Library delegation - should the path to parent library be the canonical path?

[Jonathan Gibbons]

Mark Reinhold wrote:
>> Date: Mon, 19 Apr 2010 14:26:49 +0100
>> From: alan.bateman at oracle.com
>>     
>
>   
>> I've been playing with the module library delegation and it looks to me that
>> the path to the parent library is stored as the canonical path at the time that
>> the library is created. I wonder if this will be a problem. Say in a cluster
>> where the canonical path might differ depending on which node that has mounted
>> the file system.
>>     
>
> Canonical paths aren't node-dependent in NFS clusters.  Are they that way
> in any other distributed filesystem?
>
> These paths don't absolutely have to be canonical, I think, but whether
> canonical, absolute, or possibly relative it will always be the case that
> moving non-leaf libraries around will be problematic.
>
>   
>>                  Also, I assume it means that a library cannot be moved to a
>> different location.
>>     
>
> Leaf libraries can be moved trivially.  My expectation is that nearly all
> non-leaf libraries will be system-defined, or at the very least always
> placed in well-known locations.  So I don't expect this to be a huge
> problem in practice.
>   

I think it will be very common for developers to have transient 
libraries.  I agree that moving them will not be a problem, but at least 
as a developer I think that such transient libraries will be if anything 
more common than system libraries.


>   
>>                     On that, does jmod need a "reparent" command to aid fix up
>> in the event that a library is moved?
>>     
>
> Yes, that's a good idea.
>
> - Mark
>