SQL Expressions¶
How do I render SQL expressions as strings, possibly with bound parameters inlined?¶
The “stringification” of a SQLAlchemy Core statement object or
expression fragment, as well as that of an ORM Query
object,
in the majority of simple cases is as simple as using
the str()
builtin function, as below when use it with the print
function (note the Python print
function also calls str()
automatically
if we don’t use it explicitly):
>>> from sqlalchemy import table, column, select
>>> t = table('my_table', column('x'))
>>> statement = select([t])
>>> print(str(statement))
SELECT my_table.x
FROM my_table
The str()
builtin, or an equivalent, can be invoked on ORM
Query
object as well as any statement such as that of
select()
, insert()
etc. and also any expression fragment, such
as:
>>> from sqlalchemy import column
>>> print(column('x') == 'some value')
x = :x_1
Stringifying for Specific Databases¶
A complication arises when the statement or fragment we are stringifying
contains elements that have a database-specific string format, or when it
contains elements that are only available within a certain kind of database.
In these cases, we might get a stringified statement that is not in the correct
syntax for the database we are targeting, or the operation may raise a
UnsupportedCompilationError
exception. In these cases, it is
necessary that we stringify the statement using the
ClauseElement.compile()
method, while passing along an Engine
or Dialect
object that represents the target database. Such as
below, if we have a MySQL database engine, we can stringify a statement in
terms of the MySQL dialect:
from sqlalchemy import create_engine
engine = create_engine("mysql+pymysql://scott:tiger@localhost/test")
print(statement.compile(engine))
More directly, without building up an Engine
object we can
instantiate a Dialect
object directly, as below where we
use a PostgreSQL dialect:
from sqlalchemy.dialects import postgresql
print(statement.compile(dialect=postgresql.dialect()))
When given an ORM Query
object, in order to get at the
ClauseElement.compile()
method we only need access the Query.statement
accessor first:
statement = query.statement
print(statement.compile(someengine))
Rendering Bound Parameters Inline¶
Warning
Never use this technique with string content received from untrusted input, such as from web forms or other user-input applications. SQLAlchemy’s facilities to coerce Python values into direct SQL string values are not secure against untrusted input and do not validate the type of data being passed. Always use bound parameters when programmatically invoking non-DDL SQL statements against a relational database.
The above forms will render the SQL statement as it is passed to the Python
DBAPI, which includes that bound parameters are not rendered inline.
SQLAlchemy normally does not stringify bound parameters, as this is handled
appropriately by the Python DBAPI, not to mention bypassing bound
parameters is probably the most widely exploited security hole in
modern web applications. SQLAlchemy has limited ability to do this
stringification in certain circumstances such as that of emitting DDL.
In order to access this functionality one can use the literal_binds
flag, passed to compile_kwargs
:
from sqlalchemy.sql import table, column, select
t = table('t', column('x'))
s = select([t]).where(t.c.x == 5)
print(s.compile(compile_kwargs={"literal_binds": True})) # **do not use** with untrusted input!!!
the above approach has the caveats that it is only supported for basic
types, such as ints and strings, and furthermore if a bindparam()
without a pre-set value is used directly, it won’t be able to
stringify that either.
To support inline literal rendering for types not supported, implement
a TypeDecorator
for the target type which includes a
TypeDecorator.process_literal_param()
method:
from sqlalchemy import TypeDecorator, Integer
class MyFancyType(TypeDecorator):
impl = Integer
def process_literal_param(self, value, dialect):
return "my_fancy_formatting(%s)" % value
from sqlalchemy import Table, Column, MetaData
tab = Table('mytable', MetaData(), Column('x', MyFancyType()))
print(
tab.select().where(tab.c.x > 5).compile(
compile_kwargs={"literal_binds": True})
)
producing output like:
SELECT mytable.x
FROM mytable
WHERE mytable.x > my_fancy_formatting(5)
I’m using op() to generate a custom operator and my parenthesis are not coming out correctly¶
The Operators.op()
method allows one to create a custom database operator
otherwise not known by SQLAlchemy:
>>> print(column('q').op('->')(column('p')))
q -> p
However, when using it on the right side of a compound expression, it doesn’t generate parenthesis as we expect:
>>> print((column('q1') + column('q2')).op('->')(column('p')))
q1 + q2 -> p
Where above, we probably want (q1 + q2) -> p
.
The solution to this case is to set the precedence of the operator, using
the Operators.op.precedence
parameter, to a high
number, where 100 is the maximum value, and the highest number used by any
SQLAlchemy operator is currently 15:
>>> print((column('q1') + column('q2')).op('->', precedence=100)(column('p')))
(q1 + q2) -> p
We can also usually force parenthesization around a binary expression (e.g.
an expression that has left/right operands and an operator) using the
ColumnElement.self_group()
method:
>>> print((column('q1') + column('q2')).self_group().op('->')(column('p')))
(q1 + q2) -> p
Why are the parentheses rules like this?¶
A lot of databases barf when there are excessive parenthesis or when parenthesis are in unusual places they doesn’t expect, so SQLAlchemy does not generate parenthesis based on groupings, it uses operator precedence and if the operator is known to be associative, so that parenthesis are generated minimally. Otherwise, an expression like:
column('a') & column('b') & column('c') & column('d')
would produce:
(((a AND b) AND c) AND d)
which is fine but would probably annoy people (and be reported as a bug). In other cases, it leads to things that are more likely to confuse databases or at the very least readability, such as:
column('q', ARRAY(Integer, dimensions=2))[5][6]
would produce:
((q[5])[6])
There are also some edge cases where we get things like "(x) = 7"
and databases
really don’t like that either. So parenthesization doesn’t naively
parenthesize, it uses operator precedence and associativity to determine
groupings.
For Operators.op()
, the value of precedence defaults to zero.
What if we defaulted the value of Operators.op.precedence
to 100,
e.g. the highest? Then this expression makes more parenthesis, but is
otherwise OK, that is, these two are equivalent:
>>> print (column('q') - column('y')).op('+', precedence=100)(column('z'))
(q - y) + z
>>> print (column('q') - column('y')).op('+')(column('z'))
q - y + z
but these two are not:
>>> print column('q') - column('y').op('+', precedence=100)(column('z'))
q - y + z
>>> print column('q') - column('y').op('+')(column('z'))
q - (y + z)
For now, it’s not clear that as long as we are doing parenthesization based on operator precedence and associativity, if there is really a way to parenthesize automatically for a generic operator with no precedence given that is going to work in all cases, because sometimes you want a custom op to have a lower precedence than the other operators and sometimes you want it to be higher.
It is possible that maybe if the “binary” expression above forced the use of
the self_group()
method when op()
is called, making the assumption that
a compound expression on the left side can always be parenthesized harmlessly.
Perhaps this change can be made at some point, however for the time being
keeping the parenthesization rules more internally consistent seems to be
the safer approach.