On Sat, 14 Feb 2009, Dan Fandrich wrote:

> I don't see any solution other than moving the call to gnutls_global_init()
> back into the only safe place, namely curl_global_init(). The problem with
> curl --version failing would have to be dealt with some other way, perhaps
> by making --version a special case and skipping the global init when it is
> given.

I completely agree. We can't sacrifice proper libcurl behavior only because of
an edge-case curl tool desire.

-- 
  / daniel.haxx.se