Summary
I have two programs; one written in Python3, another in C++. Both perform the same task; read from a serial port, filter out the two header floats, and print the remaining message. The Python script works properly (see the output below for the proper numbers); the C++ one, using a serial library, does not and I can't figure out why. (Running on Raspberry Pi 4, Raspbian Buster).
In order to save anyone from reading this whole thing; if my decision to use this library for serial reading is bad, how can I properly go about reading from a serial port in C++?
What I've tried
I'm very new to C++, so perhaps I'm looking in all the wrong places, but I couldn't find a universally accepted library for reading from a serial port, so I picked the one with the most stars on github (serial). This answer gives a sample for windows and links to a couple libraries, however they either are intended to work with windows, or are in C, not C++. This uses another library. This is in C (my code will be compiled alongside a Simulink-based C++ class, so I think I need to stick with C++ (?))
My Code
Python
Here is the fully functional Python code:
import serial
import struct
import time
PORT_NUMBER = '/dev/ttyACM0'
BAUDRATE = 115200
MESSAGE_LENGTH = 8
HEADER_NUMBER = float(112)
header_1_received = False # Has the first header byte been received
header_2_received = False # Has the second header byte been received
dat = [] # The actual message
# Open serial port
ser = serial.Serial(
port=PORT_NUMBER,
baudrate=BAUDRATE,
parity=serial.PARITY_NONE,
stopbits=serial.STOPBITS_ONE,
timeout=0.001)
ser.isOpen()
readings = 0
print('Receive data from Pixhawk using 2 header floats and {} message floats (32-bit) \nIf you wish to close the program, hit \"Ctrl+C\" on your keyboard and it (should) shut down gracefully.'.format(MESSAGE_LENGTH))
start_time = time.time() # Start time for the program
try:
# Main loop
while (True):
# Read 4 bytes (32-bits) to get a full float number
buffer = ser.read(4)
# Only proceed if the buffer is not empty (an empty buffer is b'')
if buffer != b'':
# Since struct.unpack() returns a tuple, we only grab the first element
try:
new_dat = struct.unpack("f",buffer)[0]
if header_1_received==True and header_2_received==True:
dat.append(new_dat)
elif new_dat == HEADER_NUMBER:
# We found a header single; treat it
if header_1_received == False:
header_1_received = True
elif header_2_received == False:
header_2_received = True
else:
# Since below, we reset headers once the full message is received, kind of pointless else
pass
else:
# If a non-header character is received, but we haven't identified the headers yet, then we're starting in the middle of a message or have lost the rest of our previous message
dat = []
header_1_received = False
header_2_received = False
except:
# struct.unpack likely failed; throw away the message and start again
header_1_received = False
header_2_received = False
dat = []
if(len(dat) == MESSAGE_LENGTH):
# Reset flags
#print(dat)
header_1_received = False
header_2_received = False
dat = []
readings += 1
except KeyboardInterrupt:
ser.close()
elapsed_time = time.time() - start_time
if readings > 0:
print("Number of readings: {}\nRun time: {}s\nAverage time per reading: {}s ({}ms)".format(readings,elapsed_time,elapsed_time/readings,(elapsed_time/readings)*1000))
C++
Here is the dysfunctional C++ code:
#include <string>
#include <iostream>
#include <cstdio>
#include <unistd.h>
#include "serial/serial.h"
using std::string;
using std::exception;
using std::cout;
using std::cerr;
using std::endl;
using std::vector;
int run(int argc, char **argv)
{
// Argument 1 is the serial port or enumerate flag
string port(argv[1]);
// Argument 2 is the baudrate
unsigned long baud = 0;
sscanf(argv[2], "%lu", &baud);
// port, baudrate, timeout in milliseconds
serial::Serial my_serial(port, baud, serial::Timeout::simpleTimeout(0.001));
cout << "Is the serial port open?";
if(my_serial.isOpen())
cout << " Yes." << endl;
else
cout << " No." << endl;
/* MY CUSTOM VARIABLES */
const float header = 112;
const int msg_size = 8;
int msg_index = 0;
float f; // the read float
float msg [msg_size] = { }; // the collected floats will be placed here, auto-filled with 0s
bool header_1_received = false;
bool header_2_received = false;
uint8_t *buffer = new uint8_t[sizeof(f)]; // buffer that will be converted to 32-bit float
int count = 0;
while (count < 1000) {
size_t number_of_bytes_read = my_serial.read(buffer, sizeof(f));
memcpy(&f, buffer, sizeof(f));
// Logic for adding new element to array
if (header_1_received and header_2_received){
msg[msg_index] = f;
msg_index += 1;
} else if (f == header) {
if (header_1_received == false){
header_1_received = true;
} else if (header_2_received == false){
header_2_received = true;
} else {
// Do nothing
}
} else {
// A non-header character was received, but headers are also not identified;
// Throw partial message away and restart
std::fill_n(msg, msg_size, 0); // Fill with zeroes
msg_index = 0;
header_1_received = false;
header_2_received = false;
}
// Check to see if message is full
if(msg_index == msg_size){
cout << "Msg: [";
for (int i = 0; i < msg_size; i += 1){
cout << msg[i] << ",";
}
cout << "]" << endl;
// Reset flags
header_1_received = false;
header_2_received = false;
std::fill_n(msg, msg_size, 0);
msg_index = 0;
}
count += 1;
}
return 0;
}
int main(int argc, char **argv) {
try {
return run(argc, argv);
} catch (exception &e) {
cerr << "Unhandled Exception: " << e.what() << endl;
}
}
The C++ library can be found here, and the documentation on the read method here. As far as I understand it, his read method writes the requested number of bytes (if available) to the buffer; there is a constant stream of incoming bytes from the other device, so I don't see this being the issue.
Expected Result
The Python3 script functions properly and outputs the following:
[0.08539174497127533, 0.17273111641407013, -9.816835403442383, 0.0, 0.0, 0.0, 0.0, 0.0]
[0.08539174497127533, 0.17273111641407013, -9.816835403442383, 0.0, 0.0, 0.0, 0.0, 0.0]
[0.08539174497127533, 0.17273111641407013, -9.816835403442383, 0.0, 0.0, 0.0, 0.0, 0.0]
[0.08539174497127533, 0.17273111641407013, -9.816835403442383, 0.0, 0.0, 0.0, 0.0, 0.0]
(it should be two very smaller numbers, followed by approximately -9.81, then 5 zeroes).
Actual Result
The C++ program can be built and executed by running the following command:
g++ serial_example.cc -lserial -L ../build/devel/lib -I../include -o test_serial
LD_LIBRARY_PATH=`pwd`/../build/devel/lib ./test_serial
and outputs the following:
[112,112,112,112,112,112,112,112,]
[112,-9.82691,-9.82691,-9.82691,-9.82691,-9.82691,-9.82691,0,]
[112,112,112,112,-9.82691,-9.82691,-9.82691,-9.82691,]
[112,112,112,112,112,112,112,112,]
If I add the following line
cout << "Float extracted from buffer: " << f << endl;
then it outputs every float it reconstructs from the read operation, resulting in a mish-mash of 9.81, 112 and 0s.
Question
What has gone wrong in my C++ program that causes it to read bytes/float differently than the Python program, and if the library is at fault, what is an alternative method or library for reading serial messages in C++?
Edit
After some troubleshooting with @Barmar and @Gaspa79, it seems that the number of bytes read by the library's read() method is inconsistent. I'll try re-writing my program and leave the new version as an answer.
