config-cascade-merge

A small, schema-driven YAML object merger and validator written in Python. It loads a base schema, validates and applies ordered overlay operations, and emits the complete configuration.

Installation

config-cascade-merge requires Python 3.10 or newer.

python -m pip install config-cascade-merge

You can also install the latest development version directly from GitHub:

python -m pip install git+https://github.com/149segolte/config-cascade-merge.git

Quick start

config-cascade-merge --base_config base.yaml --overlays_dir overlays/

The complete configuration object is written to standard output as YAML. To select specific overlays and control their order explicitly, pass their paths with --overlays:

config-cascade-merge --base_config base.yaml \
  --overlays overlays/common.yaml overlays/workstation.yaml

--overlays and --overlays_dir are mutually exclusive. Directory contents are processed in lexical filename order; an explicit list is processed in the order given.

The executable also supports short options:

config-cascade-merge -b base.yaml \
  -o overlays/common.yaml overlays/workstation.yaml

The module form is equivalent:

python -m config_cascade_merge -b base.yaml --overlays_dir overlays/

Library usage

Create a reusable Schema, validate each Overlay against it, and compose an immutable MergePlan:

from config_cascade_merge import ConfigError, MergePlan, Overlay, Schema

try:
    schema = Schema.from_file("base.yaml")
    common = Overlay.from_file("overlays/common.yaml", schema)
    workstation = Overlay.from_file("overlays/workstation.yaml", schema)
    plan = MergePlan(schema).with_overlays([common, workstation])
except ConfigError as error:
    print(f"Invalid configuration: {error}")
else:
    config = plan.create_object()
    print(config)

Schema, Overlay, and MergePlan are immutable public objects. Their factories do not configure logging or exit the calling process. Schema, overlay, and execution failures derive from ConfigError, with the more specific SchemaError, OverlayError, and MergeError types available when callers need to distinguish them.

Calling plan.create_object() constructs the default schema-shaped object and applies each overlay in order. Fixed object fields without values are None; maps and lists start empty.

Incremental composition and branching

Composition returns a new plan and leaves prior plans unchanged:

base_plan = MergePlan(schema)
common_plan = base_plan.with_overlay(common)

workstation_plan = common_plan.with_overlay(workstation)
server_plan = common_plan.with_overlay(server)

with_overlays(iterable) appends several overlays in iterable order. The read-only plan.schema and plan.overlays properties make plans easy to inspect and reuse. Schemas expose their optional source; overlays expose their name, optional source, and defensively copied operations.

In-memory inputs

All input objects support file, YAML-text, and already-decoded forms:

overlay = Overlay.from_file(path_to_overlay)
overlay = Overlay.from_yaml(yaml_text)
overlay = Overlay.from_data(
    {
        "name": "runtime",
        "operations": [
            {"action": "set", "path": ".profile.name", "data": "Ada"},
        ],
    },
    schema,
    source="runtime settings",
)

Use Schema.from_file(...), Schema.from_yaml(...), and Schema.from_data(...) for similar behavior. Optional source labels are included in validation errors.

Starting from an existing value

Pass an existing complete configuration with initial:

updated = plan.create_object(initial=existing_config)

The starting value is defensively copied and fully validated before any overlay operation executes. Missing fields, unknown fields, and invalid types fail immediately. Neither successful nor failed execution mutates the supplied object.

Base schema

The base file describes the allowed configuration shape rather than containing configuration values.

# base.yaml
type: object
keys:
    profile:
        type: object
        keys:
            name:
                type: string
            active:
                type: boolean
                optional: true
    packages:
        type: list
        id: name
        value:
            type: object
            keys:
                name:
                    type: string
                version:
                    type: integer
    labels:
        type: map
        value:
            type: string

Schema types

The merge policy is either append (default) or override. An id identifies values for identity-based merging.

Every schema node accepts a boolean optional field, which defaults to false. Complete objects and tagged unions may omit named fields marked optional. Empty plans do not materialize optional object fields, and remove deletes them instead of replacing them with null. optional does not allow an explicit null value. On roots, list items, map values, and union branches the flag is retained for structural equality but has no effect unless the node is later used as a named fixed field. A field used by an object or list id cannot be optional.

Example tagged union:

type: tagged_union
keys:
    label:
        type: string
tag:
    name: kind
    options:
        file:
            path:
                type: string
        service:
            port:
                type: integer
        disabled: null

Overlays

Each overlay has a non-empty name and an ordered list of operations. Paths start with .; . addresses the schema root.

# overlays/10-workstation.yaml
name: workstation
operations:
    - action: set
      path: .profile
      data:
          name: Ada
          active: true

    - action: merge
      path: .packages
      data:
          - name: ruff
            version: 1

    - action: test
      path: .profile.name
      data: Ada
      on_fail: warn
      message: unexpected profile

    - action: remove
      path: .labels.legacy

    - action: clear
      path: .packages

Operations

test.on_fail accepts:

• error — stop execution (default)
• warn — report the optional message and continue
• skip — keep prior operations from this overlay and skip the remainder
• drop — discard all operations from this overlay

These failure behaviors are applied to each explicit Overlay group while creating the object. drop rolls back earlier changes from that overlay, while skip preserves earlier changes and skips its remaining operations. Separate overlay documents remain separate groups even when they use the same name.

Validation and errors

The validator rejects malformed YAML, invalid schemas, unknown paths or fields, incompatible values, and unsupported operations. Errors include source filenames and line numbers when available:

overlays/10-workstation.yaml:8: Data at '.packages[0].version' must be integer, got str

The CLI exits with status 1 for invalid paths, schemas, overlays, or empty schema documents.

Development

Clone the repository, install development dependencies, and run the test suite:

git clone https://github.com/149segolte/config-cascade-merge.git
cd config-cascade-merge
uv sync --dev
uv run pytest

Build both source and wheel distributions:

uv build

Project layout:

src/config_cascade_merge/api.py         high-level library API
src/config_cascade_merge/cli.py         CLI entry point
src/config_cascade_merge/engine.py      merge-plan execution
src/config_cascade_merge/schema.py      schema parsing and normalization
src/config_cascade_merge/overlay.py     overlay loading and validation
src/config_cascade_merge/yaml_loader.py YAML loading with source locations
tests/                                  pytest test suite

See CONTRIBUTING.md for the contribution and release process.

License

Licensed under the Mozilla Public License 2.0.