Pathname#to_s stack overflow on Rubinius about fssm HOT 7 CLOSED

I've been working mostly with python and java these days and am pretty swamped ATM. The soonest I could get to this is probably the weekend, but I definitely want to support rubinius. All of my testing and use so far has been on MRI 1.8.{6,7}, 1.9.{1,2}, and various releases of jruby... with most of my real usage being in jruby (of course).

Supporting rubinius is probably going to be less complicated than macruby (and still support fsevent).

from fssm.

Evan kindly applied a patch to Rubinius that fixes that issue, but there's yet another recursion loop:

  /home/jonathan/.rvm/gems/rbx-head/gems/haml-3.0.24/vendor/fssm/lib/fssm/pathname.rb:43:in `<<'
  kernel/common/file.rb:628:in `join'
  kernel/bootstrap/array.rb:71:in `each'
  kernel/common/file.rb:606:in `join'
  /home/jonathan/.rvm/gems/rbx-head/gems/haml-3.0.24/vendor/fssm/lib/fssm/pathname.rb:373:in `join'
  /home/jonathan/.rvm/gems/rbx-head/gems/haml-3.0.24/vendor/fssm/lib/fssm/pathname.rb:405:in `join'

I think Rubinius' handling is actually very sensible, since it treats String and the other primitives simply as other Ruby objects, not special C entities. I'd be glad to try my hand at refactoring Pathname to something that doesn't rely on implementation-specific quirks, if you don't have the time to spare.

from fssm.

The entire purpose of the custom pathname library is that when I profiled FSSM ages ago, most of its' time was being spent there. No lie. So, at the time at least, there was a method to my madness.

The old results are still in the tree actually... go team crufty cruft.

Using pathname (27.9s): https://github.com/ttilley/fssm/blob/master/profile/prof-plain-pathname.html

Using FSSM's pathname (14.4s): https://github.com/ttilley/fssm/blob/master/profile/prof-fssm-pathname.html

That's pretty damn close to half. Those tests, however, haven't been run since early September 2009. And other than indentation, that code really hasn't been looked at since then either. For all I know, it might not make that much of a difference any more. It might also be the case that the implementation of pathname in rubinius performs better than what's in FSSM now.

I'd urge you to look into it and would gladly accept any contributions to the codebase that make it more accessible. ^_^

from fssm.

I've just forked FSSM and run the specs: "26 examples, 14 failures". I'll see what I can do to turn those fails into passes!

from fssm.

I have a fix that keeps the current class, but the fix is a bit... icky to say the least. I'd be interested to see how it performs versus using just Pathname on rubinius... or elsewhere, now that the change has been made.

from fssm.

using "-Xjit.sync=true -Xjit.call_til_compile=1 -Xprofile" FSSM::Pathname had a runtime of 4.9s, while Pathname had a runtime of 8.9s.

from fssm.

i'm going to increase the iterations and commit the script shortly.

from fssm.