Merge remote-tracking branch 'upstream/main' into aijams-take-function-invalid-dtype

Author: aijams

doc/source/whatsnew/v3.0.0.rst

@@ -1087,6 +1087,7 @@ I/O
 - Bug in :meth:`HDFStore.select` causing queries on categorical string columns to return unexpected results (:issue:`57608`)
 - Bug in :meth:`MultiIndex.factorize` incorrectly raising on length-0 indexes (:issue:`57517`)
 - Bug in :meth:`read_csv` causing segmentation fault when ``encoding_errors`` is not a string. (:issue:`59059`)
+- Bug in :meth:`read_csv` for the ``c`` and ``python`` engines where parsing numbers with large exponents caused overflows. Now, numbers with large positive exponents are parsed as ``inf`` or ``-inf`` depending on the sign of the mantissa, while those with large negative exponents are parsed as ``0.0`` (:issue:`62617`, :issue:`38794`, :issue:`62740`)
 - Bug in :meth:`read_csv` raising ``TypeError`` when ``index_col`` is specified and ``na_values`` is a dict containing the key ``None``. (:issue:`57547`)
 - Bug in :meth:`read_csv` raising ``TypeError`` when ``nrows`` and ``iterator`` are specified without specifying a ``chunksize``. (:issue:`59079`)
 - Bug in :meth:`read_csv` where the order of the ``na_values`` makes an inconsistency when ``na_values`` is a list non-string values. (:issue:`59303`)

pandas/_config/config.py

@@ -944,3 +944,11 @@ def is_callable(obj: object) -> bool:
     if not callable(obj):
         raise ValueError("Value must be a callable")
     return True
+
+
+# import set_module here would cause circular import
+get_option.__module__ = "pandas"
+set_option.__module__ = "pandas"
+describe_option.__module__ = "pandas"
+reset_option.__module__ = "pandas"
+option_context.__module__ = "pandas"

pandas/_libs/lib.pyx

@@ -41,6 +41,7 @@ from cython cimport (
 from pandas._config import using_string_dtype
 
 from pandas._libs.missing import check_na_tuples_nonequal
+from pandas.util._decorators import set_module
 
 import_datetime()
 
@@ -154,6 +155,7 @@ def memory_usage_of_objects(arr: object[:]) -> int64_t:
 # ----------------------------------------------------------------------
 
 
+@set_module("pandas.api.types")
 def is_scalar(val: object) -> bool:
     """
     Return True if given object is scalar.
@@ -255,6 +257,7 @@ cdef int64_t get_itemsize(object val):
         return -1
 
 
+@set_module("pandas.api.types")
 def is_iterator(obj: object) -> bool:
     """
     Check if the object is an iterator.
@@ -1095,6 +1098,7 @@ def indices_fast(ndarray[intp_t, ndim=1] index, const int64_t[:] labels, list ke
 
 # core.common import for fast inference checks
 
+@set_module("pandas.api.types")
 def is_float(obj: object) -> bool:
     """
     Return True if given object is float.
@@ -1128,6 +1132,7 @@ def is_float(obj: object) -> bool:
     return util.is_float_object(obj)
 
 
+@set_module("pandas.api.types")
 def is_integer(obj: object) -> bool:
     """
     Return True if given object is integer.
@@ -1172,6 +1177,7 @@ def is_int_or_none(obj) -> bool:
     return obj is None or util.is_integer_object(obj)
 
 
+@set_module("pandas.api.types")
 def is_bool(obj: object) -> bool:
     """
     Return True if given object is boolean.
@@ -1202,6 +1208,7 @@ def is_bool(obj: object) -> bool:
     return util.is_bool_object(obj)
 
 
+@set_module("pandas.api.types")
 def is_complex(obj: object) -> bool:
     """
     Return True if given object is complex.
@@ -1237,6 +1244,7 @@ cpdef bint is_decimal(object obj):
     return isinstance(obj, Decimal)
 
 
+@set_module("pandas.api.types")
 def is_list_like(obj: object, allow_sets: bool = True) -> bool:
     """
     Check if the object is list-like.
@@ -1520,6 +1528,7 @@ cdef object _try_infer_map(object dtype):
     return None
 
 
+@set_module("pandas.api.types")
 def infer_dtype(value: object, skipna: bool = True) -> str:
     """
     Return a string label of the type of the elements in a list-like input.

pandas/_libs/parsers.pyx

@@ -144,7 +144,7 @@ cdef extern from "pandas/parser/tokenizer.h":
         SKIP_LINE
         FINISHED
 
-    enum: ERROR_OVERFLOW
+    enum: ERROR_OVERFLOW, ERROR_INVALID_CHARS
 
     ctypedef enum BadLineHandleMethod:
         ERROR,
@@ -1051,7 +1051,7 @@ cdef class TextReader:
         if col_dtype is not None:
             col_res, na_count = self._convert_with_dtype(
                 col_dtype, i, start, end, na_filter,
-                1, na_hashset, na_fset)
+                1, na_hashset, na_fset, False)
 
             # Fallback on the parse (e.g. we requested int dtype,
             # but its actually a float).
@@ -1062,30 +1062,34 @@ cdef class TextReader:
             return self._string_convert(i, start, end, na_filter, na_hashset)
         else:
             col_res = None
+            maybe_int = True
             for dt in self.dtype_cast_order:
-                if (dt.kind in "iu" and
-                        self._column_has_float(i, start, end, na_filter, na_hashset)):
+                if not maybe_int and dt.kind in "iu":
                     continue
 
                 try:
                     col_res, na_count = self._convert_with_dtype(
-                        dt, i, start, end, na_filter, 0, na_hashset, na_fset)
-                except ValueError:
-                    # This error is raised from trying to convert to uint64,
-                    # and we discover that we cannot convert to any numerical
-                    # dtype successfully. As a result, we leave the data
-                    # column AS IS with object dtype.
-                    col_res, na_count = self._convert_with_dtype(
-                        np.dtype("object"), i, start, end, 0,
-                        0, na_hashset, na_fset)
+                        dt, i, start, end, na_filter, 0, na_hashset, na_fset, True)
+                except ValueError as e:
+                    if str(e) == "Number is not int":
+                        maybe_int = False
+                        continue
+                    else:
+                        # This error is raised from trying to convert to uint64,
+                        # and we discover that we cannot convert to any numerical
+                        # dtype successfully. As a result, we leave the data
+                        # column AS IS with object dtype.
+                        col_res, na_count = self._convert_with_dtype(
+                            np.dtype("object"), i, start, end, 0,
+                            0, na_hashset, na_fset, False)
                 except OverflowError:
                     try:
                         col_res, na_count = _try_pylong(self.parser, i, start,
                                                         end, na_filter, na_hashset)
                     except ValueError:
                         col_res, na_count = self._convert_with_dtype(
                             np.dtype("object"), i, start, end, 0,
-                            0, na_hashset, na_fset)
+                            0, na_hashset, na_fset, False)
 
                 if col_res is not None:
                     break
@@ -1133,7 +1137,7 @@ cdef class TextReader:
                              bint na_filter,
                              bint user_dtype,
                              kh_str_starts_t *na_hashset,
-                             set na_fset):
+                             set na_fset, bint raise_on_invalid):
         if isinstance(dtype, CategoricalDtype):
             # TODO: I suspect that _categorical_convert could be
             # optimized when dtype is an instance of CategoricalDtype
@@ -1174,14 +1178,14 @@ cdef class TextReader:
 
         elif dtype.kind in "iu":
             try:
-                result, na_count = _try_int64(self.parser, i, start,
-                                              end, na_filter, na_hashset)
+                result, na_count = _try_int64(self.parser, i, start, end,
+                                              na_filter, na_hashset, raise_on_invalid)
                 if user_dtype and na_count is not None:
                     if na_count > 0:
                         raise ValueError(f"Integer column has NA values in column {i}")
             except OverflowError:
                 result = _try_uint64(self.parser, i, start, end,
-                                     na_filter, na_hashset)
+                                     na_filter, na_hashset, raise_on_invalid)
                 na_count = 0
 
             if result is not None and dtype != "int64":
@@ -1344,59 +1348,6 @@ cdef class TextReader:
             else:
                 return None
 
-    cdef bint _column_has_float(self, Py_ssize_t col,
-                                int64_t start, int64_t end,
-                                bint na_filter, kh_str_starts_t *na_hashset):
-        """Check if the column contains any float number."""
-        cdef:
-            Py_ssize_t i, j, lines = end - start
-            coliter_t it
-            const char *word = NULL
-            const char *ignored_chars = " +-"
-            const char *digits = "0123456789"
-            const char *float_indicating_chars = "eE"
-            char null_byte = 0
-
-        coliter_setup(&it, self.parser, col, start)
-
-        for i in range(lines):
-            COLITER_NEXT(it, word)
-
-            if na_filter and kh_get_str_starts_item(na_hashset, word):
-                continue
-
-            found_first_digit = False
-            j = 0
-            while word[j] != null_byte:
-                if word[j] == self.parser.decimal:
-                    return True
-                elif not found_first_digit and word[j] in ignored_chars:
-                    # no-op
-                    pass
-                elif not found_first_digit and word[j] not in digits:
-                    # word isn't numeric
-                    return False
-                elif not found_first_digit and word[j] in digits:
-                    found_first_digit = True
-                elif word[j] in float_indicating_chars:
-                    # preceding chars indicates numeric and
-                    # current char indicates float
-                    return True
-                elif word[j] not in digits:
-                    # previous characters indicates numeric
-                    # current character shows otherwise
-                    return False
-                elif word[j] in digits:
-                    # no-op
-                    pass
-                else:
-                    raise AssertionError(
-                            f"Unhandled case {word[j]=} {found_first_digit=}"
-                            )
-                j += 1
-
-        return False
-
 # Factor out code common to TextReader.__dealloc__ and TextReader.close
 # It cannot be a class method, since calling self.close() in __dealloc__
 # which causes a class attribute lookup and violates best practices
@@ -1793,7 +1744,8 @@ cdef int _try_double_nogil(parser_t *parser,
 
 cdef _try_uint64(parser_t *parser, int64_t col,
                  int64_t line_start, int64_t line_end,
-                 bint na_filter, kh_str_starts_t *na_hashset):
+                 bint na_filter, kh_str_starts_t *na_hashset,
+                 bint raise_on_invalid):
     cdef:
         int error
         Py_ssize_t lines
@@ -1815,6 +1767,8 @@ cdef _try_uint64(parser_t *parser, int64_t col,
         if error == ERROR_OVERFLOW:
             # Can't get the word variable
             raise OverflowError("Overflow")
+        elif raise_on_invalid and error == ERROR_INVALID_CHARS:
+            raise ValueError("Number is not int")
         return None
 
     if uint64_conflict(&state):
@@ -1863,7 +1817,7 @@ cdef int _try_uint64_nogil(parser_t *parser, int64_t col,
 
 cdef _try_int64(parser_t *parser, int64_t col,
                 int64_t line_start, int64_t line_end,
-                bint na_filter, kh_str_starts_t *na_hashset):
+                bint na_filter, kh_str_starts_t *na_hashset, bint raise_on_invalid):
     cdef:
         int error, na_count = 0
         Py_ssize_t lines
@@ -1883,6 +1837,8 @@ cdef _try_int64(parser_t *parser, int64_t col,
         if error == ERROR_OVERFLOW:
             # Can't get the word variable
             raise OverflowError("Overflow")
+        elif raise_on_invalid and error == ERROR_INVALID_CHARS:
+            raise ValueError("Number is not int")
         return None, None
 
     return result, na_count

pandas/_libs/src/parser/tokenizer.c

@@ -1620,9 +1620,9 @@ double precise_xstrtod(const char *str, char **endptr, char decimal, char sci,
   }
 
   double number = 0.;
-  int exponent = 0;
-  int num_digits = 0;
-  int num_decimals = 0;
+  long int exponent = 0;
+  long int num_digits = 0;
+  long int num_decimals = 0;
 
   // Process string of digits.
   while (isdigit_ascii(*p)) {
@@ -1671,39 +1671,26 @@ double precise_xstrtod(const char *str, char **endptr, char decimal, char sci,
     if (maybe_int != NULL)
       *maybe_int = 0;
 
-    // Handle optional sign
-    negative = 0;
-    switch (*++p) {
-    case '-':
-      negative = 1;
-      PD_FALLTHROUGH; // Fall through to increment position.
-    case '+':
-      p++;
-      break;
-    }
+    // move past scientific notation
+    p++;
 
-    // Process string of digits.
-    num_digits = 0;
-    int n = 0;
-    while (num_digits < max_digits && isdigit_ascii(*p)) {
-      n = n * 10 + (*p - '0');
-      num_digits++;
-      p++;
-    }
+    char *tmp_ptr;
+    long int n = strtol(p, &tmp_ptr, 10);
 
-    if (negative)
-      exponent -= n;
-    else
-      exponent += n;
+    if (errno == ERANGE || checked_add(exponent, n, &exponent)) {
+      errno = 0;
+      exponent = n;
+    }
 
     // If no digits after the 'e'/'E', un-consume it.
-    if (num_digits == 0)
+    if (tmp_ptr == p)
       p--;
+    else
+      p = tmp_ptr;
   }
 
   if (exponent > 308) {
-    *error = ERANGE;
-    return HUGE_VAL;
+    number = number == 0 ? 0 : number < 0 ? -HUGE_VAL : HUGE_VAL;
   } else if (exponent > 0) {
     number *= e[exponent];
   } else if (exponent < -308) { // Subnormal
@@ -1718,9 +1705,6 @@ double precise_xstrtod(const char *str, char **endptr, char decimal, char sci,
     number /= e[-exponent];
   }
 
-  if (number == HUGE_VAL || number == -HUGE_VAL)
-    *error = ERANGE;
-
   if (skip_trailing) {
     // Skip trailing whitespace.
     while (isspace_ascii(*p))
@@ -1812,8 +1796,6 @@ double round_trip(const char *p, char **q, char decimal, char Py_UNUSED(sci),
     *maybe_int = 0;
   if (PyErr_Occurred() != NULL)
     *error = -1;
-  else if (r == Py_HUGE_VAL)
-    *error = (int)Py_HUGE_VAL;
   PyErr_Clear();
 
   PyGILState_Release(gstate);
@@ -1907,7 +1889,7 @@ int64_t str_to_int64(const char *p_item, int *error, char tsep) {
   int64_t number = strtoll(p, &endptr, 10);
 
   if (errno == ERANGE) {
-    *error = ERROR_OVERFLOW;
+    *error = *endptr ? ERROR_INVALID_CHARS : ERROR_OVERFLOW;
     errno = 0;
     return 0;
   }
@@ -1967,7 +1949,7 @@ uint64_t str_to_uint64(uint_state *state, const char *p_item, int *error,
   uint64_t number = strtoull(p, &endptr, 10);
 
   if (errno == ERANGE) {
-    *error = ERROR_OVERFLOW;
+    *error = *endptr ? ERROR_INVALID_CHARS : ERROR_OVERFLOW;
     errno = 0;
     return 0;
   }