
Boost : 
From: Bill Seymour (billattheoffice_at_[hidden])
Date: 20020904 14:32:03
Beman Dawes wrote:
>
> I wrote some accounting software once for a petroleum distributor.
> Price (dollars per gallon, 4 places) times quantity (in gallons,
> 1 place) yielded invoice amount (in dollars, 2 places).
>
And Jeff Garland responded:
>
> Sure, but you could do all of this using 4 places and then round
> off to two with the final result. This would at most require a
> couple of conversion operations which I was arguing might be
> handled by an explicit user cast instead of providing overloaded
> operators for every possible combination.
>
There's no right answer to this. Part of the "spirit of C" that
C++ inherited is automatic conversion between arithmetic types
(the "usual arithmetic conversions"); but that creates a very
hard problem as Stroustrup points out in 11.1 of D&E.
My usual choice when writing a class that acts in some way like
an arithmetic type is making all oneargument constructors
explicit so that simple conversions require explicit casts,
then writing all the operators I need to let the user do things
like Beman described in simple, unsurprising ways. Usually,
all I need are templated versions of the operators for the
general case, then overloads or specializations for sametype
righthandside arguments where they're obviously more efficient.
If I find myself going in a direction that leads to a combinatorial
explosion of operators, I usually find that it's my design that's
wrong.
Bill Seymour
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk