Aliases break if there are temporary objects

[tim.hochberg@…]

If an object being passed to represet_data has a shorter lifespan than the representer, the results are unpredictable, but generally bad. In my case, I'm trying to represent a custom omap class. I register a representer like so:

def omap_representer(dumper, data):
    items = [[x, y] for (x, y) in data.iteritems()] 
    return dumper.represent_sequence(u'!omap', items)
    
yaml.add_representer(omap, omap_representer)

If I then dump something that contains multiple omaps, such as [one_omap,another_omap], the representer get's confused because it sees distinct objects that have the same id. Here's an actual example:

>>> a # Note that these are omaps not dictionaries, despite appearances.
[{1: 2, 2: 4}, {1: 99, 2: 88}]

>>> print yaml.dump(a)
- !omap
  - &id001 [1, 2]
  - &id002 [2, 4]
- !omap
  - *id001
  - *id002

Two approaches come to mind to fix this.

1. Hold onto a reference to the original object as well as to the serialized object in represented_objects. This is what I did as a temporary fix.
2. Do some sort of weak reference magic to track if an object dies and then delete it from represented_objects. This is more complicated and I'm not entirely sure it would work well. The upside is it's possibly more frugal with memory.

-tim

[xi]

Fixed in [222].

It's a subtle bug. Thanks for the report.

I've made representer to keep references to the represented objects. Perhaps, not very efficient, but should always work.