pd.read_json May Not Maintain Numeric String Index

[WillAyd]

>>> df = pd.DataFrame(range(3), index=list("123"))
>>> df.to_json(orient="split")
'{"columns":[0],"index":["1","2","3"],"data":[[0],[1],[2]]}'
>>> pd.read_json(df.to_json(orient="split"), orient="split").index
Int64Index([1, 2, 3], dtype='int64')

Note that the string nature of the values should be preserved via roundtrip here, but ends up being lossy anyway. Noted during refactor of #28510

[jmg7173]

Can I work on this issue?

[jmg7173]

The reason why pd.read_json converts dtype of object index into int64 is as belows.

pandas/pandas/io/json/_json.py

Lines 916 to 923 in f08a1e6

	if data.dtype == "object":
	# try float
	try:
	data = data.astype("float64")
	result = True
	except (TypeError, ValueError):
	pass

Do we need to convert object typed data as coarsed type like float, int64 and so on?

If I comment out this line, it breaks tests so much.

[jmg7173]

How about change default value of convert_axes as False at this condition of read_json?

pandas/pandas/io/json/_json.py

Lines 580 to 581 in c4de906

	if convert_axes is None and orient != "table":
	convert_axes = True

[sathyz]

@jmg7173 are you still looking at this?

[scratchmex]

I'm sorry but, why the strings values should be preserved by default? It isn't more comfortable to parse string numbers as integers?

[jmg7173]

@sathyz No, you can take this issue.

[theoniko]

Hello @WillAyd,
Would it be possible to review my pr and give me feedback?
Thanks in advace.