Format std::time_point in ISO 8601 format with fractional seconds / microseconds in C++

A C++ function to format an std::time_point as an ISO 8601 string.

The C++ std chrono stuff is very useful but a bit of a head-wreck!  One of the things I had problems with was how to take an std::time_point value and format it as a string with the fractional seconds / microseconds included.  This sort of time resolution is often required for accurately time stamping machine vision images, especially when acquiring at  high rates from multiple cameras – accurate timestamps allow you to compare images from different cameras that were taken at the ‘same time’.

Anyway if you’re happy to wait for C++20 then you will have access to a format() function; but if you’re more eager to format your time strings now, then here is a function which may fit the bill, it (the function) has to jump through some hoops, but gets there in the end.

Note: This function uses the std::chrono::system_clock but it could be converted (or templated) for other std clocks..

//  Licensed under the Apache License, Version 2.0 (the "License");
//   you may not use this file except in compliance with the License.
//   You may obtain a copy of the License at
//   Unless required by applicable law or agreed to in writing, software
//   distributed under the License is distributed on an "AS IS" BASIS,
//   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
//   See the License for the specific language governing permissions and
//   limitations under the License.
// Format an std::time_point as an ISO 8601 string with fractional seconds to 6
// decimal places, e.g. 2014-08-30T08:18:51.867479
// Warning will not work for any date/times before the start of the UNIX epoch.
inline std::string to_iso_8601(std::chrono::time_point t) {
	// convert to time_t which will represent the number of
	// seconds since the UNIX epoch, UTC 00:00:00 Thursday, 1st. January 1970
	auto epoch_seconds = std::chrono::system_clock::to_time_t(t);
	// Format this as date time to seconds resolution
	// e.g. 2016-08-30T08:18:51
	std::stringstream stream;
	stream << std::put_time(gmtime(&epoch_seconds), "%FT%T");
	// If we now convert back to a time_point we will get the time truncated
	// to whole seconds 
	auto truncated = std::chrono::system_clock::from_time_t(epoch_seconds);
	// Now we subtract this seconds count from the original time to
	// get the number of extra microseconds..
	auto delta_us = std::chrono::duration_cast(t - truncated).count();
	// And append this to the output stream as fractional seconds
	// e.g. 2016-08-30T08:18:51.867479
	stream << "." << std::fixed << std::setw(6) << std::setfill('0') << delta_us;
	return stream.str();

Microsoft Teams stuck in a loop saying it needs sign in after a password change

An amusing MS Teams related problem hit me this morning after I changed my Domain Password – The MS Teams App piped up and said it need Sign In, when I hit OK it seemd to restart and then repeated the message, we were in an infinite loop of fun!

Now as much as I can happily live without Teams bugging me, I unfortunately needed it for a conference call so I had to coax it back to life.  What worked for me in the end was to wait until Teams came up with its message and then right-clicked on the teams icon in the task bar and chose the ‘Sign Out’ option.  I then shutdown and restarted Teams and everything was OK again.


Launching Flask on Port 80 without using sudo

Flask is a great python based HTTP server that’s really small and easy/fast to setup, it is really useful for deploying small Web based User Interfaces for IoT type devices.  By default flask will attach itself to port 5000.  To get my flask script to attach to port 80 I use:

if __name__ == '__main__':'', port=80, debug=True)

This works fine, the only problem is that in order to successfully attach to port 80 the script must be run as root, so instead of just running:


I have to run:

sudo ./

Which isn’t great.

To get around this problem I used the tool: authbind

To Install:

sudo apt-get install authbind

And configure it for access to port 80:

sudo touch /etc/authbind/byport/80
sudo chmod 777 /etc/authbind/byport/807

(Not sure if the very loose 777 permission is required, must experiment)

Now I just have to launch my flask UI script using authbind and it will take care of the script’s permissions to bind to port 80 and I can connect from a browser:

authbind -deep python3 ./

More info on authbind here

Octave – Can’t scroll Window, Workaround

Strange problem with Octave (my version is 4.4.0), in the GUI I can’t scroll the ‘Command Window’, so if some code outputs lots of info I can’t scroll up to see it all, the window keeps jumping to the bottom as I try to scroll!

I am not sure why its happening, but a workaround is to enter the ‘pause’ command within the window, once pause executes I can scroll to my heart’s content.  I  then hit ctrl-c to exit pause.

It is possible that installing a newer version of Octave would fix this but I am not bothered to upgrade at the moment as all of the Maths bits seem to work very well!


Paho Javascript Client – Figure out received message’s MQTT topic

When using the Paho Javascript client from MQTT; when a message arrives via client.onMessageArrived(), how can we figure out the message’s topic.

This threw me for a bit as the documentation for the message object doesn’t mention ‘topic’ at all – but it turns out that the topic name is stored in the message.destinationName field! (panic over)

KDevelop hangs during C++ build on ARM Odroid N2

I have been using Hardkernel’s new Odroid N2 to develop a computer vision system with multiple Basler USB3 cameras. I have been using KDevelop on the Odroid to engineer the C++ code and in general all has been going very well (the Odroid N2 is a fantastic device). I did however hit an occasional problem where KDevelop would cause the Odroid to ‘hang’ during a build – especially when re-building a lot of files.

On investigation it seemed like KDevelop was just using up too much memory and putting the Odroid into a very bad place, sometimes the system would free up after a (long) period of time but most often it wouldn’t, it just ground to a halt thrashing memory (I presume). I am using the boost libraries and I think that a lot of IDEs and build systems have problems with boost as it’s very big! (Clion does especially!)

Anyway the solution was to limit how many parallel build instances (or jobs as it calls them) that KDevelop can run, to do this open the ‘Project / Open Configuration..’ menu and click on the ‘Make’ tab in the left-hand column, now check the ‘Override number of simultaneous jobs’ checkbox, and enter a number in the ‘Number of Simultaneous jobs’ box. A value of 3 works well for me, the builds are a little slower but no longer hang the system! I might try increasing it to see if I can get away with quicker builds…

Boost ASIO Simple UDP Send Packet Example

Update: I have written a simple Fire-And-Forget wrapper class for sending datagrams via UDP can be found here. It handles simple transmission use cases while hiding the (sometimes confusing) boost::asio details. However, if you are interested in the details then read on!

Boost.ASIO is great but if you don’t use it everyday it can be hard to remember how to use it to do even the simplest of things. I have included below a sample of simply sending a packet via UDP (ipv4), see the function called send_message(), this example code aims to be as minimal as it can be:

Those spouting software engineering dogma will often tell you to steer well clear of UDP for the usual, well understood reasons, but for a certain type of application where very low latency is important, it just can’t be beat!!

using namespace boost::asio;
// Send a string via UDP to the specified destination
// ip addresss at the specified port (point-to-point
// not broadcast)
bool send_udp_message(const std::string& message, const std::string& destination_ip,
						const unsigned short port) {
	io_service io_service;
	ip::udp::socket socket(io_service);
	// Create the remote endpoint using the destination ip address and
	// the target port number.  This is not a broadcast
	auto remote = ip::udp::endpoint(ip::address::from_string(destination_ip), port);
	try {
		// Open the socket, socket's destructor will
		// automatically close it.;
		// And send the string... (synchronous / blocking)
		socket.send_to(buffer(message), remote);
	} catch (const boost::system::system_error& ex) {
		// Exception thrown!
		// Examine ex.code() and ex.what() to see what went wrong!
		return false;
	return true;

This is the bare-bones code, no error reporting etc. Also it won’t broadcast, to allow for broadcast you need to include the following two lines and supply a broadcast ip address when calling the function. Be careful if broadcasting a lot of data as it can really overload & mess-up network equipment!

socket_base::broadcast option(true);

Fixed – KDevelop not stopping at breakpoints on Ubuntu Mate

I couldn’t get KDevelop to stop at breakpoints even on simple ‘hello world’ C++ projects. It appeared that CMAKE_BUILD_TYPE was being correctly set and GDB worked fine from the command-line, but from within kdevelop breakpoints were never respected! I think the problem stemmed from the Cache Value for CMAKE_BUILD_TYPE being empty, this value can be seen in the Project / Open Configuration… menu:

kdevelop breakpoint broken

Following the advice from this post I added the following into the project’s CMakeLists.txt file:

# Set a default build type if none was specified
  message(STATUS "Setting build type to 'Debug' as none was specified.")
  set(CMAKE_BUILD_TYPE Debug CACHE STRING "Choose the type of build." FORCE)
  # Set the possible values of build type for cmake-gui
  set_property(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS "Debug" "Release"
    "MinSizeRel" "RelWithDebInfo")

[Originally from here]

I ran clean & rebuild etc. and then cache value was set correctly to ‘Debug’ and the debugger happily stops at breakpoints!

I hope this post helps folks as I spent _ages_ trying to get the breakpoints to work!!

Bash – send data to serial (rs232) port and wait for response

Sending data to a serial port is quite easy in Bash, for example:

echo "my packet data" > /dev/ttyS0

And you can read from a serial port using cat:


cat /dev/ttyS0

However cat must typically be run from a different shell instance as it blocks waiting for data. So is it possible to write and then read the response from a single shell instance?

Well, it is, but it requires a bit of sleight-of-hand. For example, if we start cat in the background and then send the command, cat will report the response as follows:

# Run cat in the background
cat /dev/ttyS0&
# Send the command, cat should print the response
echo "my packet data" > /dev/ttyS0

Which works but it a bit of a mouth-full! cat continues to run in the background, and will print more responses as they arrive.

But what if you want to just send one packet and then wait for a single response?

This is a bit harder, but if your response ends with an end of line character, or another known character then we can use read to help with this…

First we setup a read command in the background, unlike cat, this command will end when a response is received or when the timeout time arrives, then we can send our command:

(read -n60 -t20 RESP < /dev/ttyS0; echo $RESP)&
echo "my packet data" > /dev/ttyS0

This gets read to wait for up to 20 seconds (-t20) for a line of data (max size, 60 characters -n60) from /dev/ttyS0, which it reads into RESP, it then echos $RESP – all of this happens in the background. echo then sends the packet which will result in a response.

If your response packed ends with a character other than an EOL character then you can specify a delimiter to read using the -d command-line option.

Again it’s all a bit long winded, so we can wrap it all up in a bash script ( as follows:

# Send a packet to the specified serial port
# and wait for, and output the response, it is assumed
# that the response will end with an EOL character.
# usage: <data-to-send> <port>
# [backgound] Wait for, and read a line from the serial port into RESP,
# max 128 characters, timeout=10s, then output $RESP
# (c) 2019 Kevin Godden
(read -n128 -t10 RESP < $2; echo $RESP)&
# Hack - use read to pause for 200ms to give previous
# command a chance to get started..
read -p "" -t 0.2
# Send command
printf "$1\r" > $2
# Wait for background read to complete

An example of using the script:

./ "my-command" /dev/ttyPS0./ "my-command" /dev/ttyPS0

Github repo is here

PCL C2988 unrecognizable template declaration/definition Visual Studio 2017

If you get this compile error:

Error C2988 unrecognizable template declaration/definition

When you:


from the Point Cloud Library (PCL) in Visual Studio 2017, then either throw the following in before the #include, like this:


or upgrade your version of Visual Studio 2017 (I haven’t tested this yet myself!)

Not sure why, it’s something to do with workarounds for VS 2017 bugs or something…