Since all Nodes have an output that provides all output values now, we now
never need to add an output from a Node derivative. These changes remove the
last of them and disable the ability to add more outputs from a derivative.
If Nodes only have the one output, we don't need to do so much differentiation
between them. Previous iteration used outputs as like a distinct function
within a Node (e.g. length output would return one result, buffer output would
produce a different result - each run different code to produce their results).
Now in this iteration, it's more accurate to say a Node is just one function
(which seems more appropriate for a node system anyway).
Largely conforming renderers to new NodeValue system. Code seems a lot cleaner
this way which is a nice advantage. Likely non-functional as this won't
compile just yet and still needs probably another day or two of testing to get
it back to where it was before.
We use references to the node directly rather than references supplied by the
object. This should be more robust anyway and will fit better in the new node
structure.
Since the nodes won't be holding any rendering data themselves in this system,
we may as well enforce some level of non-write access by setting all the
functions to const. They were already const-friendly, they just weren't
labelled as such.
Fixes a "pure virtual method" crash. The inputs will try to send invalidate
cache signals through the node while it's being destroyed if these are left
connected. There's no purpose to sending invalidate cache signals since any
nodes affected by this one will automatically invalidate the cache from its
outputs being disconnected and the node being destroyed so this shouldn't
cause issues.
Fixes recurring design issue that the Blocks were a frequent exception to the
DAG concept. The Blocks connecting to each others inputs/outputs while not
necessarily being "dependent" on each other to produce an image continually
causes issues while trying to create a rendering code path. This redesign
should provide a more "directed" approach to the directed acyclic graph.
NodeParams used to be destroyed automatically through the QObject system, but
now that NodeParams perform actions that access the parent Node object in
their destructor, the NodeParam destructor needs to be called before the Node
has been fully destroyed. By destroying explicitly in the Node destructor we
accomplish this.
If a NodeParam is destroyed (presumably as the result of its parent Node being
destroyed), any edge objects connecting to another param should be destroyed
or else the edge will be floating referencing a now destroyed object.
It seems unwise to let NodeInputs take more than one value, but at times
it makes sense to create a list or array of inputs. This class will create
sub-parameters as an easy way to provide variable inputs while also enforcing
one connection per input parameter.
This is also the primary motivation for the previous commit (disambiguating
when a NodeParam's parent is a Node vs some other type).
Old code assumes that a NodeParam's parent will always be a Node. The function
has been separated off and tweaked in the event that this is not the case.
It was becoming inevitable to get rid of this, it was continuously an exception
that had to be made to the DAG concept seeing as blocks are never really
"dependents" on each other, i.e. the images they produce have nothing to do
with other blocks. The only exception is a transition which will also be
easier to accomplish with this slightly different design.
Since we're now working with a separate proxy copy of the original node graph,
if the user changes a parameter in one of those nodes (triggering an
InvalidateCache signal), the values in our copied graph need to be updated
with these new values too.
A signal emitted and received in the same thread will call other functions
before returning to the one that emitted the signal. If these other functions
try to lock a mutex while a mutex is already locked by the emitting function,
we get stuck in a deadlock. These changes ensure that a signal is never emitted
by any function until all the nodes locked by it are unlocked.
Previous iterations would use mutexes to prevent changing of the graph
mid-render, however several user actions would need to capture these mutexes
causing the main thread to hang until the current render job (frame/range of
samples) was complete. We now copy the nodes necessary as part of the "compile"
process so that the main thread shouldn't need nearly as much blocking while
caching occurs.
Useful for detecting when a graph needs recompiling, in tandem with the
"InvalidateCache()" signal when necessary, this signal ripples through the
nodes when any of the connections change which will likely need handling by
the renderer.
Sometimes we'll be copying nodes without wanting to copy their connections,
since we'll want to connect them to equivalent copies rather than connecting
them to the same nodes the originals were connected to. We now have an extra
parameter to distinguish such operations.
Both audio and video renderers were working off the same invalidation signal
(i.e. changing audio would also trigger a re-render of the video). This is
obviously suboptimal and now they are separate.
Many changes were made throughout the codebase to support audio, these are
most of the small changes necessary.
The audio support still is not perfect. I still need to write in resampling
support. After that it should work correctly with all audio types.
commit
For testing the new iteration, the texture cache disk download was written
into the main thread instead of into the separate threads. Now they're back
in separate threads again.
Also I think some of these files probably should have been in the previous
commit.
Once again, conceptually this system should work, however it does not seem to
be the most efficient and it wouldn't surprise me if the multithreading was
eventually upgraded to an even more coherent system one day. However for
"core principles" this should be fairly decent.
The new rendering pipeline strives to simplify the nodes themselves as much
as possible and move much of the logic to an external rendering engine. This
change removes all of the responsibilities that no longer belong to the
nodes themselves and will soon be folded into the renderer.
This design, while not completely illogical, seems to betray the DAG design
and borders on impossible to actually recompile. The plan is to build this
functionality into the renderer instead.
itsmattkc