Lack of Specifications (was my work...)

Werner Almesberger werner at
Fri Jun 6 13:02:02 CEST 2008

Andy Green wrote:
> I should've just sat on my hands.

The hidden benefit of having an upstream to merge with: you can deflect
a lot of silly ideas with "I'd *love* to do this, but there's absolutely
no way we could get this past the evil guys on lkml." ;-)

> This pretty nicely makes my point, we should not be arguing at this late
> stage about simple device behaviours that should have been specified
> from the start.

Yup. Cf. resume on AUX. That plane's taken off half a year ago.

  This is a huge morale suckage that puts us in a place
> that actually doing anything that has to make any assertions into any of
> the huge undefined spaces invites nothing but armchair opinions about
> how it should have been done differently.

Worse yet, technical reasons for the way things currently are can be
overlooked when making hasty changes. Then, perhaps months later,
this great new feature is falling apart and you realize that it can't
be done, and everyone would be better off not having had it in the
first place. (Most likely, someone would have looked for and found an
alternative by then that was at least as good.)

> In particular when it comes to the next platform, we can participate in
> specification formulation but ultimately once that is done, the decision
> about MPU or not falls out of the needs of a truly detailed functional
> spec, not out of your personal opinion formed some distance from, eg,
> pcf50633.c.

I think there's both to it. Even a very detailed spec sometimes doesn't
tell you that you need something like an MPU, particularly if you also
want to use it to get rid of glue logic. (Glue logic is treacherous. It
looks simple, but can have amazing ramifications.)

So there's also a bit of "engineering instinct" involved. If several
experienced engineers that don't have too much of a common history (i.e.,
no groupthink) share the gut feeling that an MPU would make our lives
easier, there's a good chance that they're right.

- Werner

More information about the openmoko-kernel mailing list