pd.read_json cryptic error on missing file

[ghost]

In [12]: pd.read_json("No Such File")
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-12-f479cae2f65a> in <module>()
----> 1 pd.read_json("No Such File")

/home/user1/src/pandas/pandas/io/json.pyc in read_json(path_or_buf, orient, typ, dtype, convert_axes, convert_dates, keep_default_dates, numpy, precise_float, date_unit)
    188         obj = FrameParser(json, orient, dtype, convert_axes, convert_dates,
    189                           keep_default_dates, numpy, precise_float,
--> 190                           date_unit).parse()
    191 
    192     if typ == 'series' or obj is None:

/home/user1/src/pandas/pandas/io/json.pyc in parse(self)
    256 
    257         else:
--> 258             self._parse_no_numpy()
    259 
    260         if self.obj is None:

/home/user1/src/pandas/pandas/io/json.pyc in _parse_no_numpy(self)
    473         if orient == "columns":
    474             self.obj = DataFrame(
--> 475                 loads(json, precise_float=self.precise_float), dtype=None)
    476         elif orient == "split":
    477             decoded = dict((str(k), v)

ValueError: Expected object or value

read_json interprets strings which are not filenames as json data, then fails to parse them
if the filename names a path that doesn't exist (due to typo, or being in wrong directory for example).

That overloading makes it impossible nasty to distinguish two distinct error cases, e.g.
missing file and malformed json. Dubious API choice to my tastes.

in any case, catch both errors and return a saner message "missing file or malformed input" etc.

[jreback]

error could be better, but this is legit, the user passed a non-existant file, so it must be a json-string (that cannot be parsed).

[ghost]

Right, what I meant by that description is that I'm aware the behavior is by choice but wish
it hadn't been : "If I make a typo, i must mean JSON" is also what this implies. These
eagerly overloaded signatures often create pathologies where there needn't be any.

Nevermind, that's the API, but a better error message would help.

[jreback]

TLTC (too-late-too-cancel)....

[ghost]

:)

[jreback]

same type of error also will occur with msgpack

[ghost]

Safety in numbers?

Oh, yeah, better error message there too then I guess.

[jreback]

haha

[ghost]

Can't fix similar problem in read_msgpack because

pd.read_msgpack("NoSuchFile")

Decodes just fine.

Can't fix it in read_json either because sometimes the Exception error message
carries valuable info on how the json is broken. Rewriting the exception message
based on the specific error message raise is fugly. No.

Sorry, but that's a really pernicious API.