What could cause ID3DXFileSaveObject::Save() to crash? I have an exporter that works fine; it exports a variety of geometry. However, if I register the XEXTENSIONS_TEMPLATES templates, and add data of type DeclData to the file, I get no error when adding the data, but the call to Save() crashes. The DeclData contains a single declaration, for a float2 coordinate stream at texture coordinate index 1. Just removing the call to add the data removes the crash. I'm using April 2006 SDK (d3dx_30.dll). The code I'm using to add the DeclData is: vector2<char> mapcoords; DWORD d = 1; mapcoords.append((char const *)&d, sizeof(d)); DX_VertexDecl vd; vd.type = D3DDECLTYPE_FLOAT2; vd.method = D3DDECLMETHOD_DEFAULT; vd.usage = D3DDECLUSAGE_TEXCOORD; vd.index = 1; mapcoords.append((char const *)&vd, sizeof(vd)); d = DWORD(texcoords.size()*sizeof(DX_Coord2)/sizeof(DWORD))-1; mapcoords.append((char const *)&d, sizeof(d)); mapcoords.append(&texcoords[sizeof(DWORD)], texcoords.size()-sizeof(DWORD)); CHECK( stored->AddDataObject(DXFILEOBJ_DeclData, 0, 0, mapcoords.size(), &mapcoords[0], &temp) ); "vector2" is just a wrapper around std::vector<>, and "texcoords" contains regular texture coordinates in .X file format (DWORD count + Coord2 data array), which saves and loads fine by itself. Again, note: this code works fine. It's when calling save->Save() that D3DX crashes.

I have to admit to not doing any work with the interface you are, but in my experience its been fairly easy to confuse/crash D3DX by passing it invalid (pointer/raw) data. I've done it quite a lot with D3D10 lately [lol]

Probably a bit obvious, but are you definitely sure there are no uninitialized variables, incorrect offsets/dimensions etc...?

I suppose the other possibility is to go digging around in the source code for things like MeshViewer/dxviewer/dxops and the various exporter plugins in the SDK. Maybe, if you're lucky, you'll find a similar example you can use as a reference?

Jack

I'm using the SDK headers, which magically link against the right version -- yes, I'm seeing this in the debug version.

Am I _absolutely_ sure my data is right? Of course not. But I'm as sure as I can be. I guess my next step is to create a small, 4-vertex isolated example and see if I can get that to crash, too.

Actually, the behavior in debug mode is slightly different -- the call to Save() works the first time around (and seems to generate acceptable data), and crashes the second time.

Turns out I was calculating the "number of DWORDS" part of the DeclData template wrong, making the save object run off the end of the data I provided it for the template. Weird that they don't do any error checking internally...

Now, next question: why doesn't this mesh load into the DirectX Viewer when DeclData is present? (I'm building that tool from source to find out)