jiter
jiter copied to clipboard
JSON parsing fails on "lone leading surrogate in hex escape" while normal json.loads don't
Hello,
In out migration to pydantic 2, we found a JSON document that pydantic 1 was able to load and pydantic 2 can't with the error:
Invalid JSON: lone leading surrogate in hex escape at line...
Here's a simple way of reproducing:
import json
from pydantic_core import from_json
data = b'{"test": "text\udce2\udc80\udc99text"}'
print(json.loads(data))
print(from_json(data))
This first print from python's json works:
{'test': 'text\udce2\udc80\udc99text'}
The second one using pydantic_core (used by pydantic2) raises
Traceback (most recent call last):
File "check.py", line 7, in <module>
print(from_json(data))
^^^^^^^^^^^^^^^
ValueError: lone leading surrogate in hex escape at line 1 column 20
Here's some versions
Python 3.12.2
pydantic 2.8.2
pydantic-core 2.20.1
Thank you!
Moving this to jiter.
We need to check what serde-json does.
Serde fails with the same error message:
https://play.rust-lang.org/?version=stable&mode=debug&edition=2021&gist=97bd7df54428c3e668c287b59565cd67
Part of the problem will be that a Python str is allowed to have invalid unicode sequences (see e.g. PEP 383 and the 'surrogateescape' handler) to contain (encoded) arbitrary byte payloads. Decoding to UTF8 (and any UTF8 operations) on these strings will fail.
Rust String data, on the other hand, strictly requires valid UTF8.
Hi, I’m encountering the same issue. Just wanted to follow up and check if there are any updates or progress on this? Thanks for your efforts!
import json
from pydantic import TypeAdapter
adapter = TypeAdapter(str)
adapter.validate_json('"\\u266a"')
# '♪'
adapter.validate_json('"\\ud83c"')
# ValidationError: 1 validation error for str
# Invalid JSON: unexpected end of hex escape at line 1 column 8 [type=json_invalid, input_value='"\\ud83c"', input_type=str]
# For further information visit https://errors.pydantic.dev/2.11/v/json_invalid
# Temporary workaround
adapter.validate_python(json.loads('"\\ud83c"'))
# '\ud83c'
+1 Any fixes for this?