[OpenAFS] Re: [OpenAFS-devel] exposing RPC code<->name mappings via rxgen extension, a library, and a new utility

Jeffrey Hutzelman jhutz@cmu.edu
Wed, 06 May 2009 10:02:27 -0400


--On Wednesday, May 06, 2009 03:05:39 PM +0200 Felix Frank 
<Felix.Frank@Desy.de> wrote:

> A proof of concept implementation is already running in DESY's
> OpenAFS+OSD test cell.

I'm confused.  I thought we were talking about supporting debugging tools 
like translate_rpc and perhaps rxdebug.  What exactly do you have "running" 
in your cell?





>> So the consesus is to
>> - generate very simple translation functions and

If by "very simple" you mean that id->name translation works exactly like 
ExecuteRequest, with a combination of switches for sparsely-populated space 
and dispatch arrays for densly-populated space.  This is very important for 
time- and space-efficiency, which is why rxgen constructs the 
ExecuteRequest function the way it does.  Given that we _have_ this code 
and it is actually _used_ in every service for dispatching RPC's, we know 
it is right and it should be easily reusable.  That's why I've been 
suggesting modifying rxgen to emit the new translation function (preferably 
as part of the libraries it already generates), rather than writing 
something new from scratch.

And of course, if by "very simple" you also mean that name->id translation 
does not exist in this form at all, since the only thing I can imagine 
needing it is a tool like translate_rpc, which will use the tables.