Tukukkk

When loading a file containing anchors at different nesting levels, the anchor seems lost and the key remains empty.

When loading the following file:

---

Dict1:

    - InnerDict: &inner

            key: val



Dict2:

    InnerDict:

        <<: *inner



Dict3:

    - InnerDict:

        <<: *inner

...

... with the code (Python 3.7, ruamel.yaml version 0.15.78):

from ruamel.yaml import YAML

with open("file.yaml") as infile:

    content = YAML(typ='rt', pure=True).load(infile)

print(content)

... gives:

{'Dict1': [ordereddict([('InnerDict', ordereddict([('key', 'val')]))])],

 'Dict2': ordereddict([('InnerDict', ordereddict())]),

 'Dict3': [ordereddict([('InnerDict', ordereddict([('key', 'val')]))])]}

... where the inner dict of Dict2 remains empty.

In safe mode, the anchor is interpreted as expected in both cases.

Is it intended or is it a bug ?

That is definitely a bug, and it has to do with the breadth-first
construction of mappings in the data structure. By the time the
InnerDict under Dict2 is constructed, the one nested under Dict1
is not completely available. The InnerDict under Dict3 lies at the
same depth, and can therefore correctly be constructed (and so can the
first merge if you remove the dash in the Dict1 construction and so
making the anchor appear at a shallower place).

One way to solve this, apart from installing ruamel.yaml>=0.15.79, is by
providing an alternative constructor, that forces depth first processing:

import sys

import ruamel.yaml



class MyConstructor(ruamel.yaml.constructor.RoundTripConstructor):

    def construct_yaml_map(self, node):

        data = ruamel.yaml.comments.CommentedMap()

        data._yaml_set_line_col(node.start_mark.line, node.start_mark.column)

        yield data

        self.construct_mapping(node, data, deep=True)

        self.set_collection_style(data, node)



MyConstructor.add_constructor(

    u'tag:yaml.org,2002:map', MyConstructor.construct_yaml_map

)





yaml = ruamel.yaml.YAML()

yaml.Constructor = MyConstructor





yaml_str = """

Dict1:

    - InnerDict: &inner

            key: val



Dict2:

    InnerDict:

        <<: *inner



Dict3:

    - InnerDict:

        <<: *inner

"""



data = yaml.load(yaml_str)

for k in data:

   print(k, data[k])

print('---------')

yaml.dump(data, sys.stdout)

Generating:

Dict1 [ordereddict([('InnerDict', ordereddict([('key', 'val')]))])]

Dict2 ordereddict([('InnerDict', ordereddict([('key', 'val')]))])

Dict3 [ordereddict([('InnerDict', ordereddict([('key', 'val')]))])]

---------

Dict1:

- InnerDict: &inner

    key: val



Dict2:

  InnerDict:

    <<: *inner



Dict3:

- InnerDict:

    <<: *inner

(As the dump of data is correct even without the above "patch", this
was not previously detected when testing round-trips)