C++ SDK issues

[awaytools]

C++ SDK issues

[bboltz]

Another issue:

The C++ SDK only writes out 2.0 AWD files. However, the AwayBuilder Live Tool writes out 2.1 AWD files.
I would like to see the SDK write out the same version as AwayBuilder = v2.1.

Version 2.1 has support normal maps and lightmaps, while v2.0 doesn't.

Thanks

[bboltz]

Another issue:

When exporting large meshes with verts/trigs greater than 65536, the AWDGeomUtil tool is not splitting the mesh into submeshes. It's writing out a single submesh with the vertex index field set to AWD_FIELD_UINT32.

By itself, I have no problem with it, but when I try to load this model into AwayBuilder Live Tool, it doesn't render the mesh properly. The mesh is deformed, vertices are pinned. I'm guessing it's ignoring the field type.

[80prozent]

Hey Bolt1

i am working on the c++ sdk (libawd) for some days now. excpect more changes/updates in the next days/weeks.

about your issues:

The AWDMetaData class allocates its own memory for generator_name and generator_version,
but I don't see it de-allocating this memory anywhere.
The override_encoder_metadata() function doesn't attempt to delete any previous allocated memory.

• yes, i need to go over all the classes-destructors, to make shure all memory is de-allocated correctly. this is on my to-do.

The Importer side is completely missing. The SDK doesn't import AWD files.

• yes, this will done once the exporter is finsihed and working for all awd2.1 features.

The C++ SDK only writes out 2.0 AWD files. However, the AwayBuilder Live Tool writes out 2.1 AWD files.

• i allready uploaded changes to export 2.1 files. there is still a lot of work to do, to have all the 2.1 features exported, but the basics are there.

When exporting large meshes with verts/trigs greater than 65536, the AWDGeomUtil tool is not splitting the mesh into submeshes. It's writing out a single submesh with the vertex index field set to AWD_FIELD_UINT32.

• i havnt tested with very large meshes yet. Of course you are right, and the AWDGeomUtil should take care about that. Thats connected to the issue, that no multipla materials are exported per mesh. This is one of the features i am working on right now...

[bboltz]

Thanks for the reply.

I'm looking at the AWD example files that come with away3d-examples-fp11
(MonsterHead.awd, PolarBear.awd, tictac.awd, etc...) and it seems there's a problem
with their geometry data streams. All of the vertex index streams have their field set
to AWD_FIELD_UINT32 (6), when it should be AWD_FIELD_UINT16 (5). These are all v2.0 files.

When I save these out using the AwayBuilder LiveTool, it sets the proper field
type, AWD_FIELD_UINT16. The C++ SDK also sets the proper field type. However, the only
problem with this is when you're exporting large files. Re-writing MonsterHead.awd with
the C++ SDK will write out a proper field of AWD_FIELD_UINT32 with large indices.
But, as I mentioned, the AwayBuilder LiveTool ignores it, and you get a mess.

So, all of the assets will be bugged if the LiveTool was changed to work correctly. :)
Fixing geomUtil to split submeshes will temporarily fix the problem, but as long
as the AWD loaders don't respect the field, the problem will still remain.

[bboltz]

BTW, thanks, I missed the Jan26 update. I downloaded the new SDK C++ and tried it out, but it crashed on me. It looks like the SDK is now writing out the metadata block, however, there are some memory bugs/leaks.

• There is no metadata class destructor declared.
• generator_name/generator_version is not initialized to NULL.
  This causes prepare_and_add_dependencies() to crash as it thinks these strings exist.
• The AWDMetaData class doesn't delete generator_name/generator_version.
• The AWDMetaData class doesn't delete encoder_name/encoder_version.
• The awd.cpp destructor is not deleting this->metadata.
• AWD::set_metadata is not deleting any previously allocated memory.
• I made some changes to get to working, and exported a test file.
  Looking at the AWD file, the metadata block was written with a block id of zero.
  I assumed all blocks start with an id of 1, since the zero id represents unused state?
  I have noticed the AwayBuilder LiveTool writes out metadata with an block id of 1.

[bboltz]

Found another metadata bug:

this->encoder_name needs to be allocated with ASCIIZ null terminator. Otherwise, functions like strlen won't work properly.
this->encoder_name = (char *)malloc(6+1);

You may want to check to see if snprintf is terminating the other string with NULL.

This line won't set a NULL terminator:
memcpy(this->encoder_name, "libawd", 6);
This is causing "libawd" to be written with a bad string length.

[bboltz]

My changes:

class AWDMetaData :
    public AWDBlock,
    public AWDAttrElement
{
    protected:
        char *encoder_name;
        char *encoder_version;
        void prepare_and_add_dependencies(AWDBlockList *);
        awd_uint32 calc_body_length(bool);
        void write_body(int, bool);

    public:
        AWDMetaData();
        ~AWDMetaData();     // ADDED
        char *generator_name;
        char *generator_version;

        void override_encoder_metadata(char *, char *);
};


AWDMetaData::AWDMetaData() :
    AWDBlock(METADATA),
    AWDAttrElement(),
    encoder_name(NULL),     // ADDED
    encoder_version(NULL),
    generator_name(NULL),
    generator_version(NULL)
{

    this->encoder_name = (char *)malloc(6+1);   // CHANGED
    memset(this->encoder_name, 0, (6+1));   // ADDED
    memcpy(this->encoder_name, "libawd", 6);

    this->encoder_version = (char *)malloc(255);
    memset(this->encoder_version, 0, 255);  // ADDED
    snprintf(this->encoder_version, 255, "%d.%d.%d%c",
             AWD::VERSION_MAJOR, AWD::VERSION_MINOR, AWD::VERSION_BUILD, AWD::VERSION_RELEASE);
}


// ADDED

AWDMetaData::~AWDMetaData() {

 if(this->generator_name) {
    free(this->generator_name);
    this->generator_name = NULL;
 }
 if(this->generator_version) {
    free(this->generator_version);
    this->generator_version = NULL;
 }

 if(this->encoder_name) {
    free(this->encoder_name);
    this->encoder_name = NULL;
 }
 if(this->encoder_version) {
    free(this->encoder_version);
    this->encoder_version = NULL;
 }
}


void
AWDMetaData::override_encoder_metadata(char *name, char *version)
{

 // ADDED

 if(this->encoder_name) {
    free(this->encoder_name);
    this->encoder_name = NULL;
 }
 if(this->encoder_version) {
    free(this->encoder_version);
    this->encoder_version = NULL;
 }

 this->encoder_name    = name;
 this->encoder_version = version;
}


AWD::~AWD()
{
    delete this->texture_blocks;
    delete this->cubetex_blocks;
    delete this->material_blocks;
    delete this->mesh_data_blocks;
    delete this->skeleton_blocks;
    delete this->skelanim_blocks;
    delete this->skelpose_blocks;
    delete this->uvanim_blocks;
    delete this->scene_blocks;
    delete this->namespace_blocks;

    if(this->metadata) {
       delete this->metadata;       // ADDED, this ok?
    }
}

void
AWD::set_metadata(AWDMetaData *block)
{

    if(this->metadata) {
       delete this->metadata;       // ADDED
       this->metadata = NULL;
    }

    this->metadata = block;
}

[bboltz]

Another issue:
It looks like the updated SDK now writes out normal textures, but geomUtil still doesn't support vertex tangents. So, the normalmap won't render correctly without them.

But, I don't even know what format Away3D is using for vertex tangents?
Is it a 3 or 4-float vector? I would think you would need at least two 3D vectors, tangent and bi-tangent. However, I only see one stream type, 5, for "vertex tangents".

[80prozent]

hey
sorry for taken so long to reply
i was working on a different part of the 3dsmax exporter over the last week, and only came back to the libawd two days ago...

i think you are 100% right about the field_type for the vertex-index-stream. i will see to fix this in the AWD2parser, so the field_type of the stream gets respected. i will take care to update this in the AWDEncoder too.

About the metadata and deconstructor bugs:
i am quite new to c++, and havnt really looked at the deconstruction side of the libawd yet. for me this compiles for 2012. I didnt searched for bugs there yet. as mentioned previeous, thats on my to do.
thanks for the changes you posted. i will review them and include them into my next update.

btw. i have been working a lot on the libawd in the past two days, and will commit a major refactor soon, including multimaterial export, splitting big meshes into subgeos, adding missing animation-blocks...

i will keep working on the libawd, so hopefully it will support all the awd2.1 stuff soon.

for the normal maps and the vertex-tangent data: i am not shure what the vertex-tangent data is really used for. i am only exporting vertex-normal-data and normal-maps.
Away3d automaticly calculates Vertextangent-data (3-float-vector).

[bboltz]

No problem. Thanks for the updates!

[80prozent]

Hey
I did a lot of changes to the AWD-SDK.
For me, VC2010 does not show any memory leaks any more.