Re: [boost] Boost libraries after C++0x acceptance

From: Robert Ramey (ramey_at_[hidden])
Date: 2006-10-09 15:51:28

Edward Diener wrote:
> Robert Ramey wrote:
>
>> The idea that making a library interface "standard" confers a sort
>> of legitimacy that says: "a library that implements this interface
>> would be a good (or necessary) thing to have" has some
>> validity.  But having a commitee say it has to have this thing
>> to be considered a comforming implementation seems to
>> be superfluous given the presumption of conforming language
>> translator and reference implementation which conforms
>> to legal C++.
>
> I think you are arguing from an emotional perspective which feels that
> there is no necessity to have standard libraries as long as libraries
> are implemented in a C++ standard conforming way. The minimalist
> approach feels good to you.

Well, I plead guilty to the preference for minimalism.  But DAMN
IT - I'm NOT emotional about it.  I don't think that the idea of the
standard library is detramental - In fact I've conceded the point
that the fact it has the "official sanction" is helpful in a number
of organizations.  But I just don't think its really very
effective in achieving what I think is the goal - making C++ more
widely useful.  Also I think it creates the impression that efforts
underway to make C++ more widely useful than I think they
could be.

> It is a nice ideal but in practical use fails because of the reality
> of compiler conformance. I too would love to have all compilers 100%
> conformant to the C++ standard. In an ideal world this would be so,
> and therefore a library which is 100% conformant would automatically
> work
> with any library which was 100% conformant, and therefore specifying a
> standard library would be a moot point, as long as there were a single
> conforming library which one could use.

This is more or less my line of reasoning.

> Unfortunately that world does
> not even approximately exist in C++, and to think it does, and
> therefore there is no need to have a C++ standard library, as opposed
> to 100% conformant C++ libraries, such as a 100% conformant reference
> implementation of the current C++ standard library, or of 100%
> conformant third party libraries like Boost, is just wishful thinking.

Ahhh - and why is that?  why is it so hard to have a 100% conformant
compiler that no one has been able to it.  That is another question
we should be asking ourselves.  As you note - this is the key source
of the difficulty.

And its even worse.  Given two 100% conforming compilers, there are
areas in the language which are undefined and left to the implementors.
Many of these don't effect language semantics - but some do.  So
even with two 100% conformant compilers there is not guarentee
that the same library can be compiled on both with the same results.

And that's a problem.  Having each vendor tweak his library around
this is A solution.  But not a very satisfactory one as far as I'm
concerned.

Hmmm, so, maybe I've answered my own original question here.

100 % comformant compilers cannot be guarenteed to be
equivalent - so a vendor MUST include implementation of a standard
library which he has verified will work on his compiler.

Of course I'm not really happy with that answer - it doesn't help
me guarentee portability of my own libraries as I would like - but
it IS an answer and probably the "real" one.  Oh Well.

> I want to also point out to you that even implementations which are,
> supposedly, 100% conformant to the particular language, such as Sun's
> Java JRE or Microsoft's .NET implementation, ship automatically with a
> set of libraries which may be called the standard library for that
> "language". Do you think that they should not, but rather say that
> there
> is a bunch of libraries out here somewhere, presumably on the
> Internet,

I have no issue with a bunch of libraries being shipped with a compiler.
It's very convenient to have pre-build libraries.  I use MFC this
way all the time.  God help me if I had to build it.  But my question
is "what is thebenefit of having a ""standard library" defined when
there is a free working implementation.  After - the free working
implementation IS a definition of the API.

Robert Ramey