Roderich Schupp wrote:
Why don't we get rid of the whole name mangling scheme and extract everything
with its "original" pathname as given in the zip? This would eliminate
a lot of (often duplicated) code. Even the ugly dynaloader hack could go
or am I missing something here?
At first glance, we have one of two simple choices:
- Extract everything into the same directory:
=> dll's being named after the last part of the module in question, we
will get lots of clashes of files called "XS.dll" (dll == so == ...)
- Extract everything in the usual directory layout. Append all the
directories to PATH or LD_LIBRARY_PATH.
=> limited env variable length is a problem.
Now, the DLL's for XS modules needn't be in PATH/LD_LIBRARY_PATH. So if
we do this:
- extract those in the usual directory structure
- extract /shared/*.dll and /*.dll to a special dir in PATH/LD_LIBRARY_PATH
Then we should be fine, right? Of course, extraction will be a bit
slower due to the extra directories. What else? You can't link against
those XS libs with -lFoo. But I'm not aware of anyone doing that, but
they all go via the name, right?
Cheers,
Steffen
