Hi @dakejahl,

I allowed myself to add a small update to avoid the console getting spammed with errors no in case of fast uart writes (which can now cause EAGAIN to be returned, as there is no wait to drain anymore). This leads to a lot of spam.

@alexcekay I'm okay with removing the warning but we should also fix the caller that is causing the warning in the first place. If 0 bytes are written it means the caller needs to back off and wait until the TX buffer has partially drained.

Just FYI @dakejahl,

I had a look at all the drivers that actually use the write and may now have problems as it is more likely now that the number of returned bytes does not match the requested number of bytes (will of course only likely occur when there is high load on the UART). Note that it already was possible before this change that the requested amount of bytes does not match the returned amount of bytes (when requesting more than the TX buffer, or interrupted).

@JoelJ18: The following function is a bit optimistic (in MicroStrain.cpp):

bool mipInterfaceUserSendToDevice(mip_interface *device, const uint8_t *data, size_t length)
{
	int res = device_uart.write(const_cast<uint8_t *>(data), length);

	if (res >= 0) {
		return true;
	}

	PX4_ERR("MicroStrain driver failed to write(%d): %s", errno, strerror(errno));
	return false;
}

The interface is happy as long as any data is written, but it may be less than the requested amount of bytes. The rest of the code seems to assume that if this returns true the send worked correctly.

@katzfey: The voxl2_io.cpp init is very optimistic:

int Voxl2IO::get_version_info()
{
	[...]
	while ((got_response == false) && (retries_left > 0)) {
		retries_left--;

		//send the version request command to the board
		if (_uart_port.write(cmd.buf, cmd.len) != cmd.len) { // <- optimistic here
			PX4_ERR("VOXL2_IO: Could not write version request packet to UART port");
			return -1;
		}

In case this one write returns less than the requested amount of bytes (which could happen, although unlikely) the function returns -1 which will cause the whole init() to fail. This causes the driver to not start at all. I think it was intended to actually have multiple tries here (retries_left), but in case the write does not directly write the complete package it instantly returns -1 and does not do any retry.

We probably need to backport this into 1.16