> OK, I've thought about this some more and I think I agree with 
> you now. I'l add type coercion when I have time, should be easy
> enough. 

I think a widely used library that does this transparently will make life 
better for everyone.  I hope the implementation I mentioned earlier can 
save you some time :)

> typed APIs. In this case, if we only do conversions between different 
> numerical types I dont see that it will hurt, as long as the corner cases 
> are handled sensibly. 

I don't think you need to worry about this, if the coercion changes the 
semantic value, not just the representation or the precision, the sender 
is in error.  Sending '10.1' to a slot that is semantically an integer should 
be undefined, though in practice you get '10'.  Receivers need to run 
domain checks in any case...

Regards,
Rohan

Reply via email to