Re: [Boost-users] [serialization] assertion inbasic_binary_oprimitive::save_binary() revisited

From: Robert Ramey (ramey_at_[hidden])
Date: 2006-01-02 14:37:16

Austin Bingham wrote:
> A while ago I brought up what I considered (and still consider) to be
> a serious problem in basic_binary_oprimitive::save_binary(). The
> problem is the assert at the end of the function:
>
>    assert(os.good());
>
> There are a number of reasons why doing this is a bad idea, but the
> discussion on this issue was dropped flat on the floor. Is there any
> possibility of discussing alternatives to this assertion? At the very
> least, can I get an explanation of why an assert is correct to which
> counter-arguments might be considered?

LOL - I remember when you brought this topic up.  I didn't have
a good answer so I didn't dwell on it.  I'm sure I put that in
without much thought so I wasn't really in a position to dispute
your proposition.  In fact, I was inclined to just accept your
view without much thought (as clearly you had thought about
it much more than I had).  This I did in my local copy which
has not yet been uploaded to CVS.

But it turns out that the issue is moot in any case.  Lately
I've been spending a little time looking at runtime performance
issues and I've concluded that I want to change the
implementation of basic_binary_oprimitive to use std::streambuf
primitives rather than std::basic_steram read/write. This
makes things faster for couple of reasons.  I've tested
locally the new implementation and it passes all tests.  So
I want to make the change "official". I do have a couple
of pending issues however:

a) add a contstructor for binary_?archive which takes
a reference to a stream buffer. This would make the
creation of a full blown stream unnecessary for
binary archives.
b) consider where binary i/o (as oppose to text i/o)
is specified and where errors in the area are detected.
c) update the relevant parts of the documentation.

So for now, I would suggest you update your local
copy to remove the offending assertion.

Robert Ramey

P.S.
If you're really  interested in this and have nothing else
to do I would be glad to send you my latest copy of
of basic_binary_i/oprimitive.* files

RR