What could it mean for a visual to have more than one depth?

[Adam Jackson]

I've been looking at cleaning up pixmap format and visual
initialization. I'm having some difficulty understanding why the code
is currently doing what it's doing. Perhaps someone has an idea.

In the initial connection block, a DEPTH is a tuple of a depth and a
list of visuals. A VISUALTYPE is a tuple of (visual-id, visual-class,
rgb-masks, bits-per-rgb, colormap-entries). Note that a VISUALTYPE does
not specify depth. The protocol contains the following enigmatic text:

"A given visual type might be listed for more than one depth or for
more than one screen."

Assuming that "visual type" here means a VISUALTYPE - which seems
pretty reasonable - the interpretation for more than one screen makes
sense. The visual ID and all its properties happen to be numerically
equal on more than one screen, and this is never ambiguous because
windows are always created relative to a particular root window.

Listing a visual for more than one depth, though, I can't figure out
what that could possibly mean. Maybe if you make a depth-16 window on a
Visual that can also do depth-24, then PutImage will automatically do
the obvious color expansion? (Or the reverse.) But if you have that,
what do you put in rgb-masks to make it make sense for both depths?
Maybe instead this is for PseudoColor hardware that wants to expose
(say) both depth-4 and depth-8, and have the initial 16 colormap
entries shared between depth-4 and depth-8 visuals, so that apps only
expecting depth 4 can work transparently?

More practically, the existing drivers - and every server I've ever
seen - only ever have one depth per visual, and applications are by now
quite prepared to deal with having tons of visuals thanks to GLX and
Composite. Would enforcing a 1:1 map from Visual to Depth be an
implementation burden on the server side for any reasonable class of
hardware?

- ajax