Correct example for compressed texture allocation?

Hi all! Can anyone point me an example for allocating in-memory texture with compression? Here is my uncompressed example (from World Wind Java)

        TextureData td = new TextureData(
                gl.getGLProfile(),    // GL profile
                GL.GL_RGBA8,          // internal format
                width, height,        // dimension
                0,                    // border
                GL.GL_RGBA,           // pixel format
                GL.GL_UNSIGNED_BYTE,  // pixel type
                this.isUseMipmaps(),  // mipmap
                false, false,         // dataIsCompressed, mustFlipVertically
                null, null);          // buffer, flusher
        Texture t = TextureIO.newTexture(td);
        t.bind(gl);
        gl.glTexParameteri(GL.GL_TEXTURE_2D, GL.GL_TEXTURE_MIN_FILTER,
                this.isUseMipmaps() ? GL.GL_LINEAR_MIPMAP_LINEAR : GL.GL_LINEAR);
        gl.glTexParameteri(GL.GL_TEXTURE_2D, GL.GL_TEXTURE_MAG_FILTER, GL.GL_LINEAR);
        gl.glTexParameteri(GL.GL_TEXTURE_2D, GL.GL_TEXTURE_WRAP_S, GL.GL_CLAMP_TO_EDGE);
        gl.glTexParameteri(GL.GL_TEXTURE_2D, GL.GL_TEXTURE_WRAP_T, GL.GL_CLAMP_TO_EDGE);

I have tried this and it used to work with DXT3/DXT5/BPTC until this year (driver update I suppose, but was I wrong?)

        TextureData td = new TextureData(
                gl.getGLProfile(),    // GL profile
                GL.GL_COMPRESSED_RGBA_S3TC_DXT3_EXT,       // internal format
                width, height,        // dimension
                0,                    // border
                GL.GL_RGBA,           // pixel format
                GL.GL_UNSIGNED_BYTE,  // pixel type
                this.isUseMipmaps(),  // mipmap
                false, false,         // dataIsCompressed, mustFlipVertically
                null, null);          // buffer, flusher
        Texture t = TextureIO.newTexture(td);
        t.bind(gl);
        gl.glTexParameteri(GL.GL_TEXTURE_2D, GL.GL_TEXTURE_MIN_FILTER,
                this.isUseMipmaps() ? GL.GL_LINEAR_MIPMAP_LINEAR : GL.GL_LINEAR);
        gl.glTexParameteri(GL.GL_TEXTURE_2D, GL.GL_TEXTURE_MAG_FILTER, GL.GL_LINEAR);
        gl.glTexParameteri(GL.GL_TEXTURE_2D, GL.GL_TEXTURE_WRAP_S, GL.GL_CLAMP_TO_EDGE);
        gl.glTexParameteri(GL.GL_TEXTURE_2D, GL.GL_TEXTURE_WRAP_T, GL.GL_CLAMP_TO_EDGE);

(I didn't set dataIsCompressed=true or assign the buffer)

mip-map is off. It used to work fine on nVIDIA card (there are artifacts on AMD's).

But now it shows complete garbage like this:

PS: The garbage parts are the compressed texture tiles. But I can still load DXT3/DXT5 from files.

https://github.com/elect86/modern-jogl-examples/blob/master/modern-jogl-examples/src/tut15/manyImages/ManyImages.java#L160-188

I use the jogl DDS

I added error checks everywhere and located the problem: glFramebufferTexture2D

The texture is to be used with framebuffer (to store whatever painted onto it), and glFramebufferTexture2D throws "invalid framebuffer operation" when the given texture's internal format is DXT3/DXT5, but no error when it's just RGBA8.

But why? It doesn't seem to be documented anywhere.

Here is the simplified code - it turns out WWJ is using TextureData for empty texture allocation and I can just skip it to save the confusion.

        int internalFormat = GL.GL_COMPRESSED_RGBA_S3TC_DXT3_EXT;
        int pixelFormat = GL.GL_RGBA;
        GL gl = dc.getGL();
        Texture t = new Texture(0, GL.GL_TEXTURE_2D, width, height, width, height, false);
        t.bind(gl);
        gl.glTexImage2D(t.getTarget(), 0, internalFormat,
                width, height, 0,
                pixelFormat, GL.GL_UNSIGNED_BYTE, null);
        gl.glTexParameteri(t.getTarget(), GL.GL_TEXTURE_MIN_FILTER, GL.GL_LINEAR);
        gl.glTexParameteri(t.getTarget(), GL.GL_TEXTURE_MAG_FILTER, GL.GL_LINEAR);
        gl.glTexParameteri(t.getTarget(), GL.GL_TEXTURE_WRAP_S, GL.GL_CLAMP_TO_EDGE);
        gl.glTexParameteri(t.getTarget(), GL.GL_TEXTURE_WRAP_T, GL.GL_CLAMP_TO_EDGE);

all right it turns out the compressed texture is not supposed to work with FBO writing to it. So it's probably bad luck last year to have it work.

UPDATE:

Found the correct way of doing it: Convert texture to GL_COMPRESSED_RGBA (works for DXTC too) Basically I just have to use a renderbuffer instead of texture as color attachment, and then copy it to texture afterwards by glCopyTexImage2D. It's slower but works.

Thank you for sharing your findings, it wasn't obvious for me.