jsonwhy 0.2.0

Explain exactly why and where Python objects fail JSON serialization

jsonwhy

Python's JSON encoder tells you which type failed, but not where that value is inside a large payload. jsonwhy adds the missing path and reports other problems it finds in the same pass.

from datetime import datetime

import jsonwhy

payload = {
    "users": [{"name": "Ada", "joined": datetime(2026, 7, 12)}],
    "tags": {"python", "json"},
}

jsonwhy.dumps(payload)

jsonwhy.JsonWhyError: JSON serialization failed with 2 issues:
1. $.users[0].joined: Object of type datetime is not JSON serializable.
   Fix: Convert it with value.isoformat(); preserve timezone information.
2. $.tags: Object of type set is not JSON serializable.
   Fix: Convert it to a list; sort first if deterministic output matters.

Install

python -m pip install jsonwhy

Python 3.10 or newer is required. The package has no runtime dependencies.

API

dumps() and dump() accept the standard json options. Valid input is passed through to the standard library; diagnosis only runs after serialization fails.

encoded = jsonwhy.dumps({"ok": True}, indent=2)

with open("data.json", "w", encoding="utf-8") as output:
    jsonwhy.dump({"ok": True}, output)

Use explain() when you want structured results without raising an exception:

issues = jsonwhy.explain(payload)

for issue in issues:
    print(issue.path)
    print(issue.json_pointer)
    print(issue.kind)
    print(issue.suggestion)
    print(issue.as_dict())

Each issue includes both a readable JSONPath-like path such as $.users[0].joined and an RFC 6901 json_pointer such as /users/0/joined. A pointer is None when the location contains a Python key that cannot exist in a JSON object.

Two shorter checks are also available:

jsonwhy.check(payload)                 # bool
jsonwhy.assert_serializable(payload)   # raises JsonWhyError

jsonwhy detects unsupported values and keys, circular references, non-finite floats when allow_nan=False, failures in custom default handlers, and overly deep structures.

Custom suggestions

Applications can register advice for their own types:

class CustomerId:
    pass


jsonwhy.register_suggestion(CustomerId, "Convert it with str(value).")

This changes the suggested fix only. jsonwhy does not silently convert the value.

Command line

The CLI checks Python literals, which is useful for values such as sets, bytes, complex numbers, and tuple keys.

jsonwhy "{'payload': b'hello', 'tags': {'a', 'b'}}"
jsonwhy --json "{'payload': b'hello'}"
jsonwhy --path-style pointer "{'payload': b'hello'}"
jsonwhy --max-issues 10 --max-depth 200 "{'payload': b'hello'}"

Exit status 0 means compatible, 1 means issues were found, and 2 means the input or command was invalid.

Notes

• A failing dump() may have written part of its output before the error.
• Diagnosis may call a custom default handler or construct a custom encoder again after the original failure. These hooks should be safe to repeat.
• Diagnostic text can contain representations of application values. Treat it with the same care as logs and tracebacks.
• Custom repr() implementations can perform work or have side effects before their resulting text is shortened.

Bug reports are welcome on the GitHub issue tracker.

License

MIT