About Movie Ballot

   As you may have guessed, this project is another one that’s made pretty much exclusively for me. I am not lacking in movies that I want to watch, so some sort of movie recommendation engine wouldn’t be that helpful to me (and also really difficult to implement well). Where I struggle is deciding what exactly I am going to watch from my list, especially when my family members start getting involved. I wanted to create a tool that would help us to quickly narrow down everyone’s selections into a single choice in a way that would feel fair and hopefully leave everyone happy. Over a single weekend I worked pretty rapidly, completing about 50% of the project before getting bored of it and pushing it onto the back-burner. A few months later I decided that enough was enough and I would finally get it done. I perhaps overestimated how close to completion I was at the time (I had thought like 70%) but I was able to get it done over about another week of sporadic work.

   Now I should warn you, this project is an amalgamation of many terrible programming practices. I wouldn’t recommend anyone else do things this way, but I thought it would be interesting and I wanted to see how far I could push it. You see, this was going to be a multiplayer tool so everyone could submit their own suggestions for what to watch. One of my main inspirations was Kahoot! which you can see a little bit in the visual design. What I thought would be interesting is if I did this multiplayer functionality without ANY backend code. I’ve written plenty of backend servers to handle requests before, but I had been messing around with Google Firebase a lot lately and it had given me some ideas. Firebase has a lot of your usual features for these types of services, you can host your code and you can run cloud functions and stuff like that. I could do everything with cloud functions and technically call it “serverless” but that just wasn’t edgy enough for me.

   My idea was to use a Firebase feature called “realtime database” which is just a non-relational database kind of like NOSQL that you can attach listeners to. You can also update it directly through client-side javascript which is pretty cool. Some of my simplest backends just serve to connect HTTP requests from a browser to a SQL query, but doing things this way would cut that out entirely. My plan was essentially to use the database as an intermediary for all the devices to send their data to one another, and one user would be designated as the host who would actually process all of this data and compute results. I would later realize that I was essentially building a peer-to-peer system with an extra step in the form of the database. Rather than sending data directly to one another which would be quite difficult to do using only client-side Javascript they would send the data to the central storage location and everyone would download it. But it’s not a server I promise! It’s just a fairly large JSON object that everyone is updating.

   Now what’s actually wrong with this? Well… a lot. For one it’s incredibly insecure (just like you). I didn’t want to force people to make accounts so anyone is able to anonymously update the database. With no backend I can’t really validate that they’re sending the right type of data or that they’re not acting outside of their scope. Someone could theoretically modify my code and read all the data from every session that’s currently active. This would never fly on a professional level, but I knew that there would be no personal information stored in this database, it’s pretty much all random strings of characters that act as IDs. There’s never a text box where you type something and it gets posted on the server. I also just thought that this would be a really interesting experiment. I’ve never really done anything multiplayer before so I was really curious to see if I could do it. One of these days I’ll create a good multiplayer system with an actual server but at least this is a start.

   So how does this system actually work? Great question. When someone presses the button to create a new room, they create a new subdirectory in the database with the room’s 4 digit random ID. They become known as the host which is stored both client-side and in the database. Anyone who joins is assigned a 10 digit ID that is unique to them. The database stores everyone who has joined a given room in a subdirectory called users and it stores a state variable that tracks the current screen the room should be on.

   I mentioned listeners before which are really cool. Firebase can trigger an event on the client-side whenever something changes in a specified directory of the database. It would be amazing if I could do everything in this project using just listeners, but there was some functionality that meant I had to query the server manually. An important feature is tracking timestamps to see who is online. A lot of the other features for choosing movies ask you to wait for your friends to finish before you can move on. If someone closes their tab or leaves then I didn’t want you to be unable to move on. My solution was a function that would be called periodically to update a lot of the data. Every player would post a timestamp next to their name and they would check everyone else’s timestamp. The number of people that I list as online is simply the number of people in the list of players who have updated their timestamp in the last 90 seconds. The host also updates one big timestamp for the room itself. When a host goes to create a room they will delete any other rooms that haven’t been updated in 3 hours. This function also checks a few other things like how many people are done and what the current state of the room is. The interval that this function happens at affects the perceived responsiveness of the whole system, specifically in things like how fast it switches between different screens. If I update too frequently I might bump myself up out of the free tier on Firebase so I settled on updating 4 times a second as a nice middle-ground.

   The last thing that the database stores is a list of movies, but I will get into that a bit more later. Here’s an example of the data structure of a given room.

the total data structure of the database

   Again, I want to clarify that this is not a good way to do things, but I was curious if it would work and I have to say I’m pretty impressed. One bad practice that I can’t justify is the atrocious CSS. I’ve been trying to get better at this, I’m even taking a class about it for the first time now, but for whatever reason when I started this project months ago I decided to hardcode pretty much everything as pixel values. I don’t usually do that but I did then for some reason and I had to stick with a lot of it because the only thing worse than writing CSS is writing CSS twice. In general I’d say the design is pretty ugly. I should have used more than two colors and one font and I also should have made a clearer decision about what type of drop shadows I wanted to use. Right now it does work on most mobile devices and it is usable on desktop but it’s not that responsive and it’s definitely not pretty.

   Okay, so that’s how the database is structured, let’s talk about what it’s actually like to use. After the host has created a room and everyone has joined it, it’s time to add your movies. Everyone is brought to a page where they can search for every movie they want to watch. For this I used a service called The Movie Database which is like an open source version of IMDB. I can just pass their API the search terms from the user and receive a bunch more JSON objects in return. TMDB has somewhat inconsistent key names for their objects, so I had to filter through the results and turn them into my own movie objects that included things like the title, the description, and a url to a poster image. TMDB also gives each movie a unique 5 digit ID which is very helpful. Here’s what one movie object looks like.

the data structure of an individual movie

   The users each add as many movies as they want to their personal list and when everyone is done (or the host decides to stop waiting) the state is updated and everyone posts their movies to the master list in the database, making sure to ignore duplicates. They get brought to a screen where they can see all the movies and then everyone pretty much just waits on the host to decide what to do. The host sees three buttons for their three options: pick randomly, ranked-choice voting, and approval swiping. These three choices were a key part of my initial design plan and I think they all complement each other nicely. The important concept is that you can use them all successively. Movies that make it through each round stay in the running and those that fail will become inactive. The next action you take will only affect the pool of movies that are still active. This lets you start with the most broad topics to narrow it down to the movies that people will generally like and then you can pick randomly from there.

   The random selection was of course the easiest to implement. Rather than just picking one winner to keep active it actually shuffles the order of the whole list of active movies. This means that if you really don’t like the result you can always take the second choice and so on. The swiping screen is a lot like Tinder. There are truly only two types of new apps that can come out these days: it’s either “Tinder for X“ or “Uber for X”. Well, I’m not about to make an exception to that rule. This screen displays large cards for each movie that have ratings and descriptions and you can either give them a thumbs up or a thumbs down on whether you’d be willing to watch it or not. Every movie that has at least 50% approval gets to move on. The trickiest part of this was letting you actually swipe the cards because I didn’t feel like doing this in a great way. I decided to use more Javascript but I bet you could even do it using the HTML drag and drop functionality. Dragging the mouse adjusts the absolute position of the card and when it passes a certain threshold it triggers the animation to send it away and bring in the next card. This works well enough but it’s a bit janky. If I was making a whole app based around this I’d be sure to do it better.

   The hardest action to implement was the voting. I wanted to do ranked-choice voting because I think it’s really interesting, but it turned out to be pretty complicated to actually program. The UX element of it was easy, it’s just creating an ordered list. When it’s time for the host to compute actual rankings from it I pass it off to a very large algorithm. Ranked-choice voting isn’t usually too complex. Voters rank their choices in order of how much they want them and if their favorite is eliminated then their vote just transfers down to their next choice in the line. What made this more challenging was that I wanted there to be an arbitrary number of winners as decided by the host. The algorithm essentially runs through this flowchart over and over until there are no more votes left to distribute.

the ranked-choice voting algorithm explained

   What also makes it harder is that you don’t have millions or even hundreds of voters. You can’t guarantee that through enough rounds of the RCV algorithm that anyone will have enough votes to win. If there still aren’t enough winners then it triggers the failsafe and just picks whoever was ahead until the threshold is met. I’m confident that my implementation of this algorithm isn’t perfect, but it’s definitely good enough for what it is. I would probably have been better off just assigning a number of points based on where the movie ranked and then totalling all of those and comparing them but that’s not nearly as fun.

   That’s all the core functionality and I’d say it definitely satisfies the job I set out to complete. Right at the end I realized that I should add a very important feature which is host transferring. Before this anyone could leave and come back and it would be fine except for the host. If the host didn’t check in for 90 seconds then they would be marked as offline and everyone would be kicked back to the lobby. This is still how some big budget multiplayer video games do things but almost all users would agree that it really sucks when that happens. I realized that I could make it so that when there is no host, one person would be singled out as the new host and be given all the powers. This also introduced the edge case where the host leaves but then eventually comes back. In that case one of the players would assert themselves behind the scenes as the true host and the other host would step down.

   One last feature I added was to make it a true progressive web app. This basically just involved adding a service worker and some extra web manifest files but it meant that I could download the website as an app and view it without the url bar. I definitely want to do more stuff with PWAs in the future. They let you do things like run background tasks, send push notifications and store things offline which are all really nice features but I didn’t need to use any of them here.

   Overall I’m impressed that this holds together as well as it does. It’s not great but I’d say it’s probably better than the sum of its individual parts and it does what I set out to do. There were some really interesting algorithms, mostly related to the host’s computation, but most of the project was just a lot of jQuery. I don’t love that kind of frontend development but it can be fun sometimes. I just don’t think the end result looks like I put in as much effort as I actually did.