[cairo] Xlib/Xcb and EXTEND_PAD

[Zack Weinberg]

Thomas Jaeger <thjaeger at gmail.com> wrote:
> I've done some more (very informal) testing, patching the intel driver
> to disable acceleration for Reflect and Pad, and I couldn't (browsing
> in firefox) tell a difference between acceleration and the server-side
> fallback.  So having the fallback in the client really hurts more than
> anything.

I agree.  The XGetImage/XPutImage cycle is what made us turn it off, not
the cost of not having this accelerated.

> > I assume you mean >= 1.4.99?
> > 
> >> then fix the drivers when problems become apparent (where fixing
> >> can mean resorting to the server's software fallback for now).
> > 
> > That might be okay for cairo in the abstract, but I'm not sure it
> > works for Mozilla, how long is it going to take for up-to-date
> > drivers to get into the hands of our userbase?
> 
> That's why I suggested 1.5.99 as the cut-off point.  xserver 1.5 is
> already too widely deployed to be doing experiments like this, but I
> think we have a chance to get the driver issues sorted out before the
> next round of distributions.

Oh, ok.  I thought you had typoed 1.*4*.99.  Yes, if we push on this it
oughta be fixable by then.

> Making this a hidden pref for mozilla will of course be fine as well,
> although the distributions can also just patch this themselves.

In general I think the last thing mozilla needs is more hidden prefs :-/

It would be really nice if cairo could *tell us* when an operation
will involve a client-side fallback; then we could keep that image data
on the client side until after that point and avoid the server
round-trip.  (I don't know what the situation is on Windows, but I have
the impression there are similar costs to copying in and out of a
Quartz surface.)

zw