Consider this scenario:

1. create a proc.
2. it gets optimized.
3. lots of calculations goes on without variable captures.
4. obtain a binding from it (via deoptimization).

now, should what be visible from the obtained binding?  Should we re-calculate the optimized-out variables?

Of course it could be anything if we are allowed to introduce incompatibility but...

> Removing the scope lookup overhead for accessing captured variables
> is possible if one considers a whole method containing the lambda for analysis.
> However, it seems much harder to optimize the scope lookup for a lambda capturing a non-optimized scope.
>
> I do not think scope lookup is the only overhead of lambdas though,
> there is a lot of implicitly captured state which needs to be restored for every call
> (but the same optimizations apply when optimizing a method containing the lambda).

I agree with those parts.

----------------------------------------
Feature #12901: Anonymous functions without scope lookup overhead
https://bugs.ruby-lang.org/issues/12901#change-63265

* Author: Richard Schneeman
* Status: Open
* Priority: Normal
* Assignee: 
* Target version: 
----------------------------------------
Right now if you are writing performance sensitive code you can get a performance bump by not using anonymous functions (procs, lambdas, blocks) due to the scope lookup overhead.

https://twitter.com/apotonick/status/791167782532493312?lang=en

I would like to be able to create an anonymous function and specify I don't need it to perform the lookup when called.

I am thinking that this code:

~~~
Proc.new(scope: false) {|var| puts var }
~~~

Would be the equivalent of 

~~~
def anon(var)
  puts var
end
~~~

If you call it while trying to access something from outside of the scope it would error

~~~
var = "hello"
Proc.new(scope: false) { puts var }.call
# => NameError: undefined local variable or method `var' for main:Object
~~~

An example of a case where this could be used is found in https://github.com/rails/sprockets/pull/417. In this case we are getting rid of anonymous functions in favor of a method created with refinements. This solution works but it was not an obvious optimization. It would be convenient to have a syntax for defining anonymous functions that do not need access to caller scope.






-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:[e-mail]?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>