BUG: pd.read_json May Not Maintain Numeric String Index

Author: theoniko

doc/source/whatsnew/v1.3.0.rst

@@ -267,7 +267,7 @@ I/O
 - Bug in :func:`to_hdf` raising ``KeyError`` when trying to apply
   for subclasses of ``DataFrame`` or ``Series`` (:issue:`33748`).
 - Bug in :func:`json_normalize` resulting in the first element of a generator object not being included in the returned ``DataFrame`` (:issue:`35923`)
-
+- Bug in :func:`read_json` does not maintan numeric string index (:issue:`28556`)
 
 Period
 ^^^^^^

pandas/io/json/_json.py

@@ -26,6 +26,7 @@
 from pandas.core import generic
 from pandas.core.construction import create_series_with_explicit_dtype
 from pandas.core.generic import NDFrame
+from pandas.core.indexes.api import Index
 from pandas.core.reshape.concat import concat
 
 from pandas.io.common import (
@@ -894,14 +895,19 @@ def _try_convert_data(self, name, data, use_dtypes=True, convert_dates=True):
             if result:
                 return new_data, True
 
-        result = False
+        result = True
 
         if data.dtype == "object":
-
+            if (
+                isinstance(data, Index)
+                and name == "index"
+                and self.orient == "split"
+                and len(data)
+            ):
+                result = False
             # try float
             try:
                 data = data.astype("float64")
-                result = True
             except (TypeError, ValueError):
                 pass
 
@@ -912,7 +918,6 @@ def _try_convert_data(self, name, data, use_dtypes=True, convert_dates=True):
                 # coerce floats to 64
                 try:
                     data = data.astype("float64")
-                    result = True
                 except (TypeError, ValueError):
                     pass
 
@@ -924,7 +929,6 @@ def _try_convert_data(self, name, data, use_dtypes=True, convert_dates=True):
                 new_data = data.astype("int64")
                 if (new_data == data).all():
                     data = new_data
-                    result = True
             except (TypeError, ValueError, OverflowError):
                 pass
 
@@ -934,7 +938,6 @@ def _try_convert_data(self, name, data, use_dtypes=True, convert_dates=True):
             # coerce floats to 64
             try:
                 data = data.astype("int64")
-                result = True
             except (TypeError, ValueError):
                 pass
 

pandas/tests/io/json/test_pandas.py

@@ -191,12 +191,13 @@ def test_roundtrip_str_axes(self, orient, convert_axes, numpy, dtype):
         # JSON objects. JSON keys are by definition strings, so there's no way
         # to disambiguate whether those keys actually were strings or numeric
         # beforehand and numeric wins out.
-        # TODO: Split should be able to support this
-        if convert_axes and (orient in ("split", "index", "columns")):
+        if convert_axes and (orient in ("index", "columns")):
             expected.columns = expected.columns.astype(np.int64)
             expected.index = expected.index.astype(np.int64)
         elif orient == "records" and convert_axes:
             expected.columns = expected.columns.astype(np.int64)
+        elif convert_axes and orient == "split":
+            expected.columns = expected.columns.astype(np.int64)
 
         assert_json_roundtrip_equal(result, expected, orient)