Thomas Lenarz wrote:[color=blue]
> Hello,
>
> please forgive me for posting this if this problem has come up before.
> I could not find any hint with google.
>
> I am porting a bunch of source-code from SUN-OS to AIX at the moment
> and have the following situation (simplified and renamed for better
> understanding)
>
>
> Source a.cc --> compiled and linked into a binary together with the
> shared object underneath
>
> .....
> AnyType *globalAnyType;
> ....
>
>
> Source b.cc --> compiled and linked into a shared object:
>
> ....
> extern AnyType *globalAnyType;
> ....
>
> globalAnyType = new AnyType(....); <----- Segment-Violation here
>
>
> On SUN the code worked fine. On AIX I get no compile-errors and no
> link errors. So I assume external references are resolved.
>
> I found out using dbx that the assignment from "new AnyType(...)" to
> globalAnyType causes the Segment-Violation. The external declared
> globalAnyType appears not to be backed up by memory (dbx shows "nil").
>
> To proove this I replaced globalAnyType by a non-extern-declared
> variable of Type AnyType and....that works.
>
> Does anyone have got an idea on why external binding of a variable
> does not work in this case? Are there circumstances you can think of?
>
> Might special compiler/linker options help?
>
> I have to mention that I port from an old compiler without
> namespace-support to a compiler with namespace-support. I already
> checked if the variables may be in different namespaces because of
> side effects of include files. But I concluded that this would be
> logically impossible.
>
> Thanks for any idea what I could try.
>
> Cheers
> Thomas[/color]
So the new and the constructor succeeded, but the pointer assignment
failed? Is there any funky operator overloading related to AnyType? Can
you look at &globalAnyType (the address of your global) in your
debugger? Try reversing the two declarations by putting the actual
variable in b.cc and the extern declaration in a.cc. Now do you have a
problem with accessing it in the latter?
I'm guessing that this might be a weird symptom of an initialization
order fiasco (cf.
http://www.parashift.com/c++-faq-lit...html#faq-10.12 and
following), and a different order of initialization allowed the code to
work on your other compiler but kills it here. Subtle things can happen
with the order of initialization. See the FAQ for a solution to such
problems.
You may also want to ask in a newsgroup related to your compiler (see
http://www.parashift.com/c++-faq-lit...t.html#faq-5.9 for some
ideas).
Cheers! --M
