From: Charles Oliver Nutter <headius@...>
Date: 2009-08-09T10:24:49+09:00
Subject: Re: Kind of ParsTree for 1.9.1

On Wed, Aug 5, 2009 at 9:56 AM, Brian Candler<b.candler@pobox.com> wrote:
> I think that's the point :-) You *have* to pass it a String source for
> it to parse. You can't take an object which Ruby has already parsed and
> installed, such as method(:foo) in this case, and get the sexp from the
> object.

Yeah, what's wrong with that?

Ignoring for the moment that different impls (and even different revs
of MRI) have (often wildly) different AST structures...

The problem with expecting you can get an AST at runtime is that it
means impls have to hold on to the AST for you. Guess what one of the
largest startup memory costs is in MRI and JRuby...that's right, it's
the AST from loading all that code into a big in-memory data
structure. Ruby 1.9 doesn't have the AST available anymore for a good
reason: it doesn't need it. JRuby won't need it soon either once our
new compiler lands, and already doesn't have it available when you
precompile code. So I guess it's a question of which is
better...getting rid of that memory use or having the AST around all
the time.

And again, that ignores that the AST is ever-changing and usually very
different across impls.

- Charlie