From: Chris Little (cslittle_at_[hidden])
Date: 2002-11-01 15:00:20
on 11/1/02 1:51 PM, David B. Held at dheld_at_[hidden] wrote:
> James Curran wrote:
>> I wouldn't be surprised if some vendors removed all #ifdef's not
>> needed for their compiler.... And several vendors have written their
>> own STL's despite the original being available.
> But how many compiler vendors write a standard library implementation
> for just their compiler? All the SCL implementations I'm aware of are
> written by library vendors, not compiler vendors. And they are all
> designed to work on multiple compilers. So I don't see how removing
> portability is an improvement. I suppose if there were significant QoI
> issues to be addressed, a library vendor might modify the implementation
> to reflect that. But it seems that metalibraries like tuples leave less
> and less room for the type of optimizations you see in, say, SGI's node
> allocators for their STL implementation. And if something like Spirit
> were to be accepted for the TR, how likely is it that any vendor would
> be able to provide a different implementation without breaking the
> library? I don't know the answers to these questions, but I'm curious.
As an FYI Metrowerks makes their own standard library.
I think that it is quite likely that some vendors (standard library or
compiler) will choose to build their own versions of the parts of boost that
are accepted so that they can have better integration, value adds and be
tailored for custom platforms.