Details

  • support for simple lists as mapping keys by transforming these to tuples
  • !!omap generates ordereddict (C) on Python 2, collections.OrderedDict on Python 3, and !!omap is generated for these types.
  • Tests whether the C yaml library is installed as well as the header files. That library doesn’t generate CommentTokens, so it cannot be used to do round trip editing on comments. It can be used to speed up normal processing (so you don’t need to install ruamel.yaml and PyYaml). See the section Optional requirements.
  • Basic support for multiline strings with preserved newlines and chomping ( ‘|‘, ‘|+‘, ‘|-‘ ). As this subclasses the string type the information is lost on reassignment. (This might be changed in the future so that the preservation/folding/chomping is part of the parent container, like comments).
  • anchors names that are hand-crafted (not of the form``idNNN``) are preserved
  • merges in dictionaries are preserved
  • adding/replacing comments on block-style sequences and mappings with smart column positioning
  • collection objects (when read in via RoundTripParser) have an lc property that contains line and column info lc.line and lc.col. Individual positions for mappings and sequences can also be retrieved (lc.key('a'), lc.value('a') resp. lc.item(3))
  • preservation of whitelines after block scalars. Contributed by Sam Thursfield.

In the following examples it is assumed you have done something like::

from ruamel.yaml import YAML
yaml = YAML()

if not explicitly specified.

Indentation of block sequences

Although ruamel.yaml doesn’t preserve individual indentations of block sequence items, it does properly dump:

x:
- b: 1
- 2

back to:

x:
-   b: 1
-   2

if you specify yaml.indent(sequence=4) (indentation is counted to the beginning of the sequence element).

PyYAML (and older versions of ruamel.yaml) gives you non-indented scalars (when specifying default_flow_style=False):

x:
-   b: 1
- 2

You can use mapping=4 to also have the mappings values indented. The dump also observes an additional offset=2 setting that can be used to push the dash inwards, within the space defined by sequence.

The above example with the often seen yaml.indent(mapping=2, sequence=4, offset=2) indentation:

x:
  y:
    - b: 1
    - 2

The defaults are as if you specified yaml.indent(mapping=2, sequence=2, offset=0).

If the offset equals sequence, there is not enough room for the dash and the space that has to follow it. In that case the element itself would normally be pushed to the next line (and older versions of ruamel.yaml did so). But this is prevented from happening. However the indent level is what is used for calculating the cumulative indent for deeper levels and specifying sequence=3 resp. offset=2, might give correct, but counter intuitive results.

It is best to always have sequence >= offset + 2 but this is not enforced. Depending on your structure, not following this advice might lead to invalid output.

Inconsistently indented YAML

If your input is inconsistently indented, such indentation cannot be preserverd. The first round-trip will make it consistent/normalize it. Here are some inconsistently indented YAML examples.

b indented 3, c indented 4 positions:

a:
   b:
       c: 1

Top level sequence is indented 2 without offset, the other sequence 4 (with offset 2):

- key:
    - foo
    - bar

Positioning ‘:’ in top level mappings, prefixing ‘:’

If you want your toplevel mappings to look like:

library version: 1
comment        : |
    this is just a first try

then set yaml.top_level_colon_align = True (and yaml.indent = 4). True causes calculation based on the longest key, but you can also explicitly set a number.

If you want an extra space between a mapping key and the colon specify yaml.prefix_colon = ' ':

- https://myurl/abc.tar.xz : 23445
#                         ^ extra space here
- https://myurl/def.tar.xz : 944

If you combine prefix_colon with top_level_colon_align, the top level mapping doesn’t get the extra prefix. If you want that anyway, specify yaml.top_level_colon_align = 12 where 12 has to be an integer that is one more than length of the widest key.

Document version support

In YAML a document version can be explicitly set by using:

%YAML 1.x

before the document start (at the top or before a ---). For ruamel.yaml x has to be 1 or 2. If no explicit version is set version 1.2 is assumed (which has been released in 2009).

The 1.2 version does not support:

  • sexagesimals like 12:34:56
  • octals that start with 0 only: like 012 for number 10 (0o12 is supported by YAML 1.2)
  • Unquoted Yes and On as alternatives for True and No and Off for False.

If you cannot change your YAML files and you need them to load as 1.1 you can load with yaml.version = (1, 1), or the equivalent (version can be a tuple, list or string) yaml.version = "1.1"

If you cannot change your code, stick with ruamel.yaml==0.10.23 and let me know if it would help to be able to set an environment variable.

This does not affect dump as ruamel.yaml never emitted sexagesimals, nor octal numbers, and emitted booleans always as true resp. false

Round trip including comments

The major motivation for this fork is the round-trip capability for comments. The integration of the sources was just an initial step to make this easier.

adding/replacing comments

Starting with version 0.8, you can add/replace comments on block style collections (mappings/sequences resuting in Python dict/list). The basic for for this is:

from __future__ import print_function

import sys
import ruamel.yaml

yaml = ruamel.yaml.YAML()  # defaults to round-trip

inp = """\
abc:
  - a     # comment 1
xyz:
  a: 1    # comment 2
  b: 2
  c: 3
  d: 4
  e: 5
  f: 6 # comment 3
"""

data = yaml.load(inp)
data['abc'].append('b')
data['abc'].yaml_add_eol_comment('comment 4', 1)  # takes column of comment 1
data['xyz'].yaml_add_eol_comment('comment 5', 'c')  # takes column of comment 2
data['xyz'].yaml_add_eol_comment('comment 6', 'e')  # takes column of comment 3
data['xyz'].yaml_add_eol_comment('comment 7', 'd', column=20)

yaml.dump(data, sys.stdout)

Resulting in:

abc:
- a       # comment 1
- b       # comment 4
xyz:
  a: 1    # comment 2
  b: 2
  c: 3    # comment 5
  d: 4              # comment 7
  e: 5 # comment 6
  f: 6 # comment 3

If the comment doesn’t start with ‘#’, this will be added. The key is the element index for list, the actual key for dictionaries. As can be seen from the example, the column to choose for a comment is derived from the previous, next or preceding comment column (picking the first one found).

Config file formats

There are only a few configuration file formats that are easily readable and editable: JSON, INI/ConfigParser, YAML (XML is to cluttered to be called easily readable).

Unfortunately JSON doesn’t support comments, and although there are some solutions with pre-processed filtering of comments, there are no libraries that support round trip updating of such commented files.

INI files support comments, and the excellent ConfigObj library by Foord and Larosa even supports round trip editing with comment preservation, nesting of sections and limited lists (within a value). Retrieval of particular value format is explicit (and extensible).

YAML has basic mapping and sequence structures as well as support for ordered mappings and sets. It supports scalars various types including dates and datetimes (missing in JSON). YAML has comments, but these are normally thrown away.

Block structured YAML is a clean and very human readable format. By extending the Python YAML parser to support round trip preservation of comments, it makes YAML a very good choice for configuration files that are human readable and editable while at the same time interpretable and modifiable by a program.

Extending

There are normally six files involved when extending the roundtrip capabilities: the reader, parser, composer and constructor to go from YAML to Python and the resolver, representer, serializer and emitter to go the other way.

Extending involves keeping extra data around for the next process step, eventuallly resulting in a different Python object (subclass or alternative), that should behave like the original, but on the way from Python to YAML generates the original (or at least something much closer).

Smartening

When you use round-tripping, then the complex data you get are already subclasses of the built-in types. So you can patch in extra methods or override existing ones. Some methods are already included and you can do:

yaml_str = """\
a:
- b:
  c: 42
- d:
    f: 196
  e:
    g: 3.14
"""


data = yaml.load(yaml_str)

assert data.mlget(['a', 1, 'd', 'f'], list_ok=True) == 196