Array subclass: uniq, flatten, other methods (?) do not clone correctly

Nice find (unfortunately ). Can I ask whether this use case came from some public library or other public code? We are interested in pointers to real-life cases of this that we can point to.

Sure - I found this while trying to do flatten and uniq on an instance of WillPaginate::Collection, which is a subclass of Array with some additional properties. See http://wiki.github.com/mislav/will_paginate

Thanks Billy...can you go ahead and add specs for this to RubySpec, and we'll see about getting it fixed? http://www.rubyspec.org/

It's more an internal dupping issue (btw, we match 1.9 here)

Marcin - Yeah, strange! I didn't get exactly the same thing when I tried 1.9, though - I compiled the latest stable release from source and got a third possibility, which is confusing.

ruby 1.9.1p129 (2009-05-12 revision 23412) [i386-darwin9.6.2]

"bar"
nil
"bar"

Charles, I would be happy to add something to the spec, but I guess I'm confused as to what should be happening here - especially since it appears that c ruby 1.8 and 1.9 don't even do the same thing. My opinion is that the 1.8.6 output is correct. If the spec advertises that a method returns a copy of the array with some filtering on the collection (such as uniq or flatten do), it should really copy that object in its entirety, including its state. In the very least, the methods should be consistent - either they should not copy the state of bolted-on properties, or they should - but what c ruby 1.9.1 appears to do is even more confusing.

Billy: If it's different in different versions, we should just add the different behaviors with version guards. Brian Ford (brixen on IRC) can point you toward good instructions to do that. And in fact if there's differences between versions, it's even more important we spec it so we clearly understand the differences.

Also, if you feel like what Ruby 1.9.1 does is in error, we should probably raise an issue with ruby-core and see if there's a reason for it.

RubySpec updates should be created for missing cases, but otherwise this may be an easy one to fix. At least repair the mentioned methods, but a review of other methods that return new Array instances would be useful. Marking as Intro.

Both C ruby 1.8.7 and 1.9.2 return the same result:

ruby 1.8.7 (2010-12-23 patchlevel 330) [i686-darwin10.6.0]

"bar"
nil
"bar"

ruby 1.9.2p136 (2010-12-25 revision 30365) [x86_64-darwin10.6.0]

"bar"
nil
"bar"

Can one of you (Charles or Brian) point me to a sample rubyspec that is using version guards? I am not on IRC all the time. I will try to catch one of you on IRC whenever I can or post questions to dev mailing list.

is this really expected? after all, even MRI docs assert that a new_ary would be returned.

I'm trying to file it at ruby's tracker, but it isn't working for now

I'm going to mark this as fixed as of 1.6 (probably was fixed earlier) since we match MRI 1.8.7 and 1.9.2. If it turns out later that ruby-core agrees it's a bug, we can look into altering the behavior.