We do so by querying the EsounD latency and estimating dwPlayedTotal based on the elapsed time since the last write.
This benefits the accuracy of the completion notifications and of wodGetPosition().

We don't know how much free space we have in the EsounD buffer, so
just write until it is full. Sleep for a fraction of the time it will
take for the buffer to drain so we can refill it in time.

This simplifies our code and avoids data copying.
Also, now wodGetVolume() returns the real EsounD volume, even if it has been modified by an external mixer (e.g. esdctl).

Put the windows application name in it to make it user friendly as it is displayed by EsounD mixers.

It is used to send/receive stream data, not to send RPCs to the sound server as esd_fd would imply.
This is a no-op.

Windows advertises it on all these formats(Nvidia Geforce 7400), and
Trackmania Nations needs it on D3DFMT_Q8W8V8U8 to accept ps_3_0 support.

shader_select takes care of compiling the shader since we have the shader
duplication infrastructure. However, there's still a motivation to skip the
shader_select call if the vertex shader is dirty.

init_materials() doesn't exist, and I don't know what the TODO means with
primitive type (the diffuse stuff is already handled properly).

rpcrt4: Free the memory allocated for the protseq and endpoint strings allocated in RpcServerUseProtseq* functions.

Make a copy of the protseq string in alloc_serverprotoseq to simplify
memory management in higher-level functions.

However, on some buggy platforms it returns
RPC_S_INVALID_ENDPOINT_FORMAT. Don't follow this behaviour our
implementation of the API.

We have the version already in the header of the winetest results.