[Intel-gfx] [PATCH 00/59] Remove the outstanding_lazy_request

[Chris Wilson]

On Fri, May 29, 2015 at 07:37:19AM -0700, Jesse Barnes wrote:
> On 05/28/2015 02:20 PM, Chris Wilson wrote:
> > On Thu, May 28, 2015 at 01:02:51PM -0700, Jesse Barnes wrote:
> >> John, Tomas, where are we with this series?  I believe this is a
> >> prerequisite for the request->fence conversion, and also the native sync
> >> work, both of which I need for some stuff I'm doing.
> > 
> > It's a 60 patch series for ~800 line patch performing a single API
> > conversion. I call that unreviewable.
> > 
> > It also misses the *bugfixes* and improvements from the original.
> 
> Can you be more specific?  The patches are relatively small, is it the
> interactions that are tough to review?  And what bugfixes are missing?
> Maybe you mentioned them on the previous thread, but I don't have it
> handy to dig them out...

The point is that the new API is never demonstrated, there is no natural
cascade of consequences of why it should be so and so the overview of
what the API should be is lost in the noise of so many patches. The
purpose here is to review the API, not the mechanical adaptation of the
code.

If you have been reading the patches, or even reviewing bug reports and
looking at regressions, the missing pieces from the first patch would be
fairly clear. The major point is that some of the bugs only become obvious
when the API is not changed piecemeal, because the new API should enforce
correctness and hence require the code to be fixed. The patch series fails
in that regard.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre