Re: Re: Boost vs. Object Space

From: Lois Goldthwaite (loisg_at_[hidden])
Date: 2001-02-15 10:31:56

The point of having a Standard library is that implementations are
supposed to be interchangeable. And, all things considered, this appears
to be a workable concept, not only in theory but also (frequently) in
practice.* As a partial proof of concept, I will mention in passing that
the other Objectspace libraries permit substitution of a different STL
-- simply define OS_USE_ALTERNATE_STD and recompile.

Having said that, however, I must be the first to admit that
Objectspace's STL is still playing catch-up to the Standard (and is not
alone in this). In particular, it is behind in supporting namespace std.
Solid namespace support from compilers is a relatively recent
development, and the installed base for this product goes back to 1995.
We don't always have the luxury of writing purist code and waiting for
compilers to catch up to coping with it.

Y'all are going to shame me into building boost with ospace STL, just to
prove it can be done. Yet another item on my to-do list....

Lois

*As we all know, in theory, theory and practice are the same. However,
in practice, they can be vastly different.

> Message: 19
>    Date: Wed, 14 Feb 2001 20:19:01 +0100
>    From: Jens Maurer <Jens.Maurer_at_[hidden]>
> Subject: Re: Re: Boost vs. Object Space
> 
> Phlip wrote:
> > But I fear Boost, being so tight with the STL backing it, must be "tuned" by
> > a booster to match the idiosyncrasies of each STL it can use.
> 
> Fear no longer.
> 
> >   Dinkumware,
> > yes. g++, yes. If Boost has never seen Objectspace before...
> 
> So what?  We're not programming for a certain standard library implementation,
> but for an interface specification (namely, the ISO C++ standard, which also
> describes the standard library).
> 
> If boost crashes with Object Space's standard library implementation,
> then either boost relies on undefined or implementation-defined behaviour
> or simply has a bug (and thus should be fixed), or Object Space does not
> implement the interface specification completely or correctly, as set forth
> by the ISO C++ standard (and thus Object Space should be fixed).
> 
> That was the ideal world, a few side remarks:
>  - Sometimes implementations are slow adjusting to ISO C++, so boost is
> prepared to work around some deficiencies.
>  - Sometimes a certain bug shows up only in a certain compiler/library/system
> combination.
>  - Sometimes the ISO standard itself still has bugs.
> 
> Jens Maurer
> 
>