[Openicc] ALL YOU NEED IS A PROFILE, THE MYTH. (WAS CC Profiles In X Specification and dispwin)

[Alastair M. Robinson]

Hi,

Robert Krawitz wrote:

> Actually, in the last release I checked in something to allow saving
> MD5 checksums from test runs (using the test pattern generator, not
> the CUPS driver).  We're not currently using it for anything, but it
> could be used to create a test that would verify predictability from
> release to release.

That could be very useful.  The only question that springs to mind - are 
any of the dither algorithms stochastic, or are they all strictly 
deterministic?

> I don't want to make something like this a release requirement,
> because we might want to make changes not related to output and having
> an absolute no changes policy may be too restrictive.  But we might be
> able to use this to at least flag changes.

Well one of the biggest problems currently is trying to figure out which 
changes will impact what.  If when distributing linearizations or 
profiles we can tag them with the MD5 sum of a repeatable test, we can 
at least verify compatibility.

> Even in that case, changes wouldn't necessarily mean that a profile or
> linearization would be invalidated.  For example, output changes
> related to High Accuracy color correction may be of no interest to
> people doing custom linearization or profiling, because they shouldn't
> be using that mode.

True.  So a "signature" to be used in this manner would be best computed 
from output done in Uncorrected mode.

> Likewise, changes to default drop sizes at 360
> DPI might not be interesting, and if we expose the drop size and light
> ink transitions, retuning these constants may also not be important to
> these users.

If such parameters are loaded from the linearization itself, then yes.
So the signature would be computed *with* the linearization applied.

> I very much like the idea of being able to hand off final rendering
> quality to others.  I think that many users won't worry about this, so
> having good (if not perfect) default tunings is important, but so is a
> way of letting people who want to perfect the tunings do so.

If those producing the "perfect" tunings share the results, these can 
*become* the defaults for future releases.  Provided the behaviour *with 
a particular linearization over-riding the defaults* doesn't change, the 
defaults can change, I think.

All the best,
--
Alastair M. Robinson