[Openicc] XICC specification draft

[Craig Ringer]

On Fri, 2005-06-24 at 09:07 +0100, Ross Burton wrote:

> Yes, this could be an issue.  At one point I was tempted with adding
> _ICC_PROFILE_FILE, which is a filename to the profile on disk (with the
> requirements that _ICC_PROFILE must be set of _ICC_PROFILE_FILE is).
> This would save traffic if the profile was large, but wouldn't save any
> memory as (lcms at least) creates a new data structure from the profile
> data, allowing the transfered data to be deleted.

It'd also be a nightmare for remote X11 use. The last thing we want to
be doing is introducing yet more new and interesting ways to break
remote X11.

Note: I use X11 thin clients at work, so I have a bit of a vested
interest in this. However, my view is that remote X11 is not just a
technical curiosity, and I suspect it'll be seeing increasing amounts of
use again now that thin clients are coming back "into fashion".

> Unless of course there is a way of working with profiles directly from
> their on-disk format, in which case applications could mmap() the file
> into memory if it exists.

Still doesn't help for remote X though. Maybe that can be left to the
app/toolkit - 'if this atom isn't locally cached already, cache it to a
file; then work with the profile by read-only mmapping the cache'.

--
Craig Ringer