> > 1. Missing checks for error conditions that the spec requires an > implementation to check for. > > 2. Newly-added state is not being saved/restored properly by meta-ops. > > 3. Overzealous error checks: some error condition is not being checked for > in the correct way, causing an error to be generated even for correct input. > > 4. API Functions neglecting to set ctx->NewState when necessary, so the > driver fails to notice that state has changed. > > IMHO, Ian's proposals, or some variant of them, would be useful because they > directly address pain points like the ones above. For instance, requiring > complete vertical slices of functionality would help with 3 and 4, because > it would ensure that at the time state was added to the front-end, we could > easily test that things propagated properly to the back-end. And requiring > a robust set of tests would help with 1 and 2, because any adequately robust > set of tests for a feature would include testing error conditions and > meta-op behaviour.
The meta-op stuff is a misnomer, only Intel use the metaops in terms of drivers people are developing. also I think the gallium stuff won't always mean we miss a ctx->NewState, until someone cleans up a lot more :) Dave. _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
