I finally found the bug: its name is "g++"

Carlo Wood carlo@alinoe.com
Thu Sep 6 18:54:00 GMT 2001

On Thu, Sep 06, 2001 at 01:31:21PM -0700, George Garvey wrote:
>    Been using it a lot longer than that.

Well, I've been using egcs too.  But 2.95.x and before doesn't count :).

> I total
> understand Markus. Been feeling about the same, and considering the same
> options.

I find that surpricing, it sounds like you think that g++-3.x is a total mess.
But it is not!  It can compile and deal with extreme large and complex
libraries (ACE, blitz, libcw[d]).

> a way to trim the problem down to make it easy enough to replicate. Can't
> believe I'm the only one who can't compile openSSL right now with 3, for
> instance.

Isn't that a problem of openSSL?  When I switched from 2.95.x to 3.0 I had
to fix VERY MUCH before my library (libcwd) worked again.  Took me two
months of work.  And that was NOT caused by a buggy compiler, it was caused
by the fact that apparently I was relying on many things that aren't
really garanteed (the main problem being that now malloc() was called before
libstdc++ was initialized - libstdc++-v2 didn't do that).  If I had to
practically rewrite my library (I am not kidding here!) then I believe that
other libraries will need to go through a debugging/test cycle as well
before you can blame the compiler.

Carlo Wood <carlo@alinoe.com>

More information about the Gcc-bugs mailing list