Re: [Rosegarden-devel] defensive programming

[Chris C. <ca...@al...>]

On Saturday 07 Jun 2003 6:45 pm, Guillaume Laurent wrote:
> On Saturday 07 June 2003 14:00, Levi Burton wrote:
> > So, basically checking every assumption that *pointer is not null
> > with an assert().  I am guessing thats a shit-load of work.

I think most of the work is in finding the places where it would=20
actually make sense to assert (for example, there's no point in=20
testing a pointer if you've just got it back from new(), there's no=20
point in testing a pointer merely for an assert if you're just about=20
to dereference it anyway, etc).  If you're going to put in that=20
effort, you might as well simply do a better job of recovering=20
correctly from such problems, rather than just asserting.

Asserts are handy in some circumstances, but they're more something=20
you write as you write the code than something you work to add=20
afterwards.  And they can cause problems -- I have twice recently had=20
to remove asserts because they were causing crashes in cases where=20
the assumption they were testing was not in fact true any more (i.e.=20
a null value was actually acceptable, or whatever).

> I don't mean to
> brag but I agree with Rich, given our features list our stability
> is fairly good.

Given that this thread was inspired by a comment of mine about how our=20
code could use some improvement, I think the subject is certainly=20
fair game.  I just don't think assert is quite the answer.


Chris