solib: URL patch
Bryce McKinlay
bryce@mckinlay.net.nz
Wed Aug 13 23:36:00 GMT 2003
On Thursday, Aug 14, 2003, at 05:23 Pacific/Auckland, Jeff Sturm wrote:
> On Wed, 13 Aug 2003, Andrew Haley wrote:
>>> Does there need to be any CNI mechanism for this? Or is Java's
>>> Class.forName() etc. adequate to load/invoke the solib?
>>
>> I wouldn't have thought so, no. At present we have a direct way of
>> invoking Java methods in a shared object from C++ code; do we want to
>> destroy that ability?
>>
>> Unless, as was suggested, we alter the C++ compiler to use the
>> indirect dispatch scheme...
>
> I don't have a strong opinion. Indirect dispatch is probably less
> important for CNI than Java code, but someone may have a good
> reason for needing/wanting it.
Eventually we'll probably want Java ABI support in C++. If CNI is aware
of the binary compatibility ABI then, aside from the obvious benefits,
we can make all the symbols private, which should improve link/load
times and reduce the size of binaries. I think we ought to get the Java
side of things fleshed out first before worrying too much about CNI
though.
>>> I'd marginally prefer "sharedlib:" over "solib:".
>>
>> DSO is becoming the de facto name, AFIAA. "dso:" might be best of
>> all.
>
> That might be a little surprising to Windows users, but yes, I tend
> to prefer dso: over solib: as well.
Mac OS X calls them "dylib"s. I'm not too worried, but how about just
lib: ?
Regards
Bryce.
More information about the Java
mailing list