Skip to content
Switch branches/tags
Go to file
Cannot retrieve contributors at this time


glom is a growing library! This document is a record of its growth.

The glom team's approach to updates can be summed up as:

  • Always maintaining backwards compatibility
  • CalVer versioning scheme (YY.MM.MICRO)
  • Stay streamlined. glom should be a well-designed bicycle, not an aircraft carrier.

Check this page when upgrading, we strive to keep the updates summarized and well-linked.


(January 24, 2022)

Adds optional * and ** expansion to the default glom access (i.e., glom.glom('a.b.*') will get all items under b as a list). ** is similar, except recursive.

This functionality is off by default in this version, as it technically would break code using '*' and '**' as plain keys. Note that this behavior will break in the next couple months, and an in-library warning has been added to this effect.

Set glom.core.PATH_STAR = True to enable the new feature and take it for a spin!


(November 3, 2020)

  • The T object now supports arithmetic operators!
  • A few internal performance improvements around chainmap, scope, and string path eval.


(August 11, 2020)

A few cleanups from the big release:

  • PathAccessError paths are now Path objects again (fixes #185)
  • Fix an intermittent infinite recursion due to a suspected bug in abc, and a bad interaction between ChainMap and weakref. (fixes #189)
  • Adjust face requirement version to be more strictly up-to-date (#188)
  • Soft launch the Pipe() spec, which is just a more explicit version of the default tuple behavior. (#191)

Huge thanks to all reporters and committers!


(July 31, 2020)

glom releases don't come bigger than this.

  • Match mode - Data validation and pattern matching. You're going to have to read the docs to believe it.
  • Complete documenation refactor
  • Traceback simplification and Target-spec trace
    • glom no longer displays glom-internal stack frames, which were not useful for developers debugging specs and data.
    • Instead, the exception message displays summarized intermediate targets and specs, to help drill down to the part of the data or spec which raised the error.
    • To restore previous behavior, or if you need to see internal stack frames, pass glom_debug=True to the top-level glom() call, or set the GLOM_DEBUG env var to 1.
    • For more details see the debugging guide.
  • Scope updates
    • S() and A (replaces the previously soft-launched Let())
    • Vars(): Non-local scopes
  • T.__() method for accessing dunder attributes (see applicable note under T)
  • glom.__version__ - importable version attribute


(May 2, 2020)

  • Added delete() and Delete(), for deletion of keys and attributes
  • Added Ref() specifier type for recursive specs
  • Added Group() for structural grouping, like a souped-up bucketize
  • Some repr improvements
  • CLI and testing upgrades


(October 29, 2019)

  • Add streaming support via Iter (#100)
  • Add better callable handling with Invoke (#101)
  • Add Fill (#110) - Also adds Auto for the default mode. (soft-launch, docs TBA)
  • Add Let for variable capture (#108) (soft-launch, docs TBA)
  • Steps in the tuple now nest scopes (#97)
  • All public specifier types now have reasonable reprs (notably, Coalesce, Check, and Assign)


(February 17, 2019)

Add Merge() spec and merge() convenience function. for turning iterables of mappings into a single mapping.

Additionally, T and Spec() instances which appear in the "key" portion of the default dict/mapping spec, are now evaluated, enabling operations like this, which demonstrates both new features:

from glom import glom, T, Merge

target = [{"id": 1, "name": "foo"}, {"id": 2, "name": "bar"}]
spec = Merge([{T["id"]: T["name"]}])

glom(target, spec)
# {1: 'foo', 2: 'bar'}


(January 20, 2019)

Added features related to folding/reducing sequences. Read more about Fold, Sum, Flatten, and flatten here.

Also switched CalVer version scheme to YY.MM.MICRO.


(December 25, 2018)

A couple features related to assign() and other minor additions and fixes.

  • Add new missing parameter to assign(), to autogenerate new datastructures at paths that don't exist. Read more at the bottom of the Assign spec docstring.
  • Allow Assign to operate on S-based specs to assign to the spec.
  • Add the STOP singleton. STOP is to SKIP what break is to continue. Useful as a default with conditional specs like Check().


(August 22, 2018)

Fix a small bug where Assign() raised a TypeError instead of a GlomError subtype. This release added PathAssignError, which is now raised instead.


(August 14, 2018)

This release introduces the Assign Spec type, and its accompanying glom.assign() "deep-set" convenience function, a feature that required the refinement of glom's Extension API. Path and T also saw improvements.

  • Assign Spec type and glom.assign() top-level function for deep setting.
  • Extensions (and advanced users) can now register new operations. For instance, Assign registers "assign", which is now a peer of "get" and "iterate", which were the only built-in operations glom provided.
  • Extensions no longer need to be registered. A glom extension is an instance of any type that provides a glomit() method. Full docs coming soon.
  • T and other TType instances now pickle correctly, fixing #48
  • Path instances now behave like strings, with indexing returning new Path objects, with full slice syntax support.
  • Path also supports .values() and .items() methods, which enable getting sequences of the data backing the Path(), for when a sub-Path object is not desirable.
  • Path objects are now comparable for equality. To compare a T, simply wrap it in a Path() use the == operator.


(July 5, 2018)

This sizable release incorporates a lot of the post-announcement feedback. Several advanced features were added, including an extension API and "Scope" support. While this involved a large refactor, all external APIs are 100% backwards-compatible.

In other good news, coverage is up over 90% and on track to go even higher. Check out all the well-tested enhancements below!

  • Introduce the glom's first-class Extension system. Docs are a work in progress, but given that all the internals are implemented in terms of the system, don't hesitate to look under the hood and start experimenting. Addresses #9.
  • The Extension system involved a runtime refactor to add a concept of "scope" to glom. Until now, glom has only supported operating on a single "target", making it the only object in scope. Now, it's also possible to add other objects to the scope, making multi-target glomming a reality for advanced users.
  • Add initial version of Check(), a specifier type aimed at providing target validation while glomming. Addresses #7.
  • Make Spec work for object-oriented glom use. Usage is similar to Python's re.compile: predefine a Spec object, call spec.glom(target) later. Addresses #14
  • Add YAML support to CLI. PyYAML was added as an "extras" dependency and is not installed by default. glom will pick up any existing installed yaml library and use that, or you can pip install glom[yaml] to explicitly install support. See #26 for details. Thanks @dfezzie!
  • Made T and Path share evaluation, repr, and exception paths, closing #6
  • add default_factory argument to Coalesce, with semantics identical to Python's built-in defaultdict.


(May 9, 2018)

Initial release.

See the announcement post for an idea of the functionality available.

This release itself added the console interface (CLI).


(May 6, 2018)

Still in semi-public beta. This release added T objects and better error messages.


(April 30, 2018)

Semi-public beta release. Most of the functionality glom had on announcement.


(April 18, 2018)

Barely more than a function definition. For the historically-oriented, here's a blast from the past, representing the core of glom functionality.