🚀 go-pugleaf

Several weeks ago I posted a notice that modules in libraries were not
initializing static constructors under g++ 2.2.2 on a SPARC, though
they did so with g++ 2.2.2 on a DECstation.  The details are included
below.  Since I received no feedback at that time, and I need several
library modules which have such constructors for a major project, I've
decided to post this query.  Am I doing something improper?  Is there
a workaround for this, short of compiling all modules with static
constructors and linking them in directly?

Brad Keister
Department of Physics
Carnegie Mellon University
Pittsburgh, PA 15213
(412) 268-2772 (office)
(412) 681-0648 (FAX)

internet:	keister@poincare.phys.cmu.edu
BITNET:		keister%poincare.phys.cmu.edu@CARNEGIE


>> A library which I have developed makes occasional use of static
>> constructors (which generate constants use frequently in
>> applications).  I've just noticed that the SPARC version of g++
>> generates code which initializes those constants to zero, but
>> initializes correctly if the .o file is linked in directly.  If I do
>> the same thing with the DECstation version of g++, I get correct
>> initialization from the code in the library, and the same is true
>> using cfront on the DECstation.
>> 
>> The code which uses the numbers from the static initialization is not
>> itself in any constructor; in fact, it's in main().  I may be
>> misunderstanding the warnings about implementation dependence of
>> static constructors, but I'm surprised that two machine versions of
>> g++ (both 2.2.2) handle it differently.