cudf.read_json returns ValueError when given file is not found

[davidwendt]

Describe the bug
The cudf.read_json reports an incorrect error when given an invalid file path.

Steps/Code to reproduce bug

>>> import cudf
>>> df = cudf.read_json("nosuchfile.json")
/conda/envs/rapids/lib/python3.10/site-packages/cudf-23.6.0-py3.10-linux-x86_64.egg/cudf/io/json.py:121: UserWarning: Using CPU via Pandas to read JSON dataset, this may be GPU accelerated in the future
  warnings.warn(
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/conda/envs/rapids/lib/python3.10/site-packages/cudf-23.6.0-py3.10-linux-x86_64.egg/cudf/io/json.py", line 143, in read_json
    pd_value = pd.read_json(
  File "/conda/envs/rapids/lib/python3.10/site-packages/pandas/util/_decorators.py", line 211, in wrapper
    return func(*args, **kwargs)
  File "/conda/envs/rapids/lib/python3.10/site-packages/pandas/util/_decorators.py", line 331, in wrapper
    return func(*args, **kwargs)
  File "/conda/envs/rapids/lib/python3.10/site-packages/pandas/io/json/_json.py", line 757, in read_json
    return json_reader.read()
  File "/conda/envs/rapids/lib/python3.10/site-packages/pandas/io/json/_json.py", line 915, in read
    obj = self._get_object_parser(self.data)
  File "/conda/envs/rapids/lib/python3.10/site-packages/pandas/io/json/_json.py", line 937, in _get_object_parser
    obj = FrameParser(json, **kwargs).parse()
  File "/conda/envs/rapids/lib/python3.10/site-packages/pandas/io/json/_json.py", line 1064, in parse
    self._parse_no_numpy()
  File "/conda/envs/rapids/lib/python3.10/site-packages/pandas/io/json/_json.py", line 1321, in _parse_no_numpy
    loads(json, precise_float=self.precise_float), dtype=None
ValueError: Unexpected character found when decoding 'null'

Using engine="cudf" gives the following result

>>> df = cudf.read_json("nosuchfile.json", engine="cudf")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/conda/envs/rapids/lib/python3.10/site-packages/cudf-23.6.0-py3.10-linux-x86_64.egg/cudf/io/json.py", line 111, in read_json
    df = libjson.read_json(
  File "json.pyx", line 50, in cudf._lib.json.read_json
  File "json.pyx", line 138, in cudf._lib.json.read_json
RuntimeError: CUDF failure at: /cudf/cpp/src/io/json/json_column.cu:958: Input needs to be an array of arrays or an array of (nested) objects

I spent way too much time trying to debug my json file format until I realized I had a typo in the path name.

Expected behavior
The other cudf.read_* functions raise FileNotFoundError if the file cannot be found

[wence-]

This is partly a consequence of pandas api compatibility, for some reason pandas read_json is special compared to all other read_XXX functions in that the input is allowed to not just name a path or be a file-like object, but also be a "VALID json string" which is treated literally.

There have been a few discussions about this on the pandas side, see pandas-dev/pandas#29102, a rejected PR to fix this pandas-dev/pandas#29104, and a PR that partially fixed it pandas-dev/pandas#46718 (by pattern-matching the input string against some subset of likely extensions that indicate a file).

I think this is a niche use-case (although happy to be corrected) and think we should go ahead and deprecate this functionality (if one wants to load json data directly from a string, it is not so difficult to wrap it in an io.StringIO).

I've opened pandas-dev/pandas#52271 to see if a deprecation could be made in pandas too.

[davidwendt]

I believe cudf.read_csv also accepts a string as input but probably must be wrapped in StringIO.
It seems reasonable for us to support read_json in the same way.

[GregoryKimball]

It's worth pointing out that I used string input extensively in my JSON blog post and user guide. 😱

[GregoryKimball]

It looks like pandas will indeed deprecate the raw string option. Watch this space:
pandas-dev/pandas#53409

[vyasr]

It looks like pandas has merged the linked PR above, so we can also stop accepting literal JSON here. @mroeschke any objections?

[mroeschke]

Agreed. cudf.read_json (and all other readers) should stop accepting any literal string data