Fix(pg): Encode binary values for PyGreSQL driver

Encode binary values for PyGreSQL driver using the same encoding as for
py-postgresql driver. This fixes the last remaining problems with the driver.

.travis.yml

@@ -70,10 +70,6 @@ env:
 
 matrix:
   allow_failures:
-    - env: TOXENV=py27-postgres-pygresql
-    - env: TOXENV=py34-postgres-pygresql
-    - env: TOXENV=py35-postgres-pygresql
-    - env: TOXENV=py36-postgres-pygresql
     - env: TOXENV=py27-firebird-fdb
     - env: TOXENV=py34-firebird-fdb
     - env: TOXENV=py35-firebird-fdb

docs/News.rst

@@ -28,7 +28,16 @@ Drivers (work in progress)
 * Encode binary values for py-postgresql driver. This fixes the
   last remaining problems with the driver.
 
-* Encode/decode binary values for PyGreSQL driver.
+* Encode binary values for PyGreSQL driver using the same encoding as for
+  py-postgresql driver. This fixes the last remaining problems with the driver.
+
+  Our own encoding is needed because unescape_bytea(escape_bytea()) is not
+  idempotent. See the comment for PQunescapeBytea at
+  https://www.postgresql.org/docs/9.6/static/libpq-exec.html:
+
+| This conversion is not exactly the inverse of PQescapeBytea, because the
+| string is not expected to be "escaped" when received from PQgetvalue. In
+| particular this means there is no need for string quoting considerations.
 
 * List all drivers in extras_require in setup.py.
 

docs/SQLObject.rst

@@ -1831,9 +1831,9 @@ PostgresConnection supports transactions and all other features.
 The user can choose a DB API driver for PostgreSQL by using a ``driver``
 parameter in DB URI or PostgresConnection that can be a comma-separated
 list of driver names. Possible drivers are: ``psycopg2``, psycopg1,
-``psycopg`` (tries psycopg2 and psycopg1), ``pygresql``, ``pygresql``,
-``pypostgresql``, ``pyodbc``, ``pypyodbc`` or ``odbc`` (try ``pyodbc`` and
-``pypyodbc``). Default is ``psycopg``.
+``psycopg`` (tries psycopg2 and psycopg1), ``pygresql``, ``pypostgresql``,
+``pyodbc``, ``pypyodbc`` or ``odbc`` (try ``pyodbc`` and ``pypyodbc``).
+Default is ``psycopg``.
 
 Connection-specific parameters are: ``sslmode``, ``unicodeCols``,
 ``schema``, ``charset``.

sqlobject/col.py

@@ -1770,9 +1770,6 @@ def to_python(self, value, state):
         if isinstance(value, str):
             if not PY2 and dbName == "mysql":
                 value = value.encode('ascii', errors='surrogateescape')
-            if dbName == "postgres" and connection.driver == 'pygresql':
-                from pg import unescape_bytea
-                value = unescape_bytea(value)
             if dbName == "sqlite":
                 if not PY2:
                     value = bytes(value, 'ascii')

sqlobject/postgres/pgconnection.py

@@ -85,14 +85,9 @@ def __init__(self, dsn=None, host=None, port=None, db=None,
             # Register a converter for psycopg Binary type.
             registerConverter(type(self.module.Binary('')),
                               PsycoBinaryConverter)
-        elif driver == 'pygresql':
-            from pg import escape_bytea as pg_escape_bytea
-            self.createBinary = \
-                lambda value, pg_escape_bytea=pg_escape_bytea: \
-                pg_escape_bytea(value)
-        elif driver in ('py-postgresql', 'pypostgresql'):
+        elif driver in ('pygresql', 'py-postgresql', 'pypostgresql'):
             registerConverter(type(self.module.Binary(b'')),
-                              PypostgresBinaryConverter)
+                              PostgresBinaryConverter)
         elif driver in ('odbc', 'pyodbc', 'pypyodbc'):
             registerConverter(bytearray, OdbcBinaryConverter)
 
@@ -541,15 +536,23 @@ def PsycoBinaryConverter(value, db):
     return str(value)
 
 
-def escape_bytea(value):
-    return ''.join(
-        ['\\' + (x[2:].rjust(3, '0')) for x in (oct(ord(c)) for c in value)]
-    )
+if PY2:
+    def escape_bytea(value):
+        return ''.join(
+            ['\\' + (x[1:].rjust(3, '0'))
+                for x in (oct(ord(c)) for c in value)]
+        )
+else:
+    def escape_bytea(value):
+        return ''.join(
+            ['\\' + (x[2:].rjust(3, '0'))
+                for x in (oct(ord(c)) for c in value.decode('latin1'))]
+        )
 
 
-def PypostgresBinaryConverter(value, db):
+def PostgresBinaryConverter(value, db):
     assert db == 'postgres'
-    return sqlrepr(escape_bytea(value.decode('latin1')), db)
+    return sqlrepr(escape_bytea(value), db)
 
 
 def OdbcBinaryConverter(value, db):