Why care about indirect rendering ?

[Olivier Galibert]

A large part of the current discussion seems to be about what
interfaces to use to do rendering in the server of graphic X commands
sent through the X protocol.  Shouldn't that be seen as obsolete at
that point?

We have DRI which somehow[1] manages to get the per-application
rendering commands directly to the video card while still
communicating with the server for the rest (window management, input
devices multiplexing...).  If we have a libX11 that did the same for
its rendering commands, would there be any point to still do any
rendering in the server?  The network transparency (which I can't live
without) could be delegated to a local direct-rendering client.

Of course, that would make DRI support in the kernel mandatory, but I
don't think requiring a hardware device that does irqs and dma to have
a minimum of kernel support is totally unheard of.

So shouldn't the target be not exactly a X protocol->GL translation
but instead a libX11->some entry point in the DRI/Mesa chain?  I
suspect the external vendors wouldn't mind having all the request-card
commands translation done in only one point.

  OG.

[1] Haven't looked at the actual code yet