Object Relational Mappers are evil (a meditation)

[Steven D'Aprano]

On Wed, 23 Dec 2009 20:11:25 -0500, Terry Reedy wrote:

>>> Instead the accepted, idiomatic Python way of writing this was to use
>>> short-circuit booleans:
>>>
>>> result = condition and x or y
>>>
>>> However this idiom is buggy! If x is a false-value (say, 0) then
>>> result gets set to y no matter what the value of condition.
> 
> This is only a bug if one expects otherwise.

I'm not saying the behaviour of `a and x or y` is buggy, but that it's 
use as a replacement for a ternary conditional expression is buggy; the 
*idiom* is buggy, not the behaviour of and/or.

If I say "you can make perfect hard boiled eggs by putting the egg in a 
glass of water in the microwave on high for eight minutes", and the egg 
explodes, that's not a bug in the microwave, that's a bug in the recipe.



>>> This buggy idiom survived many years of Python development, missed by
>>> virtually everyone.
> 
> The last statement is false. The hazard of using and/or was well-known
> back in '97 or so when I discovered or learned it and I believe it was
> mentioned in the FAQ entry on the subject. 

We can argue about how well-known it was for somebody like Raymond 
Hettinger to miss it, and for whoever did a code-review of his 
application to also miss it.


> The new alternative has the
> hazard that the condition and if-branch must be written and read in a
> backwards order.

If you had asked me a couple of years ago, I would have agreed, but I've 
now come to the conclusion that `x if condition else y` is not only 
perfectly natural, but at least as natural as the conventional order of 
`if condition then x else y` (at least for expressions, not for if 
statements).

"Steven, what are you doing on Monday night?"
"Going to the movies if I can get away from work on time, otherwise 
sitting at home answering questions on comp.lang.python."




-- 
Steven