revert: fingerprint_data back to canonical-json hash + stringify + callable guard

the custom normalization layer (_fingerprint_normalize / _fingerprint_key /
_fingerprint_default + type-tagged markers + determinism allowlist) guarded against
non-json inputs that never arrive by contract, and in exchange crashed on ordinary
value types (uuid/decimal/path/objectid) and still collided (bytes vs ["bytes",hex]).
worse than the four-liner on both axes, and it regressed across four iterations.

fingerprint_data takes json-shaped record data, so json.dumps(sort_keys=True,
separators, default=str) is the correct tool: order-independent, whitespace-stable,
non-json scalars stringified rather than crashing. the one real footgun default=str
introduces - a callable's str() embeds a memory address, non-deterministic across
processes - is guarded with an explicit TypeError. non-str dict keys still raise from
json natively (str-key contract, matching encrypt_data's own guard and the original).

deletes the whole machinery. closes the fingerprint regressions R1/R2/R3 and the
marker-encoding docstring nit at the root. digest VALUE changes vs the marker-based
version - any persisted fingerprints must re-baseline.

Signed-off-by: disqualifier <dev@disqualifier.me>
This commit is contained in:
2026-07-06 20:55:03 -04:00
parent 4b9820b18e
commit 1edd7ddfea
+12 -68
View File
@@ -630,77 +630,21 @@ def decrypt_record(crypto: EnvelopeCrypto, record, traversal_level: int = 2) ->
return result return result
_DETERMINISTIC_SCALARS = (str, int, float, bool, type(None))
def _fingerprint_default(value: Any) -> Any:
"""json.dumps default= handler for values fingerprint_data can't natively serialize
handles datetime (isoformat) and bytes (hex) deterministically, returning a 2-element
``["type", value]`` list so the encoding is a JSON ARRAY that no genuine str value can
ever equal (a str encodes as a JSON string) - collision-resistant by construction.
anything else - a callable, an arbitrary object, a generator - has no guaranteed stable
representation across processes (its repr usually embeds a memory address), so it fails
loud with a TypeError rather than yield an unstable digest.
"""
if hasattr(value, "isoformat"):
return ["datetime", value.isoformat()]
if isinstance(value, (bytes, bytearray)):
return ["bytes", value.hex()]
raise TypeError(
f"fingerprint_data: {type(value).__name__} has no deterministic representation; "
"serialize it to a stable str/bytes/datetime before fingerprinting"
)
def _fingerprint_key(key: Any) -> str:
"""derive a deterministic, type-qualified tag for a dict key
scalars tag by type + value; set/frozenset keys canonically sort their members; any other
key type (a callable, an identity-repr object, a nested container) has no stable key form
and raises TypeError, mirroring the value-side determinism guard.
"""
if isinstance(key, _DETERMINISTIC_SCALARS):
return f"{type(key).__name__}:{key!r}"
if isinstance(key, (set, frozenset)):
normalized = _fingerprint_normalize(key)
return f"{type(key).__name__}:" + json.dumps(normalized, sort_keys=True, default=_fingerprint_default)
raise TypeError(
f"fingerprint_data: dict key of type {type(key).__name__} has no deterministic "
"representation; use a scalar or set/frozenset key"
)
def _fingerprint_normalize(value: Any) -> Any:
"""recursively normalize dict keys and sets so the encoding is deterministic across processes"""
if isinstance(value, dict):
return {
_fingerprint_key(key): _fingerprint_normalize(sub_value)
for key, sub_value in value.items()
}
if isinstance(value, (set, frozenset)):
members = [_fingerprint_normalize(item) for item in value]
members.sort(key=lambda item: json.dumps(item, sort_keys=True, default=_fingerprint_default))
# keep the concrete type in the marker so a set value and a frozenset value with the
# same members stay distinct; the \x00 prefix keeps it uncollidable with a user str key
return {f"\x00{type(value).__name__}:members": members}
if isinstance(value, (list, tuple)):
return [_fingerprint_normalize(item) for item in value]
return value
def fingerprint_data(data: dict) -> str: def fingerprint_data(data: dict) -> str:
"""return a deterministic, collision-resistant SHA-256 hex fingerprint of a dict """deterministic sha256 fingerprint of json-shaped dict data
keys of different types that would coerce to the same JSON string are kept distinct via the input is a record sent over the wire, so it is json-serializable by contract:
a type-tagged pre-pass; set/frozenset keys and values are canonically sorted so their keys are canonically sorted for an order-independent digest, separators are fixed for
unordered members hash stably; datetime/bytes go through a type-qualified default= handler whitespace stability, and any non-json scalar (uuid/decimal/path/datetime/objectid) is
whose ``\\x00``-prefixed tags cannot collide with a genuine str value. any key or value stringified rather than crashing the encode. a callable value raises - its str() embeds a
with no deterministic representation (a callable, an identity-repr object) raises TypeError memory address, which would make the digest vary across processes. never logs the data.
rather than yield a digest that varies across processes. never logs the data.
""" """
normalized = _fingerprint_normalize(data) def _stringify(value: object) -> str:
encoded = json.dumps(normalized, sort_keys=True, default=_fingerprint_default) if callable(value):
raise TypeError(f"fingerprint_data: cannot fingerprint a callable: {value!r}")
return str(value)
encoded = json.dumps(data, sort_keys=True, separators=(",", ":"), default=_stringify)
return hashlib.sha256(encoded.encode()).hexdigest() return hashlib.sha256(encoded.encode()).hexdigest()