Re: wined3d: Remove duplicate assignment of NULL to map_ptr in wined3d_buffer_map.

[Henri Verbeet]

On 17 February 2017 at 04:43, Sebastian Lackner <sebastian(a)fds-team.de> wrote:
> BTW: Is someone currently working on wined3d_buffer_map? To me it looks like
> there are multiple issues related to location handling in there.

The code certainly has issues, at least in part that's because it
mostly predates the current resource location management scheme.
> * For mappings without !(flags & WINED3D_MAP_READONLY), shouldn't the code invalidate
>   the sysmem location, to force a download when it is required later?

If you're referring to the case I think you're referring to, no,
because WINED3D_BUFFER_PIN_SYSMEM isn't set and the system memory
location has been evicted.
> * In the "Falling back to doublebuffered operation." case with
>   !(flags & WINED3D_MAP_READONLY) the buffer location is never invalidated, which
>   means no upload will happen. For testing purposes, this can be triggered by
>   replacing the if conditions and running d3d9 visual tests. Adding something like:
>   buffer->locations &= ~WINED3D_LOCATION_BUFFER; fixes it (using a function would
>   add the modified range twice).

That path should only be reachable for read-only maps, in which case
there's nothing to upload.

The general issue that the code is overly complicated and hard to
reason about is certainly true though.