Re: [boost] Re: Boost implementation of the Library TR?

From: Douglas Gregor (gregod_at_[hidden])
Date: 2003-05-04 10:06:36

On Sunday 04 May 2003 07:39 am, Gennaro Prota wrote:
> On Fri, 2 May 2003 09:44:05 -0400, Douglas Gregor <gregod_at_[hidden]>
>
> wrote:
> >I don't see any reason to rereview Joel & Jaakko's new implementation of
> >tuples. We don't have a policy of reviewing changes to implementations
> > once a library is initially accepted
>
> Which I've always considered a silly thing. Note that I'm speaking, so
> to say, against myself here, because dynamic_bitset is probably the
> library that has been changed most in the last months (there have been
> changes in the implementation, and there will be others; there have
> been changes to semantics -about which I've asked opinions here, with
> no success-  and there will probably be even backward compatible
> changes in the interface). This fact, that a library can be radically
> changed after acceptance, kinda defeats the review process.

Yes, major changes should probably be reviewed, but I don't know that we need 
another formal review. Perhaps there is a middle ground, for instance, where 
we designate, say, a two-day period for careful consideration of a major 
change before it's permanently committed. Library authors are generally quite 
careful when it comes to interface changes. FWIW, I've only once been bitten 
by a backward-incompatible change in Boost (you _know_ who you are! <g>).

> let me say, we
> should not be too afraid to break code if that is done to fix design
> errors: boost components are likely to be what we'll be in the
> standard in the next years; if we discover an error, and leave it
> there because of backward compatibility, what will we standardize?

There are some minor tweaks that can wait for standardization, and often lots 
of broken-compiler fat to trim away (e.g., the functionN class templates in 
Function), but otherwise I agree fully.

        Doug