Add 3rd party integration issues FAQ; add numpy

Users of numpy and pandas should refer to this section
where we will try to catalog the common issues that don't
really have an automatic solution.

Fixes: #5552
Change-Id: I02990e328616ccb21933ffda1167c52256bace07

diff --git a/doc/build/faq/index.rst b/doc/build/faq/index.rst
index 5961226ce4f3072896e2ac12fd4d46bf1f1c5bc7..5238490a4070b2022ba72541fc07c6ffa5f755b6 100644 (file)
--- a/doc/build/faq/index.rst
+++ b/doc/build/faq/index.rst
@@ -16,4 +16,5 @@ observed questions to well-known issues.
     ormconfiguration
     performance
     sessions
+    thirdparty
 

diff --git a/doc/build/faq/thirdparty.rst b/doc/build/faq/thirdparty.rst
new file mode 100644 (file)
index 0000000..27c8fbf
--- /dev/null
+++ b/doc/build/faq/thirdparty.rst
@@ -0,0 +1,75 @@
+Third Party Integration Issues
+===============================
+
+.. contents::
+    :local:
+    :class: faq
+    :backlinks: none
+
+.. _numpy_int64:
+
+I'm getting errors related to "``numpy.int64``", "``numpy.bool_``", etc.
+------------------------------------------------------------------------
+
+The numpy_ package has its own numeric datatypes that extend from Python's
+numeric types, but contain some behaviors that in some cases make them impossible
+to reconcile with some of SQLAlchemy's behaviors, as well as in some cases
+with those of the underlying DBAPI driver in use.
+
+Two errors which can occur are ``ProgrammingError: can't adapt type 'numpy.int64'``
+on a backend such as psycopg2, and ``ArgumentError: SQL expression object
+expected, got object of type <class 'numpy.bool_'> instead``; in
+more recent versions of SQLAlchemy this may be ``ArgumentError: SQL expression
+for WHERE/HAVING role expected, got True``.
+
+In the first case, the issue is due to psycopg2 not having an appropriate
+lookup entry for the ``int64`` datatype such that it is not accepted directly
+by queries.   This may be illustrated from code based on the following::
+
+    import numpy
+
+    class A(Base):
+        __tablename__ = "a"
+
+        id = Column(Integer, primary_key=True)
+        data = Column(Integer)
+
+    # .. later
+    session.add(A(data=numpy.int64(10)))
+    session.commit()
+
+
+In the latter case, the issue is due to the ``numpy.int64`` datatype overriding
+the ``__eq__()`` method and enforcing that the return type of an expression is
+``numpy.True`` or ``numpy.False``, which breaks SQLAlchemy's expression
+language behavior that expects to return :class:`_sql.ColumnElement`
+expressions from Python equality comparisons::
+
+    >>> import numpy
+    >>> from sqlalchemy import column, Integer
+    >>> print(column('x', Integer) == numpy.int64(10))  # works
+    x = :x_1
+    >>> print(numpy.int64(10) == column('x', Integer))  # breaks
+    False
+
+These errors are both solved in the same way, which is that special numpy
+datatypes need to be replaced with regular Python values.  Examples include
+applying the Python ``int()`` function to types like ``numpy.int32`` and
+``numpy.int64`` and the Python ``float()`` function to ``numpy.float32``::
+
+    data = numpy.int64(10)
+
+    session.add(A(data=int(data)))
+
+    result = session.execute(
+        select(A.data).where(int(data) == A.data)
+    )
+
+    session.commit()
+
+.. _numpy: https://numpy.org
+
+SQL expression for WHERE/HAVING role expected, got True
+-------------------------------------------------------
+
+See :ref:`numpy_int64`.
\ No newline at end of file