netbsd@precedence.co.uk said:
> http://www.wiili.org/forum/bluesoleil-solution-for-not-supported-adapte
> rs-t794.html 

Ah thanks - I concur.

> Given this coupled with the evidence above that this isn't a single wrong 
> device, I think the best we can do (without overhauling the whole 
> device/class matching intrastructure) to make aue(4) pickier is match 
> class=UDCLASS_IN_INTERFACE as per my previous patch. If I write a more 
> verbose comment, can we agree for me to commit it?

I think this is OK. I'd however limit the class=0 check to the
known problematic 050d/0121 ID: comment out that table entry
in the big table and add a single match for vid/did/class.
While 3 data points are good, who knows what the other 15 or so
vendors have done. At least an 0xff wouldn't be illegal.

The reason I'm so uneasy about that is that I'm thinking
about on-demand loading of drivers. For that, the matching
should follow simple rules so that all information is in
tables which can be shared between kernel and userland.

best regards
Matthias