[ On Wednesday, November 24, 2004 at 07:05:56 (-0500), Steven J. Dovich wrote: ]
> Subject: Re: Threading problems 
>
> To solve the pkgsrc issues, you cannot portably dlopen() a threaded
> object from a non-threaded application. Threads support has to be
> in place before main() is called, and probably before C++ static
> constructors get invoked.

Indeed.

This "threat" of some program someday eventualy dynamic loading a module
that expects to be able to start threads when the program that loaded it
wasn't already expecting to need a fully thread-safe environment really
must be stomped out.

I for one am extremely opposed to forcing everything into paying the
cost of running in a fully thread-safe environment just on the off
chance that somehow some thread-needy module might be loaded by some
other associated application.  Those who want the freedom to plug&play
with anything any time might be willing to pay price, but for those of
us who have well defined requirements which allow us to know in advance
when and where we'll use various components there's no need.

-- 
						Greg A. Woods

+1 416 218-0098                  VE3TCP            RoboHack <woods@robohack.ca>
Planix, Inc. <woods@planix.com>          Secrets of the Weird <woods@weird.com>