On Tue, Nov 20, 2001 at 01:39:21PM -0500, Christopher C. Chimelis wrote:
> I believe that it is incorrect to rely on that. It's possible that the
> new operator was contained in libgcc in 2.95.4, meaning that it could
> satisfy the symbol without libstdc++ (I just checked...2.95.x's libgcc has
> a builtin new operator, while 3.0.x doesn't, which explains why it worked
> with 2.95.x).
[...]
> I tried your example with both gcc 2.95.x and 3.0.x and confirmed the
> above. With 2.95.4 (unstable), the new operator is, in fact, using
> __builtin_new, which is linked into the binary from libgcc.a. 3.0's
> libgcc has no builtin new, so it needs libstdc++ to satisfy the symbol
> at link time. In fact, the 2.95.4 version isn't linked to libstdc++ at
> all, fyi, since it uses no symbols from it.
All true. Just as an addendum: if a user only needs support code (new,
delete, etc) and doesn't feel like linking against the full libstdc++,
the support code also exists in a separate library, libsupc++.
http://gcc.gnu.org/onlinedocs/libstdc++/faq/index.html#2_5
And like you said, this didn't exist at all in the 2.x days.
Phil
--
If ye love wealth greater than liberty, the tranquility of servitude greater
than the animating contest for freedom, go home and leave us in peace. We seek
not your counsel, nor your arms. Crouch down and lick the hand that feeds you;
and may posterity forget that ye were our countrymen. - Samuel Adams
