Hi Jeroen,
Thanks for raising this point!
On Mon, Sep 24, 2012 at 09:24:37AM +0200, Jeroen Demeyer wrote:
> In Sage, I have seen some quite some places where assert/AssertionError
> is used for control flow, where an
>
> if condition:
> raise ValueError("...") # or other exception
>
> would be the right thing to use instead of an assert.
>
> I find this bad, because an assert checks something which should always
> be true, a failed assertion is always a bug in the program. It's not
> something which should be raised and caught like other exceptions. Also
> remember that assertions are something which can be turned off, and
> doing that should not change the program.
Definitely +1. For the same reason, I particularly dislike calling a
function with random garbage to see if it can handle it, hopping that
the function will properly catch and report bad values. We are doing
it a lot in the coercion code, and this is sooo brittle and unexplicit.
> That being said, assertions certainly have their place if used well.
> But if there is any user input to a public function which can cause an
> assertion, that is by definition a bug. Just like any segmentation
> fault in the program is by definition a bug (even if it's caused by
> bogus input).
I personally consider that using ``assert`` for checking the
preconditions of a function has two strong values:
- It's very concise, readable and to the point. Compare:
assert i > 0
With:
if i <= 0:
raise ValueError("i should be a positive integer")
An specific point is that it is hard to enforce consistency across
all hand written error messages. Also note that, upon error, the
`assert i > 0` line appears in the traceback.
- It can be turned off
This point is *crucial*: it relieves the programmer from the
dangerous temptation of not testing the input to save on speed.
Granted, it's on the harsh side on the user to raise an assertion, so
I see the point of writing nice ValueError error messages for the most
public or non speed-critical functions. But for other functions I
think we should allow for using asserts.
That being said, if there is a way to raise ValueError in a way that
can be turned off like for assertions, I am happy to change my mind!
(I am missing this from MuPAD; we could even change this at runtime
which was really cool),.
By the way, we should encourage more assert's everywhere in the
code. And to start with, implement ``sage -O`` which passes the -O
option to python to disable assertions (#13523).
Cheers,
Nicolas
--
Nicolas M. Thiéry "Isil" <nthi...@users.sf.net>
http://Nicolas.Thiery.name/
--
You received this message because you are subscribed to the Google Groups
"sage-devel" group.
To post to this group, send email to sage-devel@googlegroups.com.
To unsubscribe from this group, send email to
sage-devel+unsubscr...@googlegroups.com.
Visit this group at http://groups.google.com/group/sage-devel?hl=en.
