Refine dlopen wrapper

[Christian Speckner]

I think it would be useful if some more control over dlopen would be available. specifically RTLD_GLOBAL would be useful for external model libraries as it is not unreasonable for those to depend on the same code as parameter.xxx.f90. At the moment, the behavior is unspecified which defaults to RTLD_LOCAL at least on linux systems. E.g., in case of the Three-Site Model, this means that both the external library _and_ the O'Mega library have to include the actual model library, and as this code is localized, both have to initialize their own copy, which is somehow awkward and might produce weird results if the default would turn out to be RTLD_GLOBAL on some other system.

As a side note: currently 1 is used as dlopen flag: is this really portable? Even if it works on linux and OSX, it might be wiser to determine this in the configure phase.

[Christian Speckner]

After some investigation, it turns out that the default behavior already is different on OSX, as are the values of the different flags - on linux, 1 means(RTLD_LAZY

RTLD_LOCAL), while it is (RTLD_LAZY

UNDEFINED) with UNDEFINED defaulting to RTLD_GLOBAL on OSX. I think this should definitely be done properly in order to avoid trouble, e.g. by compiling a small C program which writes a module source containing the proper values to stdout.

[Juergen Reuter]

CS has implemented not the original scope of the ticket, but the functionality described here, in the NLO branch. After this branch has been merged back to the trunk (scheduled for mid-June) this ticket can be closed.

[Juergen Reuter]

All to be done is done in the NLO branch. As soon as this is merged back to the trunk, this ticket will definitely be closed.

[Juergen Reuter]

We merged. All that is needed, is now in the trunk. Closing.