Implement generic Double and related fixed types

Added :class:`.Double`, :class:`.DOUBLE`, :class:`.DOUBLE_PRECISION`
datatypes to the base ``sqlalchemy.`` module namespace, for explicit use of
double/double precision as well as generic "double" datatypes. Use
:class:`.Double` for generic support that will resolve to DOUBLE/DOUBLE
PRECISION/FLOAT as needed for different backends.

Implemented DDL and reflection support for ``FLOAT`` datatypes which
include an explicit "binary_precision" value. Using the Oracle-specific
:class:`_oracle.FLOAT` datatype, the new parameter
:paramref:`_oracle.FLOAT.binary_precision` may be specified which will
render Oracle's precision for floating point types directly. This value is
interpreted during reflection. Upon reflecting back a ``FLOAT`` datatype,
the datatype returned is one of :class:`_types.DOUBLE_PRECISION` for a
``FLOAT`` for a precision of 126 (this is also Oracle's default precision
for ``FLOAT``), :class:`_types.REAL` for a precision of 63, and
:class:`_oracle.FLOAT` for a custom precision, as per Oracle documentation.

As part of this change, the generic :paramref:`_sqltypes.Float.precision`
value is explicitly rejected when generating DDL for Oracle, as this
precision cannot be accurately converted to "binary precision"; instead, an
error message encourages the use of
:meth:`_sqltypes.TypeEngine.with_variant` so that Oracle's specific form of
precision may be chosen exactly. This is a backwards-incompatible change in
behavior, as the previous "precision" value was silently ignored for
Oracle.

Fixes: #5465
Closes: #7674
Pull-request: https://github.com/sqlalchemy/sqlalchemy/pull/7674
Pull-request-sha: 5c68419e5aee2e27bf21a8ac9eb5950d196c77e5

Change-Id: I831f4af3ee3b23fde02e8f6393c83e23dd7cd34d

2 files changed, 95 insertions, 10 deletions

diff --git a/lib/sqlalchemy/dialects/oracle/__init__.py b/lib/sqlalchemy/dialects/oracle/__init__.py
index c83e0573d..8994b4a10 100644
--- a/lib/sqlalchemy/dialects/oracle/__init__.py
+++ b/lib/sqlalchemy/dialects/oracle/__init__.py
@@ -24,6 +24,7 @@ from .base import NUMBER
 from .base import NVARCHAR
 from .base import NVARCHAR2
 from .base import RAW
+from .base import REAL
 from .base import ROWID
 from .base import TIMESTAMP
 from .base import VARCHAR
diff --git a/lib/sqlalchemy/dialects/oracle/base.py b/lib/sqlalchemy/dialects/oracle/base.py
index 578450d24..1ae58b8f4 100644
--- a/lib/sqlalchemy/dialects/oracle/base.py
+++ b/lib/sqlalchemy/dialects/oracle/base.py
@@ -558,10 +558,12 @@ from ...sql import visitors
 from ...types import BLOB
 from ...types import CHAR
 from ...types import CLOB
+from ...types import DOUBLE_PRECISION
 from ...types import FLOAT
 from ...types import INTEGER
 from ...types import NCHAR
 from ...types import NVARCHAR
+from ...types import REAL
 from ...types import TIMESTAMP
 from ...types import VARCHAR
 
@@ -625,8 +627,49 @@ class NUMBER(sqltypes.Numeric, sqltypes.Integer):
             return sqltypes.Integer
 
 
-class DOUBLE_PRECISION(sqltypes.Float):
-    __visit_name__ = "DOUBLE_PRECISION"
+class FLOAT(sqltypes.FLOAT):
+    """Oracle FLOAT.
+
+    This is the same as :class:`_sqltypes.FLOAT` except that
+    an Oracle-specific :paramref:`_oracle.FLOAT.binary_precision`
+    parameter is accepted, and
+    the :paramref:`_sqltypes.Float.precision` parameter is not accepted.
+
+    Oracle FLOAT types indicate precision in terms of "binary precision", which
+    defaults to 126. For a REAL type, the value is 63. This parameter does not
+    cleanly map to a specific number of decimal places but is roughly
+    equivalent to the desired number of decimal places divided by 0.3103.
+
+    .. versionadded:: 2.0
+
+    """
+
+    __visit_name__ = "FLOAT"
+
+    def __init__(
+        self,
+        binary_precision=None,
+        asdecimal=False,
+        decimal_return_scale=None,
+    ):
+        r"""
+        Construct a FLOAT
+
+        :param binary_precision: Oracle binary precision value to be rendered
+         in DDL. This may be approximated to the number of decimal characters
+         using the formula "decimal precision = 0.30103 * binary precision".
+         The default value used by Oracle for FLOAT / DOUBLE PRECISION is 126.
+
+        :param asdecimal: See :paramref:`_sqltypes.Float.asdecimal`
+
+        :param decimal_return_scale: See
+         :paramref:`_sqltypes.Float.decimal_return_scale`
+
+        """
+        super().__init__(
+            asdecimal=asdecimal, decimal_return_scale=decimal_return_scale
+        )
+        self.binary_precision = binary_precision
 
 
 class BINARY_DOUBLE(sqltypes.Float):
@@ -742,6 +785,7 @@ ischema_names = {
     "RAW": RAW,
     "FLOAT": FLOAT,
     "DOUBLE PRECISION": DOUBLE_PRECISION,
+    "REAL": REAL,
     "LONG": LONG,
     "BINARY_DOUBLE": BINARY_DOUBLE,
     "BINARY_FLOAT": BINARY_FLOAT,
@@ -760,6 +804,9 @@ class OracleTypeCompiler(compiler.GenericTypeCompiler):
     def visit_float(self, type_, **kw):
         return self.visit_FLOAT(type_, **kw)
 
+    def visit_double(self, type_, **kw):
+        return self.visit_DOUBLE_PRECISION(type_, **kw)
+
     def visit_unicode(self, type_, **kw):
         if self.dialect._use_nchar_for_unicode:
             return self.visit_NVARCHAR2(type_, **kw)
@@ -795,24 +842,50 @@ class OracleTypeCompiler(compiler.GenericTypeCompiler):
         return self._generate_numeric(type_, "BINARY_FLOAT", **kw)
 
     def visit_FLOAT(self, type_, **kw):
-        # don't support conversion between decimal/binary
-        # precision yet
-        kw["no_precision"] = True
+        kw["_requires_binary_precision"] = True
         return self._generate_numeric(type_, "FLOAT", **kw)
 
     def visit_NUMBER(self, type_, **kw):
         return self._generate_numeric(type_, "NUMBER", **kw)
 
     def _generate_numeric(
-        self, type_, name, precision=None, scale=None, no_precision=False, **kw
+        self,
+        type_,
+        name,
+        precision=None,
+        scale=None,
+        _requires_binary_precision=False,
+        **kw,
     ):
         if precision is None:
-            precision = type_.precision
+
+            precision = getattr(type_, "precision", None)
+
+        if _requires_binary_precision:
+            binary_precision = getattr(type_, "binary_precision", None)
+
+            if precision and binary_precision is None:
+                # https://www.oracletutorial.com/oracle-basics/oracle-float/
+                estimated_binary_precision = int(precision / 0.30103)
+                raise exc.ArgumentError(
+                    "Oracle FLOAT types use 'binary precision', which does "
+                    "not convert cleanly from decimal 'precision'.  Please "
+                    "specify "
+                    f"this type with a separate Oracle variant, such as "
+                    f"{type_.__class__.__name__}(precision={precision})."
+                    f"with_variant(oracle.FLOAT"
+                    f"(binary_precision="
+                    f"{estimated_binary_precision}), 'oracle'), so that the "
+                    "Oracle specific 'binary_precision' may be specified "
+                    "accurately."
+                )
+            else:
+                precision = binary_precision
 
         if scale is None:
             scale = getattr(type_, "scale", None)
 
-        if no_precision or precision is None:
+        if precision is None:
             return name
         elif scale is None:
             n = "%(name)s(%(precision)s)"
@@ -1964,8 +2037,19 @@ class OracleDialect(default.DefaultDialect):
                 else:
                     coltype = NUMBER(precision, scale)
             elif coltype == "FLOAT":
-                # TODO: support "precision" here as "binary_precision"
-                coltype = FLOAT()
+                # https://docs.oracle.com/cd/B14117_01/server.101/b10758/sqlqr06.htm
+                if precision == 126:
+                    # The DOUBLE PRECISION datatype is a floating-point
+                    # number with binary precision 126.
+                    coltype = DOUBLE_PRECISION()
+                elif precision == 63:
+                    # The REAL datatype is a floating-point number with a
+                    # binary precision of 63, or 18 decimal.
+                    coltype = REAL()
+                else:
+                    # non standard precision
+                    coltype = FLOAT(binary_precision=precision)
+
             elif coltype in ("VARCHAR2", "NVARCHAR2", "CHAR", "NCHAR"):
                 coltype = self.ischema_names.get(coltype)(length)
             elif "WITH TIME ZONE" in coltype: