dazzle-lib

The DazzleLib stack's bedrock: shared Protocols, TypedDict payload schemas, the exception root, and the pure stdlib primitives (Continuum + the state system) every dazzle-* tool composes on.

Every dazzle-* library (the stack) builds on this package: it defines what stack objects can be expected to do (view themselves, serialize themselves), what shapes cross-layer payloads have, and (as of v0.3.x) the pure computational primitives they compose on.

Types + pure primitives -- by charter this package is stdlib-only and SIDE-EFFECT-FREE forever (no I/O, no path handling, no platform probing, no subprocess). Pure computation over data (a Continuum's ordering / stepping / slicing) is in-charter; side effects never are. (The charter evolved 0.2 -> 0.3 from "types only" to "types + pure primitives" -- the stack needs shared composable primitives, not only contract types. The hard guarantees, enforced by tests/test_charter.py, are unchanged.)

pip install dazzle-lib

What's inside (all of it)

New to the Continuum model? The ladder explains the four base objects (Unified / Groupable / Continuum / ContinuumSpace), the implicit-carry chain, the signed invariant-bearing zero, and aligned-vs-product composition.

The idea: the dict is the interface

Rich objects in upper layers know how to become plain dicts; lower-layer functions take and return those dicts. The TypedDicts here are the agreed shapes, so a manifest object in dazzle-preservelib and a metadata collector in dazzle-filekit speak the same payload without sharing a class hierarchy:

from dataclasses import dataclass
from dazzle_lib import DazzleDataMixin, Serializable, FileMetadataDict

@dataclass
class TransferResult(DazzleDataMixin):
    SCHEMA_VERSION = 1
    path: str
    metadata: FileMetadataDict

    def to_dict(self):
        return {"schema_version": self.SCHEMA_VERSION,
                "path": self.path, "metadata": dict(self.metadata)}

    @classmethod
    def from_dict(cls, data):
        # from_dict is the boundary -- VALIDATE here, don't just reconstruct.
        # A bare cls(**data) re-imports the "nothing to validate against" bug.
        return cls(path=data["path"], metadata=data["metadata"])

result = TransferResult("a.txt", {"mode": 0o644, "size": 10, "timestamps": {}})
assert isinstance(result, Serializable)   # structural -- no subclassing needed
print(result.summary())                   # one-liner for logs

And one catchable root for the whole stack:

from dazzle_lib import DazzleError
try:
    ...  # any dazzle-* library call
except DazzleError as e:
    ...  # caught, whichever layer raised it

Dict at the boundary, typed object inside

"The dict is the interface" is a boundary rule, not a working-representation rule. Confuse the two -- start computing against the dict instead of just handing it across a seam -- and you give up the schema that makes a mistake surface at the boundary instead of silently, three layers downstream. The plain dict is the wire format between independently-versioned libraries, so dazzle-filekit and dazzle-preservelib exchange a FileMetadataDict without sharing a class hierarchy. It is not an invitation to thread raw dicts through a library's internals: each side reconstructs its OWN typed object at from_dict and works with that. Pass a dict across the boundary; never reach for d["key"] deep in your code.

Two consequences, stated outright because the stack learned them the hard way:

• from_dict should VALIDATE, not just reconstruct. A TypedDict is statically-checkable but has no runtime enforcement -- at runtime it is a plain dict. The boundary is the safety net: from_dict is where an incoming payload becomes a checked typed object. A bare cls(**data) reintroduces exactly the "nothing to validate against, miss it downstream" failure that typed objects exist to prevent. (Reference pattern: dazzlecmd reconstructs every entity through Pydantic validate_python at its manifest boundary.)
• The computational primitives are typed objects, not dicts. Continuum, the state-system types (StateAxis / EntityState / Transition), and Receipt are frozen dataclasses -- the dict idiom above is for cross-library payloads, not for the shared machinery. Reach for a dict only at a serialization boundary or a genuinely open/extensible keyspace (validated at construction); reach for a typed object everywhere else.

The charter (enforced by tests)

This package is stdlib-only forever and contains no behavior. tests/test_charter.py fails on any banned import (os, shutil, pathlib, subprocess, ...) anywhere in the package -- a PR that needs to weaken that test is adding something that belongs in a higher layer. Admission follows the rule of two: a Protocol or TypedDict enters the bedrock only when two or more stack libraries need it.

The stack

Full architecture contract: STACK-MAP.md. API stability policy: docs/api-stability.md.

Contributing

Contributions welcome! Please open an issue or submit a pull request.

pip install -e ".[dev]"
python -m pytest tests/ -v

Before proposing additions, note the two house rules this package lives by:

• The charter: types only -- no I/O, no path handling, no behavior (tests/test_charter.py enforces it; a PR that needs to weaken that test belongs in a higher layer)
• The rule of two: a Protocol or TypedDict enters the bedrock only when two or more stack libraries need it
• API changes follow docs/api-stability.md (locked surface, noisy-shim deprecation policy)

Like the project?

License

This project is licensed under the MIT License - see the LICENSE file for details. The bedrock sits beneath MIT and GPL stack members alike, so it carries the permissive license.