Re: [PATCH] USB: typec: tps6598x: use device 'type' field to identify devices

[Heikki Krogerus <heikki.krogerus@linux.intel.com>]

Hi Roger,

> > Why not just match against the structures themselves?
> > 
> >         if (tps->data == &tps25750_data)
> >                 ...
> 
> Then you need to declare tps25750_data and friends at the top of the file?
> 
> A better approach might be to have type agnostic quirk flags for the special
> behavior required for different types. This way, multiple devices can share
> the same quirk if needed.
> 
> e.g.
> NEEDS_POWER_UP instead of TIPD_TYPE_APPLE_CD321X
> SKIP_VID_READ instead of TIPD_TYPE_TI_TPS25750X
> INIT_ON_RESUME instead of TIPD_TYPE_TI_TPS25750X
> 
> Also rename cd321x_switch_power_state() to tps6598x_switch_power_state().

No. Functions like that isolate cd321x specific functionality into an
actual "function" just like they should.

Quirk flags mean that if something breaks, it will almost always break
for everybody (there is no real isolation with quirk flags), and when
things are fixed and when features are added, we are forced to always
"dance" around those quirk flags - you always have to consider them.

Platform/device type checks are just as bad IMO, but in one way they
are better than quirk flags. There is no question about what a
platform check is checking, but quirk flags can so easily become
incomprehensible (just what exactly does it mean when you say
NEEDS_POWER_UP, SKIP_VID_READ and so on (you would need to document
those quirks, which is waste of effort, and in reality nobody will do).

In case of tipd/code.c, it should be converted into a library that
only has the common/shared functionality. CD321, TPS2579x, TPS6598x
and what ever there is, then will have a glue driver that handles
everything that specific for their controller type.

Before this driver is reorganised like that (any volunteers?), we'll
have the PD controller type checks, but quirk flags we will not have.

In general, you should only use quirk flags if there is no other
way to move forward - they are the last resort. They are dangerous,
and even in the best case they reduce the maintenability of the code.

thanks,

-- 
heikki