|
Boost : |
From: Peter Dimov (pdimov_at_[hidden])
Date: 2002-09-17 15:59:27
From: "Beman Dawes" <bdawes_at_[hidden]>
> At 12:47 PM 9/17/2002, Peter Dimov wrote:
>
> >> > Comments? Rationale for the original text?
> >>
> >> All I can say was that it was true with the original non-threadsafe
> >> shared_ptr design.
> >
> >Yes, I know that, it's just that the standard library practice is to not
> >enumerate the possible exceptions. There might have been a particular
> >reason for documenting things that way for smart_ptr.
>
> Because the Boost smart pointers are not part of the standard library,
they
> aren't automatically covered by the standard's clause 17 front matter. It
> may have been an attempt to boil all of clause 17 down to a single
> sentence. If so, that wasn't a very good idea on my part.
Part of the problem is that we are writing both an interface and an
implementation. It would be unwise to couple the interface to the (current)
implementation. We can't adopt 17.4.4.8's wording, though, as it says that
functions may throw implementation-defined exceptions unless otherwise
specified. "Implementation defined" means that we must document the exact
exception types since we are the implementors!
I think that we should consider imposing a boost-wide "Chapter 17" saying
that functions may throw exceptions derived from standard exception classes
unless otherwise specified.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk