Blender Git Loki

Temporary remove derivedRender

Couldn't say the idea was wrong, but some nasty bugs were
discovered with the approach and to make life simpler for
now it's easier to do hacks needed for boolean modifier
from the render pipeline by tagging needed operands for
the update.

Update metaballs in a single thread

Mballs evaluation uses BKE_scene_base_iter_next which calls
duplilist for all objects in the scene. This leads to conflict
accessing and writing same data from multiple threads.

Ideally Mballs shouldn't do such an iteration and use DAG
queries instead. For the time being we've got new DAG
let's keep it simple and update mballs in a ingle thread.

Code cleanup

- Add some additional cases to mallocn_intern
to cover FreeBSD and add missing include for OSX.

- Move DAG_init/DAG_exit functions to the top,
they're kind of global and it's not so much
logical to keep them under "Threaded Update"
comment.

Merge branch 'master' into soc-2013-depsgraph_mt

merge with master a930fb154f47a7473848254ad0b0b83c9edd1850

Cycles Volume: materials without a surface shader now act transparent when
there is a volume shader.

This is the behavior you would expect I think and makes setup easier. More
optimizations are possible in this case to avoid surface shading altogether, but
it's probably not so noticeable altogether as this will early out and not get to
any expensive computations as far as I know.

Note that these still count as transparent bounces even if there is no BSDF. Not
entirely happy with that, but there should be some way to limit the depth?

Cycles Volume: remove use volumetrics option.

In general it should not be needed to turn on such features, cycles should
auto-detect if there is a volume shader and just use it. It would be useful
at some point to have global options to turn off effects, but we better group
those options somewhere then.

Cycles Volume: rename isotropic to scatter volume, g to anisotropy, scattering to volume bounces.

Merge branch 'master' into soc-2013-dingto

Skeleton files for all remaining point cache user types, so we can start
replacing the old point cache entirely.

Moved object/modifier/sim resets out of the BKE_ptcache_id_reset
function. This is a design cleanup, the point cache methods should only
write to the DNA when retrieving state data. There were already a number
of cases where such resets occured outside of this function (such as
with rigid body worlds or smoke sims).

The BKE_ptcache_object_reset function still exists and includes these
reset calls, but there the cache user data types are accessed directly.

Added XXX comment about the BAKING flag, left it in for now to not lose
track of the purpose.

Removed the BAKED flag from point cache state. The cached now has a new
flag LOCK_SETTINGS instead, which should indicate more clearly that it
is about preventing cache modification by disabling user settings
changes.

The cache should not have to care about this setting in the first place.
It becomes increasingly difficult to control changed simulation settings
anyway (scripting!), and the cache should not even try to exert control
over sims this way. Eventually this flag should be a feature of
simulations and modifiers, using it as a 3rd option beside
enabled/disabled and using only existing cached data.

Removed the SIMULATION_VALID flag entirely. This was only used by the
bake operators.

Completely removed the "bake" operators for point cache. This feature
will get a complete replacement, removing it altogether makes this much
easier.

Marked the PTC_STATE_SIMULATION_VALID flag as deprecated. This is only
used by the bake operator, but keeping track of the sync state of cache
data vs. simulation settings needs to be radically simplified.

The API should provide a way of detecting whether a valid simulation
state can be obtained for a certain frame.

/* check if a frame can be read from the cache (no data is modified) */
PTCReadSampleResult PTC_test_sample(struct PTCReader *reader, float frame);
/* try to obtain a cache sample and write to DNA data */
PTCReadSampleResult PTC_read_sample(struct PTCReader *reader, float frame);

Beyond that the cache should be oblivious of the simulation state. When
settings or dependencies are changed which invalidate the simulation
cache state this should invalidate all cache samples (or possible a
certain frame range), but the cache itself should not need to keep track
of the simulation as a whole.

Build the archive filename inside the base Reader/Writer constructor
from info in the PointCache. This way we can replace it more flexibly
with other data streams, packed files, memory buffers etc.

Make the archive file name construction based on PointCache settings
internal to the path utility function.

Partial cleanup of the point cache UI panel, using a number of booleans
to store all the stupifying special cases. Not complete and will need
a general overhaul anyway ...

Removed or disabled remaining BKE_pointcache calls in particle_system.
Only DNA read(!) access to PointCache is allowed (otherwise would have
to wrap all the flag checks etc., this could be done later still).

Particle edit mode uses a lot of badly designed point cache DNA writing,
so for now is disabled (or broken).