On Tue, Feb 04, 2014 at 02:31:19PM +0100, Lars-Peter Clausen wrote:
currently find in drivers with more than one interface. And there is the thing that you can actually build a config that results in a build error which triggers the occasional randconfig build failures for for ASoC. Once the last three drivers that still soc-io have been
Those randconfig failures are for configurations that no human would ever generate, it's really hard to care.
converted to regmap I think we should move the selection of REGMAP_I2C and REGMAP_SPI to the individual drivers rather letting the core select them unconditionally. Using this new scheme will
That's the idea, it does need the drivers handling first though. The fact that not everything works through select (there's the ignored dependencies thing and I think some other stuff) isn't helpful with this stuff either, it makes it all too fragile.
make it easier to avoid the situation where you end up selection REGMAP_I2C or REGMAP_SPI even though I2S or SPI_MASTER is not
The core only enables the regmap code if the bus is being built, we don't have a problem with that bit - that bit at least is fine.
selected. I don't have a problem with updating the other drivers to use the new scheme, but it's not like we have to update them all at once in one atomic step. And you have to start somewhere.
Right, but that somewhere shouldn't be doing it in a new driver without even mentioning that it's being done - it's not something that should be coming as a surprise in the middle of a review of something that should be unrelated. That's the biggest problem here, it wasn't even called out in the changelog so people wouldn't be able to figure out why this driver was being different.
An explicit conversion patch for at least one driver (possibly even this one on the end of the series) would be ideal, or adding a note in the changelog saying something like "We should do this in general because X but I didn't get round to starting yet" so that it's visible.