…an codebase, for more general consistency with other renderer code

In general, moved #include of CoronaGraphics.h out of other headers

Sometimes a forward declaration was enough

Shader's DrawState now uses a const CoronaShaderDrawParams* (and points to a static dummy when none is assigned)

The geometry extension attribute type is stored as a U16 (rather than an enum), and cast back in the cpp

Similary, IsFloat()'s definition is moved to cpp-side

A program's extra uniforms are now retrieved through a reference rather than on the stack, the type being defined on the cpp-side

Command buffers' AddCommand() now takes a pointer, and stores the result in a LightPtrArray (owned by the renderer)

A program's shell transform cleanup is stored in non-typedef's form (equality with the typedef is asserted on an is-same trait

State blocks and commands are now stored in PtrArray values, themselves held in a forward-declared structure defined on the cpp-side

The pending commands are now a LightPtrArray

Whoops, scratch that :D There is now a synced counter, rather than a command count, and values after the synced counter are forwarded to the other buffer (then the counter is updated)

Removed CustomOp from Renderer (wasn't used any more)

Bug found in Array's PadToSize() method, where old data wasn't copied over and then removed on growing the buffer

…e and Group subtypes), converting it from a struct to class with private members

Using debug new on Windows (in debug mode, of course)

Fixed leak in ShaderData

Redid some stuff with std vectors and strings to use Array and String types, in a couple list functions (need to test compiles and runs)

Added new files to projects

… the library's PushImage() calls

CoronaGraphics.h includes revised for Apple projects

32-bit version used by format extension lists

…es, now using a hash to enforce validity

Brought in wyhash (right now just inlined in CoronaObjects.cpp) for updates to same and an RNG

Revised CoronaGroupObjectGetParent() and *GetChild() revised to accept stack object as out parameter, in keeping with no-storage policy

A couple state blocks functions revised to early-out if list empty (to avoid setting up box scope)

Objects all passed by value

No CoronaStackObject, since now superfluous

Types are lazily gathered

No actual box object, only reified in array

And of course, all of it crashes anyhow :(

Now using hijacked pointers rather than pushed-by-value structs

These contain a xor'd-with-hash version of the value in a scope's array (quick to check and robust against absence)

Using wyhash, now in external

Rename of ObjectBoxList to ObjectHandle

Slimming down of previous details

External refs and engine-supplied ones use same array, but the former set a flag

Rather than return a handle, often the same one, and also due to the odd reference thing, just reserve a slot with CoronaObjectGetAvailableSlot()

Same is now an out value in GetChild() and GetParent()

These slots are now encoded in a bit of the box pointer, before mixed with the hash

Some redesign around the above

Added Clear() method and an RAII type that can invoke it

This is for the possiblility of customizing some method scopes (restrict to if clause versus both before and after calls, say)

On that side of things, made object params booleans full ints and reordered them (needed to fix offsetof() in GenericParams code too

Now have declarations and also deferred Add()'ing

Using optional separated scopes for display objects, only loading them if their before or after is present (cuts down on spurious hashes)

Those use the ClearIf RAII objects (initialized with a separateScopes member)

OnMessage handlers may also opt to retain the current scope

Switched some allStored checks to asserts (only really needs checking in open situations)

…ta) and replaced with one to set the geometry writer, a callback

Same in Main.cpp and Rtt_ApplePlatform.mm

Added writer list to renderer, which defaults to a memcpy-based one

Can supply others (tested with text objects; should be in supplementary diff in PR)

Minor changes in CoronaShaderDrawParams's before and after functions to allow injection of custom writers

Vertex copying in renderer logic slightly adjusted for writer list approach

…ese as well as wyhash)

Reworked how TypeNode was initialized since static member being shared among derived types was broken; seemed to work with Visual Studio build but maybe just not stress-tested enough (still untested)

Returning nil if extended vertex data not present rather than creating it just to read it (still untested; related to _properties)

Rtt_STATIC_ASSERT() likewise

Added Rtt_ObjectHandle to various platforms (and removed box list from Apple ones)

Added a few that I missed in Linux projects (I assume these are IDE-related)