Search This Blog

Wednesday, October 23, 2019

Designing With State Machines

I can still remember the first time I was introduced to state machines. It was around 1988 when I took a course on compiler design. We used "The Dragon Book,"otherwise known as "Compilers: Principles, Techniques, and Tools."

State Machines

One of the assignments was to build a lexical analyzer (a piece of code that reads texts and turns it into meaningful tokens) to scan through a language called "Babbage."  To do this, we were tasked to build a state machine. If you aren't familiar with state machines, the best way to describe them is that you have a graph of different "states" with "transitions" that move between them.  The following shows a simple state machine that recognizes the words "cat," "cap," and "cub."


State machines are applicable across many different fields.  We already have seen them in relation to tokenizing input, but they can also be used in machine control.  Consider the following state machine that could be part of a simple thermostat.
In this state machine, there are three states: Heating, Cooling, and Idle.  If the system is in the idle state, and "TooCold" is declared, the state machine transitions to the Heating state. Likewise, if the system is in the Heating state and the target temperature is reached ("TargetReached"), the system transitions back to the Idle state.

Compound States

If you look at the thermostat state machine, you will see that it only handles heating and cooling. There is nothing in the state machine that controls the fan on the system. In the following diagram, the FanOn and FanOff states have been added along with a Wait state to introduce hysteresis to the system.

Reading the diagram, we see that when we are in the Idle state we are also in the FanOff state. When we are in the Heating, Wait, or Cooling states, we are also in the FanOn state. We don't need extra transitions to go between the FanOff and FanOn states. The transition to FanOn is automatically made when entering Heating or Cooling, and the transition to FanOff is made automatically when entering Idle.

Knowing what a state machine is is one thing. Knowing how to use them in code is another.

Working With State Machines in Code

There are several ways that we can implement state machines in code. I'll start with a brute force method, then move on successively simpler methods.

Logic Driven State Machines

The first method is through brute force coding of the state machine in the code itself. Switch statements and conditionals are used to control the state transitions. This is one of the first ways I started using state machines, and the method can still be found some "big loop" style control code. Let's look at what this would look like in terms of our first (simplest) thermostat:

    /// Brute Force
    switch (currentState) {
    case Idle:
        if (TooCold) {
            nextState = Heating;
        }
        else if (TooHot) {
            nextState = Cooling;
        }
        break;
        
    case Heating:
    case Cooling:
        if (TargetReached) {
            nextState = Idle;
        }
        break;
    }
    currentState = nextState;
The code doesn't look too bad for only three states, but consider what it would be like for our simple lexical analyzer!  There must be a better way!

Table Driven State Machines

I need to be honest here, for the lexical analyzer assignment I drew a state machine. Instead, I used a piece of software called "lex." "lex" builds a lexical analyzer state machine for you. It takes a set of words (tokens) to recognize and generates C code with state tables. Some might say I cheated, but the professor did not dock my grade.

Most people who code state machines will resort to a table approach. It works well for simple state machines with few states.

    /// Table
    const states stateTable[NUM_STATES][NUM_CONDITIONS] {
        /*              TooHot   TooCold  TargetReached */
        /* Idle    */  {Heating, Cooling, Idle},
        /* Heating */  {Idle,    Heating, Idle},
        /* Cooling */  {Cooling, Idle,    Idle}
    };

    newState = stateTable[currentState][condition];

The table is relatively concise and fast. There are no conditional tests, just simple table lookups.

Problems come when we get more states, need to deal with combined states (e.g. FanOn, FanOff), or try to change the table. Introduce a new state and/or condition, and everything needs to change.

What if we could have the speed of a table driven state machine and we could easily modify the state machine without having manually regenerate or modify the tables?

Graphically Driven State Machines

In case you missed it: I love programming for the Qt Framework. There are many reasons, but here is one of my favorites: 

Starting with Qt 5.7, you can draw state machines in Qt Creator and then easily export and use them in your code.

Modifying your state machine only costs you the rebuilding of your code, not hours of work re-coding brute force machines or modifying state tables.  Even combined states are handled gracefully.

Once you have graphically drawn your state machine, you just need to use it. You start by instantiating the state machine, then you simply connect the signals and slots as appropriate. The following code from my book Hands-On Embedded Programming with Qt shows how the thermostat state machine can be used.

    // Connect state machine states to UI indicators
    m_hvacSM.connectToState("Heating", ui->heatingInd, &QLabel::setEnabled);
    m_hvacSM.connectToState("Cooling", ui->coolingInd, &QLabel::setEnabled);
    m_hvacSM.connectToState("FanOn",   ui->fanInd,     &QLabel::setEnabled);

    // Connect state machine states to the HVAC controller
    m_hvacSM.connectToState("Heating", &m_hvacCtrl, &HVACController::setHeatingOn);
    m_hvacSM.connectToState("Cooling", &m_hvacCtrl, &HVACController::setCoolingOn);
    m_hvacSM.connectToState("FanOn",   &m_hvacCtrl, &HVACController::setFanOn);
    m_hvacSM.start();   // start the state machine

. . .
    // update the state machine based on current temp and the settings
    QString transition;
    if (temperature > ui->maxTemperatureSpinBox->value()) {
        transition = "TooHot";
    } else if (temperature < ui->minTemperatureSpinBox->value()) {
        transition = "TooCold";
    } else {
        transition = "TargetReached";
    }
    m_hvacSM.submitEvent(transition);
While it seems like the code is more complex, it is actually doing a lot more than the other examples. In the other examples, we only advanced the state machine. In this code example, we are also turning on GUI indicators, and we are causing the system to control the fan and heating and cooling elements. Best of all, changing the behavior of the system (like inserting a wait state) doesn't require re-writing the code! Only the state machine changes!

Behind the Scenes

Truth be told, both the second and third implementations are using tables. The real difference is in the second method we have to create and maintain the tables ourselves. In the third method, tooling takes our drawings, creates the tables, and provides a easy to use Qt/C++ Interface.

Conclusion

Hopefully, now that you have learned about state machines you will start using them in your own designs and maybe even use Qt for its built in state machine support.

If you want to learn more about graphically working with state machines in Qt, get a copy of my book Hands-On Embedded Programming with Qt and check out "Designing with State Machines in Chapter 7.  You can find it either on the Packt Website or on Amazon.

Tuesday, October 22, 2019

Meet Qt Pittsburgh 2019 Presentation

I recently presented a Lightning Talk at the Meet Qt event held in Pittsburgh, PA on October 17, 2018.  the topic of my talk was Embedded Development without Hardware. While I plan do a full blog entry about the topic in the future, I thought I would share the presentation with you now.

The presentation file can be be viewed here: http://bit.ly/NoHardwareDev-2019

Enjoy,
-John

Monday, October 7, 2019

Self Promotion, Paying the Bills, or Just some Good Resources

I've spent the last 28 years developing embedded software, the last 10 years working with Qt, and the last year writing a book on how they can be used together.  The book has been published by Packt.

If you haven't checked out Packt's Hands-on books and are interested in learning new software development ideas, you really should see what they have. I have bought several books myself so that I learn things like GPU programming.

For a limited time, Packt is offering some great discounts on their books in print and e-formats, including mine. Check it out: https://bit.ly/2Moffku.

embedded programming with QT

-John

Friday, September 20, 2019

Upcoming Event: Meet Qt Pittsburgh, PA Greater Area

Someone once said, "When it rains, it pours."  Usually, that is a bad thing, but in my case it appears to be a good thing.  I already hinted that I may soon be working on another book about Qt.  That project seems to be going well.

Just after finding out about that, I stumbled into an opportunity to do a Lightning Talk at a Qt event.

I will be giving a Lightning Talk at Meet Qt in Pittsburgh, PA on October 17, 2019.

The "Meet Qt Pittsburgh Greater Area" is a free, half day conference about Qt.

As soon as I heard about the event, I immediately thought of a couple of topics.

The first one was working with State Charts (Machines) in Qt Creator. With a couple of clicks you can turn them into C++ code that you can add to your project.  We have done it several times on a couple of projects here at Caliber Imaging & Diagnostics, and it is really nice to be able to simply update a drawing and fix the behaviour of a system.

While I really like to talk about Qt's SCXML support, I didn't think I could justice to it in 10-15 minutes.

The other topic came out of a question asked during a presentation of Qt's new level of MCU support. Imagine running Qt directly on an MCU without needing an OS!

The question was simple, "How can I develop Qt code if I don't have the hardware?" Great question, and the topic I will be dealing with at Meet Qt, Pittsburgh.

If you are interested in more information about this free event, check out the event page for it at https://www.qt.io/events/meet-qt-pittsburg-1565888110.

I hope to see you there!
-John

Friday, August 30, 2019

And Now For Something Completely Different...

In restarting this blog, my intent is to publish useful, software engineering related material every week or so.  To that point, I have almost 2 months of ideas planned out.  They including everything from using drawings to build state machines to some of the more useful design patterns I have found.

But something came up. Something I think you will enjoy when it is completed, but something that needs some immediate attention.

I have been contacted to write another book for Packt, and I am working to develop the proposal.

What's all this about writing another book?

One of the more frequent questions I when I tell people that I am a published author is "How did that happen?"

There are a couple of ways to get a book published. One method is to write the book and self-publish.

The next is to write a book and then shop it around to different publisher. This is the way many authors start out. The trick is to find a publishing house with an editor that will say yes. It is not easy.

In many ways it is like trying to find a job.  You network, send out "resumes" (or samples of your writing), make calls, etc. etc.

When I was looking for my last two jobs, I did all of that stuff, and it got me nowhere.  In the case of my last two jobs, the companies reached out to me (through a recruiter).  In the case of my first book, the same thing happened.

For "Hands-on Embedded Programming with Qt," an Acquisitions Editor reached out to me to see if I would be interested in writing a book on the book.  I actually declined at first, but he talked me into it, and I am glad he did.  (I like to say I was commissioned to write the book.) It was hard work, and there were some sleep deprived days, but in the end I really enjoyed it.

Recently, the Acquisitions Editor reached out to me with two book ideas.  I read both them and discovered that one proposal is almost exactly the next book I was thinking I should write.

So there you have it.  I've been commissioned to write another Packt book titled "REDACTED" that will be tentatively published in REDACTED of 2020.  I'm really looking forward to it, and I am hoping that now that I know the process, it will go quite easily.

And In the Meantime...

In the meantime, I still plan on covering the topics I had planned for the blog, but will also probably slip hints in about my progress on the new project.

Selling the Merch...

Please Check-out "Hands-on Embedded Programming with Qt".  It is available on both Amazon and Packt websites.

Monday, August 19, 2019

Time to Rethink Security Questions?

You have probably done it many times. You set up a new account, be it online or in a bank, and they ask you to give them an answer to several Security Questions so they can verify that it is really you.
  • What is your mother's maiden name?
  • What city were you born in?
  • What is the name of the high school you graduated from?
  • What is the name of your oldest sibling?
Have you ever thought about those questions?

In the Internet Connected Age, those are probably the most insecure questions they could be asking. Thanks to social (i.e. Facebook, LinkedIn) and Genealogy sites, most of the answers are easily accessible with little to no work.

Solving the Problem

While I would like to say that by writing this blog I am going to magically get the banking world to change its questions, I really doubt it.  So what can you do?

Personally, I use answers that I know that can't be easily found, but I can remember. For example, I could use "Alfred E. Neuman" as the basis
  • What is your mother's maiden name? Neuman
  • What city were you born in? Flint, Michigan
  • What is the name of the high school you graduated from? MAD
  • What is the name of your oldest sibling? Alfred
For those of you who don't know who I am talking about, take a look at this Wikipedia article: https://en.wikipedia.org/wiki/Alfred_E._Neuman.

Thursday, August 15, 2019

Why "Newest" Is Not Always "Best"

When I wrote Hands on Embedded Programming with Qt (https://www.amazon.com/dp/1789952069), I had a choice to make. Which version of Qt should I use? I chose to go with Qt 5.12. In the 8 months it took to write the book, Qt release several minor updates (5.12.1, 5.12.2, etc.) and then finally 5.13. So why didn't I update the book to use Qt 5.13? It wouldn't have taken much work. Qt 5.12 is a LTS (Long Term Support) version of Qt. That means that it will be getting updates long after Qt 5.13 has reached its end of life. If you are building an embedded system, especially a medical or other regulated system, you need to plan on supporting the device for years. Using a LTS solution is a great way to do that. It avoids major upgrades (and retesting) of the base framework on devices in the field. hashtagqt5 hashtagqt hashtagprogramming hashtagembeddedsoftware hashtagmedicaldevices hashtagmedicalsoftware

Tuesday, August 13, 2019

The Quest for the Perfect Tool

I have recently been involved in an effort to pick "the perfect Code Review Tool." That's really a funny idea when you think about it. "Perfect" could mean a lot of things to a lot of different people, even if they all agree what is.

Consider the case of find "The Perfect Car." The perfect car for what? The perfect car for carrying your teenage daughter and her friends around? The perfect car for to commute 50 miles a day to work and not spend half a weeks pay on fuel? Maybe it's the perfect car for running brisk TSD rallies on the snow covered back roads of central NY?

The same problem happens when looking for the perfect software tool. Even limiting it down to a specific tool, a Code Review Tool, isn't enough. There are still too many questions that come up: Does the whole team do reviews or only a couple of dedicated reviewers? Pre- or post-commit reviews? How does code review fit into the rest of your process? How big is the code set? Is only one project going to use this, or is it becoming the standard for many projects? What do the other projects want? What other tools are in use that it has to integrate with? How much money is there to spend? What platform does it have to run on? The list just keeps going on.

Sometimes the quest for "The Perfect " becomes less about finding the and more about finding out what you really want. As Joe Jackson once sung "You can't get what you want, until you know what you want."

So how do you find out what you want? That's a good question to blog about in the future.