On Mon, Mar 31, 2014 at 10:37:39AM -0600, Stephen Warren wrote:
I really wish we would make up our minds about this.
For I2C (and SPI and perhaps others) the I2C match table works fine as a replacement for the of_match table. The only issue might be different manufacturers with the same chip names. If this is a problem, why is fallback to the I2C match table even allowed any more; we should mandate that OF matching only works via the OF match table.
When DT was young, Grant tried to require of_match for everything for completeness, and then I tried enforcing that for reviews, and then Grant said not to bother with that, so I stopped, and now you're saying it's required again. I really wish I could get consistency in how this kind of thing is supposed to work. It's difficult for contributors to know what to do if reviewers keep flip-flopping over time.
Well, *I've* not been flip flopping on this, frankly I was unaware that anyone thought it was a particularly good idea to actively not include the match table. It's true that as a matter of practicality you don't need to bother at the minute but I think especially once you're adding any explicit code at all to the driver the explicit match strings ought to be there too.
I suspect this may have been a pragmatic suggestion due to all the complaints about churn generated by DT.