Does oc.dict.values preserve order?

[will-gp]

I am wondering if oc.dict.values preserves order.

Given the following example:

pets.yaml

zebra:
    - name: Bob
    - age: 24
cat:
    - name: Louis
    - age: 5

and another yaml file which uses pets.yaml:

defaults:
  - pets

profile:
   - ${oc.dict.values:pets}

will oc.dict.values preserve the order of the objects in pets.yaml?
ie. the order will always be the values of [zebra,cat] and never [cat, zebra]

[Jasha10]

TLDR: Yes, oc.dict.{keys,values} does iterate over keys/values in order (for python version >= 3.7)

---

oc.dict.keys and oc.dict.values iterate over the keys/values in the same order that they appear in the DictConfig object. (You can see the implementation of oc.dict.keys and oc.dict.values in this file) Under the hood, DictConfig uses a python dictionary to store its contents (i.e. its keys and values). For Python version >=3.7, python dictionaries are guaranteed to preserve order of keys/values. This means that, for python >= 3.7, DictConfig itself is stable in terms of the order of its keys, and oc.dict.{keys,values} will preserve that order.

I noticed that your example above uses yaml files and a hydra-like defaults list. Note that according to the yaml spec, yaml mappings (i.e. "yaml dicts") are technically unordered. OmegaConf and Hydra use the pyyaml library to handle yaml parsing, and I'm not sure if pyyaml guarantees that the keys/values from yaml files will be parsed in order. In any case, looking at your Hydra example above, the order of cfg.pets.values() will be the same as the order of cfg.profile[0], even if that order is backwards when compared with the order of keys in the source yaml file pets.yaml.

[will-gp]

Thanks @Jasha10. It does seem, based on a number of discussions, that pyyaml preserves insertion order in python >= 3.7. In this case, even though the yaml mappings would be ordered, this would still represent a valid yaml.

I've included the link(s) below for reference:
yaml/pyyaml#110
yaml/pyyaml#523

I would note that although loading the yaml preserves order, yaml.dump() does not, which is a source of confusion for many.