>>  - Why isn't (at least thats my impression from your talk with Charles) before
>>    any file system access is done the name of a searched module matched 
>>    against everything in the list of already loaded modules?  (Windows does
>>    cheat here and only compares the basename, which is enough in any sane
>>    setup anyway)
>
> That's what exactly my patch does. If you take a look at the code, you'll
> know that unconditional file system access is performed with the current
> code. My patch addresses that.

No, your patch does not solve that problem.

And, in fact, you can't do that anyway.  The ELF semantics allow two
modules with the same basename to be loaded, if the modules that
require them have different embedded search paths.

This is why I said the cache should be associated with the search
path.