UART timeout and timeout_char is not working in ESP32
Hi,
I noticed that UART timeout_char is not working. Regardless of the value of this parameter, UART always waits for around 8x the time of a character to identify that communication has ended.
With the help of a forum contributor, we were able to locate and fix the problem.
In the machine_uart.c we add the following command to line 208:
uart_set_rx_timeout(self->uart_num, self->timeout_char);
==================================================
Now that code snippet looked like this:
// set timeout_char
// make sure it is at least as long as a whole character (13 bits to be safe)
self->timeout_char = args[ARG_timeout_char].u_int;
uint32_t min_timeout_char = 13000 / baudrate + 1;
if (self->timeout_char < min_timeout_char) {
self->timeout_char = min_timeout_char;
}
uart_set_rx_timeout(self->uart_num, self->timeout_char);
==================================================
I don't know if timeouts have been disabled for a specific reason, but I would like to report that this minor fix resolves the issue, at least for timeout_char.
machine.UART: Fix the double timeout with uart.read(), uart.write() and uart.readinto().
Summary
When uart.read(), uart.write() and uart_readinto() timed out because less data was transferred than expected, but some, the delay was uart.timeout + uart.timeout.char instead of just uart,timeout_char().
This PR changes the behavior into the expected one. It consists of 3 commits. The first changes uart.read() and uart.write(). The impact of that change is limited to the machine.UART class. The second commit changes stream.readinto() and may affect all uses of that function.
The third commit changes the ESP32's machine_uart.c to deal properly with the timeout_char argument for uart.read().
uart.readline() is unchanged.
This PR addresses issue #17611.
Testing
Tested with STM32(Pyboard V1.1 and PYBD_SF6), RP2 Pico, MIMXRT(Teensy 4.1), SAMD51, ESP32, NRF (nrf52840) and a CC3200 board using strings of varying length. Typical results:
RP2
UART(0, baudrate=38400, bits=8, parity=None, stop=1, tx=0, rx=1, txbuf=256, rxbuf=256, timeout=200, timeout_char=50, invert=None, irq=0)
uart.read() msg="" timeout=201 ms
uart.read() msg="123" timeout=52 ms
uart.read() msg="12345" timeout=1 ms
uart.readline() msg="" timeout=201 ms
uart.readline() msg="12345" timeout=202 ms
uart.readline() msg="12345\n" timeout=2 ms
uart.readinto(bytearray(5)) msg="" timeout=201 ms
uart.readinto(bytearray(5)) msg="123" timeout=52 ms
uart.readinto(bytearray(5)) msg="12345" timeout=1 ms
CC3200
UART(1, pins=(Pin("GP16"), Pin("GP17")))
uart.read() msg="" timeout=36 ms
uart.read() msg="123" timeout=38 ms (from 74 ms without the change)
uart.read() msg="12345" timeout=5 ms
uart.readline() msg="" timeout=35 ms
uart.readline() msg="12345" timeout=41 ms
uart.readline() msg="12345\n" timeout=6 ms
uart.readinto(bytearray(5)) msg="" timeout=35 ms
uart.readinto(bytearray(5)) msg="123" timeout=39 ms
uart.readinto(bytearray(5)) msg="12345" timeout=5 ms
Trade-offs and Alternatives
If the impact of the changes to a stream module seems unsafe, the second commit may be omitted, causing the behavior of uart.readinto() unchanged.