YAML BUG

[haldane]

Does anyone know why this program causes YAML to go into an infinite
loop until it runs out of stack?

http://194.100.133.72/rubytest/t.rb.txt


I tried it with the latest cvs snapshot - ruby 1.9.0 (2004-06-29)
[i686-linux]

(I'm trying to get yaml to sucessfully dump an instiki wiki.)

Stephen Sykes

 

[haldane]

Does anyone know why this program causes YAML to go into an infinite

loop until it runs out of stack?

http://194.100.133.72/rubytest/t.rb.txt

I have a much shorter example now:

require 'yaml'
class A
attr_accessor :b,:c
def initialize
@f = "h" * 2000
@g = $x
end
end

$x = A.new

a = A.new
a.b = A.new
a.b.b = A.new
a.b.c = a

# yaml goes into an infinite loop
y a

My theory is that yaml keeps a track of objects, and only assigns them
an id number if it sees the object for the second time (in emitter.c).
If it does this, it tries to go back and write that id where the
object first occurred. But if the 4k buffer has already been exceeded
and flushed, then the id cannot be written, and the whole object is
written out again. Causing an infinite loop.

This is a serious problem. If my theory is right, I guess it could be
fixed by always writing an id for every object in the expectation that
it may be met again?

_why: any comments?

Stephen Sykes

 

[why the lucky stiff]

My theory is that yaml keeps a track of objects, and only assigns them
an id number if it sees the object for the second time (in emitter.c).
If it does this, it tries to go back and write that id where the
object first occurred. But if the 4k buffer has already been exceeded
and flushed, then the id cannot be written, and the whole object is
written out again. Causing an infinite loop.

This is definitely the problem.

This is a serious problem. If my theory is right, I guess it could be
fixed by always writing an id for every object in the expectation that
it may be met again?

This solution would make for a lot of ugly YAML. Too big of a tradeoff
for the lesser case of large recursive elements.

I'll just collect IDs for all the collections in a tree before emitting.
Ok, it's etched into the schedule. Thank you!

_why