@occheung When you're free again, would you please show us how to test that this behaviour has been completely eliminated?
Agh! Finally found someone else complaining about a similar situation as we do: https://github.com/rust-embedded/embedded-hal/issues/264
I'm going to read through it and see what steps we can…
When Dip comes back, I'd like to know the actual reason behind this delay between CS assertion/dessertion and the actual SPI transfer. Must the delay be in nanoseconds?
In 232a08f11012fd749b68882140a8b9dc1a653363, I added a pair of newlines, one before the comment line (that explains polling), and one after self.poll(). Would that be good?
Resolved in 232a08f11012fd749b68882140a8b9dc1a653363 by copying this original line of comment verbatim.
Just added 2eadb652ff1d8e043736006aa74170dbb6d67f15 to deal with full TX buffer. We could check smoltcp::socket::TcpSocket::can_send() before doing send_slice(), but by just checking the number…
@astro Thanks a lot for reviewing this. I didn't update the PR description early enough before you pushed your review, and I just updated it to show some corrections.
The main thing I've changed…
Perhaps @occheung meant that the internal_socket is the socket stored "internally" in the smoltcp::socket::SocketSet within the NetworkStack, which returns the socket needed using the socket…
Right. Before, poll() is simply part of update(), and I decided to separate them simply for the sake of clarity.
What do you think: is it good to keep them separated?