Thumbs up or thumbs down: stl_config.h ?

Matt Austern
Fri Feb 16 14:29:00 GMT 2001

Phil Edwards wrote:
> Okay, I'll hack on this while waiting for responses on other topics.
> > | eventually we want to remove the uncesessary code from the STL itself. I
> > | think Matt is doing this anyway, so I don't want to duplicate efforts.
> >
> > Is there a chance to have V3 maintainers to have closer look at what
> > is going on on Matt's side?
> And, will SGI allow someone working at AT&T to continue to maintain SGI's
> implementation?  We all know how wacky those folks are.  :-)

Probably not, any more than HP would allow people working at SGI to
maintain HP's implementation.

Yes, I should talk more closely to you people about just what I'm doing.
 - I'm writing a new implementation of the standard C++ library, using
   SGI's library as a code base.  Many files have HP, SGI, and AT&T 
 - One difference between this and the SGI library is that I'm no
   longer attempting to support obsolete compilers.  That means that
   almost all of the macros in stl_config.h are going away.  I think
   I do still need a config file for just one or two things, though---
   for example, the presence or absence of long long.
 - Another difference between this and the SGI library is that I'm
   no longer attempting to keep allof the STL code in header files.
   There are a lot of ".cxx" files, so many things are cleaner.
 - Yet another difference is that I'm taking standard conformance
   more seriously.  Extensions will be in their own headers and in 
   their own namespace.  
 - Something I haven't yet done is go carefully over the I/O part
   of the library and incorporate all of the performance enhancements
   from the sfio library.  (sfio is a replacement for stdio, written
   by another AT&T person, Phuong Vo. There are some very clever
   things in sfio.)
 - Another thing I haven't yet done is redo the low-level locale 
   interface.  The version I did for the SGI library was a reasonable 
   first attempt, but it's pretty clunky in some ways.


More information about the Libstdc++ mailing list