jsonify emits NaN and Infinity, which are not valid JSON

[PranavAchar01]

jsonify serialises float("nan") and float("inf") as the bare literals NaN and Infinity. Those are not valid JSON (RFC 8259 allows only decimal numbers), so the response is served as application/json but cannot be parsed by a strict parser, including the browser's JSON.parse.

This inherits from json.dumps, whose allow_nan defaults to True. DefaultJSONProvider sets default, ensure_ascii and sort_keys, but never touches allow_nan, so the permissive default applies.

The path to it is ordinary: any average, ratio or division that can meet an empty input produces nan without anything looking wrong on the server.

Replication:

from flask import Flask, jsonify

app = Flask(__name__)

@app.route("/stats")
def stats():
    values = []
    mean = sum(values) / len(values) if values else float("nan")
    return jsonify({"count": len(values), "mean": mean})

with app.test_client() as c:
    r = c.get("/stats")
    print(r.headers["Content-Type"])      # application/json
    print(r.get_data(as_text=True))       # {"count":0,"mean":NaN}

The response advertises application/json with a 200, but the body is not JSON:

import json
json.loads('{"count":0,"mean":NaN}')      # accepted, Python is permissive by default
json.loads('{"count":0,"mean":NaN}',
           parse_constant=lambda x: (_ for _ in ()).throw(ValueError(x)))
# ValueError: NaN     <- what a strict parser, and JSON.parse in a browser, does

So the failure surfaces in the client rather than the server, which makes it awkward to trace back.

Expected: either the same behaviour with it documented, or serialisation raises so the problem shows up server-side, where ValueError: Out of range float values are not JSON compliant: nan names the cause immediately.

There is currently no class attribute for this, although the plumbing already works, since dumps forwards **kwargs to json.dumps:

app.json.dumps({"a": float("nan")}, allow_nan=False)
# ValueError: Out of range float values are not JSON compliant: nan

DefaultJSONProvider exposes sort_keys, ensure_ascii and compact as class attributes, so an allow_nan attribute alongside them would fit the existing pattern and let applications choose, without changing the default for anyone. I could not find NaN mentioned in the DefaultJSONProvider or flask.json.dumps docstrings, so at minimum a documentation note seems worthwhile.

Happy to send a PR for whichever you prefer, an allow_nan provider attribute or just the docs.

Environment:

• Python 3.14
• Flask 3.1.3
• Werkzeug 3.1.8

[Omacrow]

Opened a PR implementing the allow_nan attribute option as suggested #6116

[davidism]

Duplicate of #2832 and others. Be sure to search first.