gcc-3_1-branch and pending patches.

[Benjamin Kosnik]

> > Also, at some point, perhaps a written guideline on what constitutes
> > a non-ABI-breaking patch should be formulated.
> 
> That would be helpful beyond words.

I'd like to do something like:


> 2002-05-27 

Description of the libstdc++ ABI.

I. What is an ABI? What's covered? What's not?
- headers and library binary must match
- library only, compiler ABI different (but affects)
- shared library only
- minimum environment that supports ABI

II. Versioning
- headers
- libraries

III. ABI changes
- (anything) changing size of exported symbol
- (anything) changing alignment of exported symbol
- (anything) changing mangling on exported symbol
- (anything) adding or deleting exported symbol

-- pending
- look at sun's C++ ABI specs
- look at glibc's ABI specs
- look at the IA64-ABI specs

-- references
- ld info pages
- ELF info

And have it linked off the main documenation page as a libstdc++ ABI
document.

> > Also, I believe that the recommendation for distributors should be
> > to use --enable-version-specific-runtime-libs and
> > --with-gxx-include-dir= to version headers. Is this correct?
> 
> To my understanding, yes.
> 
> For the record, I feel these should be on by default somehow.  Gaby
> also thinks that versioning should be done by default, but has a
> different approach.

Yes. I'm aware that the details need to be resolved, but I think we all
agree in principle on the fact that the includes should be versioned as
well as the shared library binary. And that the versioning should be on
by default.

How it's done, of course, is still up in the air. Gaby, I'm a bit out of
it, but did you ever post patches for your idea?

-benjamin