Re: stl implementations--does anybody use SGI STL anymore?

[Randy Smith]

Ulrich Eckhardt wrote:
> Randy Smith wrote:
>
> > I am curious what the dominant stl implementations available are
> > nowadays. 
>
>
> Only one, the STL. In case you wonder, the STL (there is only one,
> remember) has strongly influenced the C++ standardlibrary. With some
> exceptions, it has been completely incorporated into the C++
> standardlibrary in fact and there it resides next to IOStreams, the C API
> and some other stuff.
>
>
> > I've used the SGI stl implementation for quite a while alongside gcc, 
> > but a recent upgrade to gcc (using 3.4.4 now) will not compile the sgi 
> > stl, and it looks like non-standards-compliant code is the culprit. 
> > Browsing around using google, I could find very little information on 
> > sgi stl, so I'm guessing it has gone into disuse.
>
>
> Right, soon after C++ was standardised, the meaning of the STL on its own
> vanished.
>
>
> > Although it has been several years since, I have experienced some pretty 
> > nasty multi-thread-triggered bugs in the gnu stl that were not 
> > problematic in the sgi stl (the bug had to do with inherent 
> > thread-safety problems in gnu stl's implementation of string creation).
>
>
> Which bug number? Did you check if it got fixed? Or did you check if it was
> a bug at all and not just misuse?

I never looked for a number, but it was bug not a misuse.  I didn't have 
the option of upgrading our entire toolchain even if it was eventually 
fixed, so when we learned from others who experienced the same thing 
that SGI's STL implementation did not have the problem, we switched STL 
impls and didn't look back.

The problem was that string creation was inherently not thread-safe; if 
a thread was in the process of creating a string, switched out, and 
another created a string, the contents of the first string could get 
corrupted.  The SGI stl avoided this problem, although string creation 
was somewhat slower.

The problem was so expensive in terms of man-hours to identify (the 
software was on an embedded box and the bug exhibited itself only under 
extreme load) that all of us involved became leery (for better or worse) 
of the gnu c++ stdlib.  This was circa 2001-2002.

Thanks for comments.
Randy Smith