ruamel.yaml

ruamel.yaml is a YAML 1.2 loader/dumper package for Python.

breaking changes, that may make future uploads to PyPI impossible

If you are interested in future upgrades of ruamel.yaml
please check the documentation on installing,
since at some point I might not be able to upload a new version to PyPI with updated information.

ruamel.yaml was intentionally named as yaml in a namespace ruamel. The namespace allows the installation
name to correspond unchanged to how the package is imported, reduces the number of links I have to create
in site-packages of a Python install during development, as well as providing a recognisable set of packages
my company releases to the public.

However, after uploading version 0.18.7, I got an email from PyPI, about having to change the project name
to ruamel_yaml to comply with PEP 625, sometime in the future. The email doesn't say if namespace packages are
no longer allowed, or how to deal with the very real clash with the pre-existing package ruamel_yaml.

I might not be able to adapt ruamel.yaml, in
a way that does not negatively affect the 0.5 million daily downloads (and my own usage of the package) in time.
My experience with other such service downgrades (Bitbucket, Readthedocs), has not been entirely positive.

This is probably the last patch version of ruamel.yaml that installs pre-compiled
ruamel.yaml.clib. This version already supports loading of the C extension
compiled using the Zig toolchain
(i.e. without you having to have a compiler environment pre-installed).

You can test this by invoking (note the final z):

python -m pip install --no-deps ruamel.yaml ruamel.yaml.clibz

You can however install both ruamel.yaml.clib and ruamel.yaml.clibz, in
which case the latter is taken (e.g. by dropping --no-deps from the invocation)

The move to using zig eliminates having to generate the 60 or so .whl files
(partly on unreliable services like AppVeyor and Github). Apart from that
the compilation on the target machine can fully optimize for that
architecture and its OS.

But the test matrix still has many dimensions:

    Python versions: 3.9 - 3.14
    OS-es:           Linux, Alpine (musl), macOS, Windows
    Architectures:   Intel/AMD, Arm (and others), in 64 and some also in 32 bit versions
    Zig version:     ziglang < 0.16 is taken from PyPI

I try to test as much as possible, multiple Python versions were macOS-Arm and Ubuntu Linux-Intel.
And at least one Python version along each of the indicated positions of the
dimensions above (e.g. Windows10 64bit was tested, but not the RISC-V architecture).

In 0.19 the loading mechanism will be reversed: defaulting to installing and using
ruamel.yaml.clibz unless ruamel.yaml.clib is installed. This backwards
compatibility will be dropped at the end of the 0.19 series.
Pin the version that you are relying on, until you can test the changes to work for you.

As announced, in 0.18.0, the old PyYAML functions have been deprecated.
(scan, parse, compose, load, emit, serialize, dump and their variants
(_all, safe_, round_trip_, etc)). If you only read this after your program has
stopped working: I am sorry to hear that, but that also means you, or the person
developing your program, has not tested with warnings on (which is the recommendation
in PEP 565, and e.g. defaulting when using pytest). If you have troubles, explicitly use

pip install "ruamel.yaml<0.18.0"

or put something to that effects in your requirments, to give yourself
some time to solve the issue.

There will be at least one more potentially breaking change in the 0.18 series: YAML(typ='unsafe')
now has a pending deprecation warning and is going to be deprecated, probably before the end of 2023.
If you only use it to dump, please use the new YAML(typ='full'), the result of that can be safely
loaded with a default instance YAML(), as that will get you inspectable, tagged, scalars, instead of
executed Python functions/classes. (You should probably add constructors for what you actually need,
but I do consider adding a ruamel.yaml.unsafe package that will re-add the typ='unsafe' option.
Please adjust/pin your dependencies accordingly if necessary.

There seems to be a CVE on ruamel.yaml, stating that the load() function could be abused
because of unchecked input. load() was never the default function (that was round_trip_load()
before the new API came into existence. So the creator of that CVE was ill informed and
probably lazily assumed that since ruamel.yaml is a derivative of PyYAML (for which
a similar CVE exists), the same problem would still exist, without checking.
So the CVE was always inappropriate, now just more so, as the call
to the function load() with any input will terminate your program with an error message. If you
(have to) care about such things as this CVE, my recommendation is to stop using Python
completely, as pickle.load() can be abused in the same way as load() (and like unlike load()
is only documented to be unsafe, without development-time warning.

Version 0.18.9 was the last one tested to be working with Python 3.7
Version 0.17.21 was the last one tested to be working on Python 3.5 and 3.6

The 0.16.13 release was the last that was tested to be working on Python 2.7.

There are two extra plug-in packages
(ruamel.yaml.bytes and ruamel.yaml.string)
for those not wanting to do the streaming to a
io.BytesIO/StringIO buffer themselves.

If your package uses ruamel.yaml and is not listed on PyPI, drop me an
email, preferably with some information on how you use the package (or a
link to the repository) and I'll keep you informed when the status of
the API is stable enough to make the transition.

For packaging purposes you can use a download of the tar balls of tagged source

0.18.17 (2025-12-17):

• try to load C functions from _ruamel_yaml_clibz first.

0.18.16 (2025-10-22):

• root level block style scalars that started with a directives-end marker or a document-end marker, are now indented 2 spaces.
• merged fix for accessing end_marks on Tokens provided by Toknak

0.18.15 (2025-08-19):

• duplicate merge keys are never allowed (not even with .allow_duplicate_keys = True
• merge keys now keep there position if a key before the merge key gets deleted (previously a key after the merge key would move before it)

0.18.14 (2025-06-09):

• Fix issue with constructing dataclasses that have a default factoryi attribute, but were missing a mapping value for that attribute. Reported by Victor Prieto
• the tagged release tar files can now also be downloaded from https://yaml.dev/ruamel-dl-tagged-releases/ please adjust if you use https://sourceforge.net/projects/ruamel-dl-tagged-releases/files/ as that repository in sourceforge will no longer be updated from some later date.

0.18.13 (2025-06-06):

• Fix line wrapping on plain scalars not observing width correctly. Issue 529, reported by Sebastien Vermeille
• Fix sha256 and length in RECORD files. Reported by Evan

0.18.12 (2025-05-30):

• fix additional issue with extra space in double quoted string. Reported by Saugat Pachhai
• fix duplicate key url, now pointing to yaml.dev. Reported by Hugo
• fix broken RECORD file, which was a problem for uv, not pip. Reported by konstin

0.18.11 (2025-05-19):

• function load_yaml_guess_indent now takes an option yaml argument so you can provide an already created/configured YAML instance
• Sequence item indicator with both comment/empty line before indicator and comment before sequence item, could not move comment and raise NotImplementedError. Reported by Karsten Tessarzik.
• missing f for f-string (reported by π, via email)
• fixed issue with extra space in double quoted dump (reported by Jan Möller)

0.18.10 (2025-01-06):

• implemented changes to the setup.py for Python 3.14 as suggested by Miro Hrončok in merge requests (MR not merged as those files are copied in from develop config)

0.18.9 (2025-01-05):

• fix issue with roundtripping 0 in YAML 1.1 reported by Peter Law

0.18.8 (2025-01-02):

• added warning to README.md that PyPI might block updates due to breaking changes

0.18.7 (2024-12-30):

• fixes for README (reported by Kees Bakker)
• fixes preserving anchor on scalar integer 0 (issue reported by (Mor Peled)[https://sourceforge.net/u/morp/profile/] and also in a question by Ravi on Stackoverflow)
• fix for formatting of README suggested by Michael R. Crusoe

0.18.6 (2024-02-07):

• fixed an issue with dataclass loading when the fields were collections (bug found as a result of a question by FibroMyAlgebra on StackOverflow)
• fixed an issue loading dataclasses with InitVar fields when from __future__ import annotations was used to delay evaluation of typing.

0.18.5 (2023-11-03):

• there is some indication that dependent packages have been pinned to use specific (tested) and just install the latest even in Python versions that have end-of-life

0.18.4 (2023-11-01):

• YAML() instance has a doc_infos attribute which is a cumulative list of DocInfo instances (one for load(), one per document for load_all()). DocInfo instances contain version information (requested, directive) and tag directive information
• fix issue that the YAML instance tags attribute was not reset between documents, resulting in mixing of tag directives of multiple documents. Now only provides tag directive information on latest document after loading. This means tags for dumping must be set again after a document is loaded with the same instance. (because of this tags will be removed in a favour of a different mechanism in the future)
• fix issue with multiple document intermixing YAML 1.2 and YAML 1.1, the VersionedResolver now resets
• fix issue with disappearing comment when next token was Tag (still can't have both a comment before a tag and after a tag, before node)

0.18.3 (2023-10-29):

• fix issue with spurious newline on first item after comment + nested block sequence
• additional links in the metadata on PyPI (Reported, with pointers how to fix, by Sorin).

0.18.2 (2023-10-24):

• calling the deprecated functions now raises an AttributeError with the, somewhat more informative, orginal warning message. Instead of calling sys.exit(1)

0.18.1 (2023-10-24):

• calling the deprecated functions now always displays the warning message. (reported by Trend Lloyd)

0.18.0 (2023-10-23):

• the functions scan, parse, compose, load, emit, serialize, dump and their variants (_all, safe_, round_trip_, etc) have been deprecated (the same named methods on YAML() instances are, of course, still there.
• YAML(typ='unsafe') now issues a PendingDeprecationWarning. This will become deprecated in the 0.18 series
  (probably before the end of 2023).
  You can use YAML(typ='full') to dump unregistered Python classes/functions.
  For loading you'll have to register your classes/functions
  if you want the old, unsafe, functionality. You can still load any tag, like `!!python/name:posix.system', safely
  with the (default) round-trip parser.
• fix for bytes-like object is required not 'str' while dumping binary streams. This was reported, analysed and a fix provided by Vit Zikmund

0.17.40 (2023-10-20):

• flow style sets are now preserved ( !!set {a, b, c} ). Any values specified when loading are dropped, including !!null "".
• potential workaround for issue 484: the long_description_content_type including the variant specification CommonMark
  can result in problems on Azure. If you can install from .tar.gz using
  RUAMEL_NO_LONG_DESCRIPTION=1 pip install ruamel.yaml --no-binary :all: then the long description, and its
  offending type, are nog included (in the METADATA).
  (Reported by Coury Ditch)
• links in documentation update (reported by David Hoese)
• Added some __repr__ for internally used classes

0.17.39 (2023-10-19):

• update README generation, no code changes

0.17.36 (2023-10-19):

• fixed issue 480, dumping of a loaded empty flow-style mapping with comment failed (Reported by Stéphane Brunner)
• fixed issue 482, caused by DEFAULT_MAPPING_TAG having changes to being a Tag() instance, not a string (reported by yan12125)
• updated documentation to use mkdocs

0.17.35 (2023-10-04):

• support for loading dataclasses with InitVar variables (some special coding was necessary to get the, unexecpected, default value in the corresponding instance attribute ( example of usage in this question)

0.17.34 (2023-10-03):

• Python 3.12 also loads C version when using typ='safe'
• initial support for loading invoking
  __post_init__() on dataclasses that have that
  method after loading a registered dataclass.
  (Originally
  asked on
  Stackoverflow by
  nyanpasu64
  and as
  ticket by
  Patrick Lehmann)

@yaml.register_class
@dataclass
class ...

0.17.33 (2023-09-28):

• added flow_seq_start, flow_seq_end, flow_seq_separator, flow_map_start, flow_map_end, flow_map_separator class attributes to the Emitter class so flow style output can more easily be influenced (based on this answer on a StackOverflow question by Huw Walters).

0.17.32 (2023-06-17):

• fix issue with scanner getting stuck in infinite loop

0.17.31 (2023-05-31):

• added tag.setter on ScalarEvent and on Node, that takes either a Tag instance, or a str (reported by Sorin Sbarnea)

0.17.30 (2023-05-30):

• fix issue 467, caused by Tag instances not being hashable (reported by Douglas Raillard)

0.17.29 (2023-05-30):

• changed the internals of the tag property from a string to a class which allows for preservation of the original handle and suffix. This should result in better results using documents with %TAG directives, as well as preserving URI escapes in tag suffixes.

0.17.28 (2023-05-26):

• fix for issue 464: documents ending with document end marker
  without final newline fail to load (reported by Mariusz
  Rusiniak)

0.17.27 (2023-05-25):

• fix issue with inline mappings as value for merge keys (reported by Sirish on StackOverflow)
• fix for 468, error inserting after accessing merge attribute on CommentedMap (reported by Bastien gerard)
• fix for issue 461 pop + insert on same CommentedMap key throwing error (reported by John Thorvald Wodder II)

0.17.26 (2023-05-09):

• fix for error on edge cage for issue 459

0.17.25 (2023-05-09):

• fix for regression while dumping wrapped strings with too many backslashes removed (issue 459, reported by Lele Gaifax)

0.17.24 (2023-05-06):

• rewrite of CommentedMap.insert(). If you have a merge key in the YAML document for the mapping you insert to, the position value should be the one as you look at the YAML input. This fixes issue 453 where other keys of a merged in mapping would show up after an insert (reported by Alex Miller). It also fixes a call to .insert() resulting into the merge key to move to be the first key if it wasn't already and it is also now possible to insert a key before a merge key (even if the fist key in the mapping).
• fix (in the pure Python implementation including default) for issue 447. (reported by Jack Cherng, also brought up by brent on StackOverflow)

0.17.23 (2023-05-05):

• fix 458, error on plain scalars starting with word longer than width. (reported by Kyle Larose)
• fix for .update() no longer correctly handling keyword arguments (reported by John Lin on StackOverflow)
• fix issue 454: high Unicode (emojis) in quoted strings always
  escaped (reported by Michal
  Čihař based on a
  question on StackOverflow).
• fix issue with emitter conservatively inserting extra backslashes in wrapped quoted strings (reported by thebenman on StackOverflow)

0.17.22 (2023-05-02):

• fix issue 449 where the second exclamation marks got URL encoded (reported and fixing PR provided by John Stark)
• fix issue with indent != 2 and literal scalars with empty first line (reported by wrdis on StackOverflow)
• updated __repr__ of CommentedMap, now that Python's dict is ordered -> no more ordereddict(list-of-tuples)
• merge MR 4, handling OctalInt in YAML 1.1 (provided by Jacob Floyd)
• fix loading of !!float 42 (reported by Eric on Stack overflow)
• line numbers are now set on CommentedKeySeq and CommentedKeyMap (which are created if you have a sequence resp. mapping as the key in a mapping)
• plain scalars: put single words longer than width on a line of
  their own, instead of after the previous line (issue 427, reported
  by Antoine
  Cotten). Caveat:
  this currently results in a space ending the previous line.
• fix for folded scalar part of 421: comments after ">" on first
  line of folded scalars are now preserved (as were those in the
  same position on literal scalars). Issue reported by Jacob Floyd.
• added stacklevel to warnings
• typing changed from Py2 compatible comments to Py3, removed various Py2-isms

0.17.21 (2022-02-12):

• fix bug in calling .compose() method with pathlib.Path instance.

0.17.20 (2022-01-03):

• fix error in microseconds while rounding datetime fractions >= 9999995 (reported by Luis Ferreira)

0.17.19 (2021-12-26):

• fix mypy problems (reported by Arun)

0.17.18 (2021-12-24):

• copy-paste error in folded scalar comment attachment (reported by Stephan Geulette)
• fix 411, indent error comment between key empty seq value (reported by Guillermo Julián)

0.17.17 (2021-10-31):

• extract timestamp matching/creation to util

0.17.16 (2021-08-28):

• 398 also handle issue 397 when comment is newline

0.17.15 (2021-08-28):

• fix issue 397, insert comment before key when a comment between key and value exists (reported by Bastien gerard)

0.17.14 (2021-08-25):

• fix issue 396, inserting key/val in merged-in dictionary (reported by Bastien gerard)

0.17.13 (2021-08-21):

• minor fix in attr handling

0.17.12 (2021-08-21):

• fix issue with anchor on registered class not preserved and those classes using package attrs with @attr.s() (both reported by ssph)

0.17.11 (2021-08-19):

• fix error baseclass for DuplicateKeyError (reported by Łukasz Rogalski)
• fix typo in reader error message, causing KeyError during reader error (reported by MTU)

0.17.10 (2021-06-24):

• fix issue 388, token with old comment structure != two elements (reported by Dimitrios Bariamis)

0.17.9 (2021-06-10):

• fix issue with updating CommentedMap (reported by sri on StackOverflow)

0.17.8 (2021-06-09):

• fix for issue 387 where templated anchors on tagged object did get set resulting in potential id reuse. (reported by Artem Ploujnikov)

0.17.7 (2021-05-31):

• issue 385 also affected other deprecated loaders (reported via email by Oren Watson)

0.17.6 (2021-05-31):

• merged type annotations update provided by Jochen Sprickerhof
• fix for issue 385: deprecated round_trip_loader function not
  working (reported by Mike
  Gouline)
• wasted a few hours getting rid of mypy warnings/errors

0.17.5 (2021-05-30):

• fix for issue 384 !!set with aliased entry resulting in broken YAML on rt reported by William Kimball)

0.17.4 (2021-04-07):

• prevent (empty) comments from throwing assertion error (issue 351 reported by William Kimball) comments (or empty line) will be dropped

0.17.3 (2021-04-07):

• fix for issue 382 caused by an error in a format string (reported by William Kimball)
• allow expansion of aliases by setting yaml.composer.return_alias = lambda s: copy.deepcopy(s)
  (as per Stackoverflow answer)

0.17.2 (2021-03-29):

• change -py2.py3-none-any.whl to -py3-none-any.whl, and remove 0.17.1

0.17.1 (2021-03-29):

• added 'Programming Language :: Python :: 3 :: Only', and
  removing 0.17.0 from PyPI (reported by Alasdair
  Nicol)

0.17.0 (2021-03-26):

• removed because of incomplete classifiers
• this release no longer supports Python 2.7, most if not all Python 2 specific code is removed. The 0.17.x series is the last to support Python 3.5 (this also allowed for removal of the dependency on ruamel.std.pathlib)
• remove Python2 specific code branches and adaptations (u-strings)
• prepare % code for f-strings using _F
• allow PyOxidisation (issue 324 resp. issue 171)
• replaced Python 2 compatible enforcement of keyword arguments with '*'
• the old top level functions load, safe_load, round_trip_load, dump, safe_dump, round_trip_dump, scan, parse, compose, emit, serialize as well as their _all variants for multi-document streams, now issue a PendingDeprecationning (e.g. when run from pytest, but also Python is started with -Wd). Use the methods on YAML(), which have been extended.
• fix for issue 376: indentation changes could put literal/folded
  scalar to start before the # column of a following comment.
  Effectively making the comment part of the scalar in the output.
  (reported by Bence Nagy)

For older changes see the file
CHANGES