如何使用CommentedMap递归排序YAML（带锚点）？

解决此类问题的方法是，首先添加期望的和必要的导入，将输入和期望的输出定义为多行字符串，然后diff向YAML实例添加有用的方法。

测试时，字符串输入比文件更易于使用，因为所有内容都在一个文件中（需要删除一些尾随空格吗？），并且您无法覆盖输入并以与第一个不同的方式开始下一个运行。

import sys
import difflib
import ruamel.yaml
from ruamel.yaml.comments import merge_attrib

yaml_in = """\
_world:
  anchor_struct: &anchor_struct
    foo:
      val: "foo"
    bar:
      val: "foo"
  string: "string"
  newmsg: &newmsg
    msg: "msg"
    foo: "foo"
    new: "new"
  anchr_val: &anchor_val famous_val
  bool: True
elem2:
  myStruct:
    <<: *anchor_struct
  anchor_val: *anchor_val
  <<: *anchor_struct
  zzz: zorglub
  www: web
  anchor_struct:
    <<: *anchor_struct
    other_elem: "other_elem"
elem1:
  <<: *anchor_struct
  zzz: zorglub
  newmsg: 
    <<: *newmsg
    msg: "msg2"
  myStruct:
    <<: *anchor_struct
  anchor_struct:
    second_elem: "second_elem"
    <<: *anchor_struct
    other_elem: "other_elem"
  www: web
  anchor_val: *anchor_val
"""

yaml_out = """\
_world:
  anchor_struct: &anchor_struct
    bar:
      val: "foo"
    foo:
      val: "foo"
  anchr_val: &anchor_val famous_val
  bool: True
  newmsg: &newmsg
    foo: "foo"
    msg: "msg"
    new: "new"
  string: "string"
elem1:
  <<: *anchor_struct
  anchor_struct:
    <<: *anchor_struct
    other_elem: "other_elem"
    second_elem: "second_elem"
  anchor_val: *anchor_val
  myStruct:
    <<: *anchor_struct
  newmsg: 
    <<: *newmsg
    msg: "msg2"
  www: web
  zzz: zorglub
elem2:
  <<: *anchor_struct
  anchor_struct:
    <<: *anchor_struct
    other_elem: "other_elem"
  anchor_val: *anchor_val
  myStruct:
    <<: *anchor_struct
  www: web
  zzz: zorglub
"""


def diff_yaml(self, data, s, fnin="in", fnout="out"):
    # dump data if necessary and compare with s
    inl = [l.rstrip() + '\n' for l in s.splitlines()]   # trailing space at end of line disregarded
    if not isinstance(data, str):
        buf = ruamel.yaml.compat.StringIO()
        self.dump(data, buf)
        outl = buf.getvalue().splitlines(True)
    else:
        outl = [l.rstrip() + '\n' for l in data.splitlines()]
    diff = difflib.unified_diff(inl, outl, fnin, fnout)
    result = True
    for line in diff:
        sys.stdout.write(line)
        result = False
    return result

ruamel.yaml.YAML.diff = diff_yaml

yaml = ruamel.yaml.YAML()
# yaml.indent(mapping=4, sequence=4, offset=2)
yaml.boolean_representation = ["False", "True"]
yaml.preserve_quotes = True

然后确保您期望的输出有效，并且可以往返：

dout = yaml.load(yaml_out)
buf = ruamel.yaml.compat.StringIO()
yaml.dump(dout, buf)
assert yaml.diff(dout, yaml_out)

既不给出输出也不给出断言错误（预期输出中包含尾随空格，以及非默认True布尔值）。如果预期输出不能往返，则ruamel.yaml可能无法转储您的预期输出。

如果您陷入困境，现在可以检查dout以确定解析后的输入应该是什么样子。

所以现在尝试 recursive_sort

def recursive_sort_mappings(s):
    if isinstance(s, list):
        for elem in s:
            recursive_sort_mappings(elem)
        return 
    if not isinstance(s, dict):
        return
    for key in sorted(s, reverse=True):
        value = s.pop(key)
        recursive_sort_mappings(value)
        s.insert(0, key, value)

din = yaml.load(yaml_in)
recursive_sort_mappings(din)
yaml.diff(din, yaml_out)

这给出了很多输出，因为recursive_sort_mappings不知道合并并在所有键上运行，尝试将合并键保持在其原始位置，此外，在弹出键时（在将其重新插入第一个位置之前）如果弹出的值存在于合并映射中，则有些魔术：

--- in
+++ out
@@ -1,8 +1,8 @@
 _world:
   anchor_struct: &anchor_struct
-    bar:
+    bar: &id001
       val: "foo"
-    foo:
+    foo: &id002
       val: "foo"
   anchr_val: &anchor_val famous_val
   bool: True
@@ -14,24 +14,38 @@
 elem1:
   <<: *anchor_struct
   anchor_struct:
+    bar: *id001
     <<: *anchor_struct
+    foo: *id002
     other_elem: "other_elem"
     second_elem: "second_elem"
   anchor_val: *anchor_val
+  bar: *id001
+  foo: *id002
   myStruct:
     <<: *anchor_struct
+    bar: *id001
+    foo: *id002
   newmsg:
     <<: *newmsg
+    foo: "foo"
     msg: "msg2"
+    new: "new"
   www: web
   zzz: zorglub
 elem2:
-  <<: *anchor_struct
   anchor_struct:
     <<: *anchor_struct
+    bar: *id001
+    foo: *id002
     other_elem: "other_elem"
   anchor_val: *anchor_val
+  <<: *anchor_struct
+  bar: *id001
+  foo: *id002
   myStruct:
     <<: *anchor_struct
+    bar: *id001
+    foo: *id002
   www: web
   zzz: zorglub

为了解决这个问题，您需要做很多事情。首先，您需要放弃.insert（），它模拟（对于Python3内置OrderedDict）方法定义的Corderdict包ruamel.ordereddict包。此仿真将重新创建OrderedDict，并导致重复。Python3 C实现的功能（不如.insert()）不那么强大，但是在这种情况下是有用的方法move_to_end（可.insert()在ruamel.yaml中的仿真更新中使用）。

其次，您只需要遍历“真实”键，而不需要合并提供的那些键，因此您不能使用for key in。

第三，如果合并键在其他位置，则需要将其移到映射的顶部。

（level已添加参数用于调试目的）

def recursive_sort_mappings(s, level=0):
    if isinstance(s, list): 
        for elem in s:
            recursive_sort_mappings(elem, level=level+1)
        return 
    if not isinstance(s, dict):
        return
    merge = getattr(s, merge_attrib, [None])[0]
    if merge is not None and merge[0] != 0:  # << not in first position, move it
       setattr(s, merge_attrib, [(0, merge[1])])

    for key in sorted(s._ok): # _ok -> set of Own Keys, i.e. not merged in keys
        value = s[key]
        # print('v1', level, key, super(ruamel.yaml.comments.CommentedMap, s).keys())
        recursive_sort_mappings(value, level=level+1)
        # print('v2', level, key, super(ruamel.yaml.comments.CommentedMap, s).keys())
        s.move_to_end(key)


din = yaml.load(yaml_in)
recursive_sort_mappings(din)
assert yaml.diff(din, yaml_out)

然后差异不再提供输出。