new STL

new STL

[Mike Stump]

The new STL should be folded in for the 1.1 release.

Re: new STL

[Joe Buck]

> The new STL should be folded in for the 1.1 release.

Well, Jeff said no. Given this, we shouldn't ask him to change his mind
without solid evidence.

So here's how we get some solid evidence: let's do some testing.
For anyone interested (and if you really, really want the new STL,
you should be), do the following:

0. Have a copy of the g++ and libstdc++ test results for the latest
   snapshot on your machine (run "make check" as described in the FAQ).
   (19980707)

1. Download the new STL from http://www.sgi.com/Technology/STL/download.html

2. Remove the following files from the new STL:
	stdexcept
	string
	char_traits.h

(this means we will use the existing string class: after further study,
I think that the SGI string class has unacceptable performance in many
applications, so we should consider whether to switch to it separately.
the SGI stdexcept is pretty much identical to the one we have).

3. Install the new STL on top of your existing egcs snapshot installation,
   by copying all the files into $prefix/include/g++ .

4. Rerun the g++ and libstdc++ tests.  Report any changes from step 0.

5. (optional) compile other STL programs you have with this new setup.

If, within a week, we have tests on the major platforms showing that the
new STL has no more test failures or other problems than the old STL,
at that point (and *only* at that point) I think we should ask Jeff and
Jason to change their minds and include the new STL.  Not before.

Alternatively, if an additional compiler bug is exposed but quickly fixed,
it could still be included.

One issue with all of this is that the existing libstdc++ tests are
extremely weak.  It would be nice if we had tests that provide better
coverage.