Hey, I'm Christopher Chedeau aka Vjeux! I started this blog to talk about the various projects I am working on and to reveal some of my programming tricks! I hope you will find some of my stuff fun if not useful 🙂

I'm a Facebook Software Engineer in the Front-end team working on React Native. Before that, I went to EPITA, a 5-year Computer Science school and majored in its R&D lab LRDE. I also worked for Curse during the nights and week-ends.

During the past few weeks, I've been working on prettier, which is a JavaScript pretty printer. We are approaching the phase where we can actually use it so this is a good time to explain how it works.

We're going to go through an example

if (!pretty) { makePretty() }

String -> AST

The first step is to take this string that represents some JavaScript and to parse it in order to get an AST out of it. An AST is a tree that represents the program. Using either Babylon or Flow we can parse this example and we get the following tree.


You can explore the full AST using astexplorer.net.


Now that we have this tree, we want to print it. For each type of node like IfStatement, UnaryExpression... we're going to output something. In the case of prettier, this something is an intermediate representation called a document as described by the paper a prettier printer by Philip Wadler.

    "if (",
    group([ indent(2, [ softline, "!", "pretty" ]), softline ]),
    " ",
    indent(2, [ hardline, "makePretty", "()", ";" ]),

You can play around with this representation on the prettier explorer.

IR -> String

The interesting thing about this representation is that it is the same no matter what the line-length is. The basic idea is that the primitives such as group, indent, softline encode the way they should look if they fit in the line or if they don't.

The most important primitive is group. The algorithm will first try to recursively print a group on a single line. If it doesn't fit the desired width, then it's going to break the outer group and keep going.

Then, we have primitives that behave differently if they are in a group that fits a single line or not: softline that does not print anything if the group it is contained in fits and a line otherwise. indent adds a level of indentation if it doesn't fit. If you are curious, you can look at the short list of available commands.

So, we just need to take this IR, send it through a solver along with the desired line width and we get the result!



Hopefully this gives you a better idea of how a pretty printer that takes into account the desired width work.

I had the opportunity to do a live fireside chat with Lee Byron to chat about React Native and GraphQL. I outlined the boring future of React Native 😉

It was pretty surreal as I got to have makeup done by a professional!


I also tested the live video feature of Facebook on the open source booth at f8. I showcased React Native running on Xbox, Windows Phone and Desktop and the F8 app being open sourced.

Yesterday, there was a big discussion on Twitter on how hard it is to start hacking on a js project. One comment by Dan Abramov struck me in particular: "Right: don’t use tools, face a problem, choose a tool or roll your own. Wrong: learn tools that don’t solve your problems, hate the tools."

This is spot on. All the solutions presented in this thread do not solve the problems I have when I'm hacking on a new project.

My dream PHP Setup

Back when I was 14, I had the best setup I've ever used in my life. Here it is:

  • Launch WinSCP and connect to fooo.fr.
  • Create a file test.php, write <?php echo 'Hello World'; locally with EditPlus 2, press cmd+s to save, wait 0.5 second for the upload.
  • Open my browser and go to http://fooo.fr/~vjeux/. Click on test.php.
  • ?????
  • Profit


I want to get the same attributes but with JavaScript and React. Here are the constraints:

  • No setup: I'm happy to have to setup something initially (dedicated server, apache, php...) but nothing should be required to create a new project. No npm install, react-native init, creating a github project, yo webapp...
  • One file: I want to write a single js file to start with. No package.json, no .babelrc, no Procfile...
  • Sharable: I want to be able to share it with a url without any extra step. No git push heroku master or git push gh-pages.
  • Keeps working: Once online, it should stay there and keep working 6 months later. No npm start to run it, no special port that's going to conflict with my 10 other prototypes...
  • Not generic: I don't care about it being generic, I will use whatever transforms you decided. Happy to write js without semi-colons and using SASS for my CSS if you checked all the boxes above.
  • Not prod-ready: This setup doesn't have to be prod-ready, support unit testing or anything that involves it being a real product. This is meant for hacking on stuff. When the project becomes good, I'll spend the time to add all the proper boilerplate.

So, that's the challenge. Can you make it happen?

One common pattern when implementing user interface optimizations is to compute some value for a node where the computation involves looking at neighbor nodes and want to keep this value updated when the tree is mutated.

On this article, I'm going to explain the pattern we implement to solve this use case on various places in React Native.

Example: background color propagation

On React Native, we implement an optimization where the background color is propagated from the parent instead of being transparent. This provides a guarantee to the GPU that it won't need to paint pixels that are underneath. You can read this release notes for a more complete explanation.

In any case, the algorithm is pretty simple, if the background color is not set on an element, we take the one from the nearest parent that has one set. Here's an example:

Now, let say that the red node background color is being unset, the example would look like:

In order to implement this behavior, we first need to traverse up the hierarchy and find the background color and then down to propagate it, but stop at nodes that have set background colors. This means that we have to implement two different algorithms: one for the initial rendering and one for the update.

The complexity of this example as explained is small enough that it is easy to maintain the same invariants. But in practice, this algorithm is a bit more complex: we don't forward the color for transparent nodes nor for the image component in certain cases... We also experimented with more conditions that we didn't end up using.

Dirty-up and execute top-down

What would be great is to just write the top-down recursive algorithm you learn in school to apply colors however you want and whenever there's a mutation just re-run it on the entire tree. The problem with that approach is that you're going to spend a lot of CPU time running this algorithm on the entire tree when you only need to update a few nodes.

Instead, you can dirty the node you are mutating and all the nodes up to the root.

Then, run your algorithm top-down starting at the root.

You need to implement a way to figure out if a non-dirty node needs to be recomputed. The strategy we use is to cache all the arguments of getColor, for example (parentColor, nodeColor, isNodeText, ...) along with the result. If we're being called with the same arguments and the node is not dirty, then we don't need to go further and can just bail. The pseudo code looks like this:

function getColor(isDirty, prevArgs, prevValue, nextArgs) {
  if (isDirty || prevArgs !== nextArgs) {
    return realGetColorRecursive(...nextArgs);
  return prevValue;  


The benefits of this technique is that you can write your business logic once in a traditional top-down recursive way. You don't need to care about the much harder update case.

The downside is that you are doing more work: instead of touching only the elements that have changed, you need to traverse everything up from the root.

Another downside is that it requires more infrastructure as you need to add a caching layer and refactor your code to add a pass that starts at the top.

We're making use of this technique for a few aspects of React Native:

  • Background color is propagation as explained in this article.
  • Computing the layout (top, left, width, height) values based on the css attributes (eg marginTop, flex: 1...).
  • Flag if a text node is the root of the tree of text nodes to be able to flatten it into a single native view.
  • Flag if a node only affect layout and doesn't paint anything so that it can safely be removed from the UI tree.

When I was 14, Philippe Pelletier reached out to me with an interesting project. He is passionate about cinema and maintains a database (lots of physical sheets of paper and a growing number of Word documents) of every movie an actor played in, the cover of the movie, a biography of that person. He had the idea of turning this into a website to share this work.

I built the website called CinéArtistes over the summer and forgot about it. Turns out, 10 years later with no code change, it is still running and growing!

Finally something is broken!

Philippe contacted me this week because the website was getting unstable and those errors started appearing more and more regularly over the last few months. It has gotten to a point where it was unusable at peak hours.


As a first glance, it looked like the mysql database was saturated. When I went to cineartistes.com, the page would take something like 10 seconds to load. My intuition was that some queries were not scaling properly with the growing number of entries in the database.

When I browsed around the website and went to an actor page, it was actually really fast. This confirmed my suspicion that this was not a site-wide issue but just a problematic query on the home page.

Going down memory lane

Now that I had an idea of what was going on, I asked Philippe how do I connect to the server and he gave me a FTP url, username and password. It's been a long time since I haven't used FTP for anything!

The entire website was a single file 2800-lines file called index.php! Mind you, there isn't any version control and the staging environment is a file called index_dev.php.

Even though today it seems crazy, it's a good reminder that tools are there to help you achieve your goal. This codebase has worked flawlessly to power the website for more than 10 years with 0 intervention.

It is also surprisingly well structured. There's a router that calls a specific function for every page, good thing that at the time having pretty URL were not a thing 🙂

switch ($_POST['page']) {
  case "admin": OnAdmin(); break;
  case "import": OnImport(); break;
  case "modif": OnModif(); break;
  case "modiffestival": OnModifFestival(); break;
  case "ajout": OnAjout(); break;
  case "ajoutfestival": OnAjoutFestival(); break;

I already structured my code as components:

Liste($rech_nom, $rech_naiss, $rech_deces, $rech_activite);

Of course, the implementation isn't fancy: echoing concatenated strings for most of the code and ending php interpolation ?><html...><?php for static things.

Finding the performance bottleneck

Time to go back to the task at hand, we need to figure out why the front-page is so slow. My intuition is that there is a query that runs slowly, so I want to instrument all the mysql_query calls and log the time each one takes.

Thankfully, there is a database abstraction in place $db_mysql->query(...)! This kind of blows my mind! I probably got super annoyed that the mysql API was so hard to use. The abstraction code doesn't look like it was coded by me, I probably copy and pasted it from somewhere else 🙂

Since all the calls are centralized there, I just need to do

$t = microtime(true /* Seriously php ... */);
// do the query
print($query, (microtime(true) - $t) * 1000);

But there's one little problem, how do I print? If I'm using echo, it's going to show it in the middle of the content and everything will look broken. I could buffer it to a global variable and print it at the end but there's a better solution: I can use JavaScript console.log!

function print() {
  echo '<script>console.log(';
  foreach (func_get_args() as $elem) {
    echo '"'.$elem.'", ';
  echo '"");</script>'; // notice the "" in order to avoid dealing with trailing comma

At first I tried to use json_encode to be safe if there was a " in the query but guess what, this 10 years old version of PHP doesn't implement it!

I was expecting to see one query takes many seconds but I was a bit disconcerted when the most costly one would only take ~100ms. It turns out that this one was being executed 80 times in a row!

SELECT COUNT( * ) FROM `images` WHERE `from`='4020' AND `type`='2' 104ms
SELECT COUNT( * ) FROM `images` WHERE `from`='4019' AND `type`='2' 117ms
SELECT COUNT( * ) FROM `images` WHERE `from`='4019' AND `type`='2' 101ms
SELECT COUNT( * ) FROM `images` WHERE `from`='4018' AND `type`='2' 125ms
SELECT COUNT( * ) FROM `images` WHERE `from`='4018' AND `type`='2' 104ms
... 80 times ...

This is a classical N+1 problem where we first query for the the elements in the list and then send one query for each one. The proper way to fix this is to refactor the code to merge the inner query in the outer one, but this is super annoying to do in practice.

At Facebook, we use GraphQL and Relay which solves this problem elegantly: it lets you write the queries in a nested fashion as it is in this example, but has a pre-process step that merges all those queries into one.

Make it fast

Anyway, I first wanted to figure out if I could optimize the query instead. It is just being used to check if an actor has at least one image of type 2 (a cover photo), it really shouldn't take 100ms.

I saw three possible improvements:

1) COUNT(*) is wasteful because we don't care about the total count, we just want to know if there's at least one. I learned the hard way that at Facebook, count is extremely expensive because you need to privacy check all the elements in order to know the real count. This means that count is as expensive as fetching all the elements. But in this case, it probably isn't doing such dramatic things.

2) While searching, I found someone writing that if you search for a string on an int field, mysql would be much much slower. That seemed to make sense so I removed ' around the value and unfortunately it made the query go 3 times slower!??!? This wasn't the quick win I was hoping to see.

3) I've read many times that adding an index is super important but had never done it before, so I google'd for it and found this command:

ALTER TABLE `images` ADD INDEX (`from`);

After running this single line of code, the time it took went from 100ms to 0.5ms! So, instead of spending 8 seconds to run those 80 queries it only took 40ms. Problem solved.



Philippe asked several people to try and fix this problem before reaching out to me and their response was along the lines of: this website is built using ancient technology, you need to port it to <insert framework/language name>.

Rewriting is usually the first response but I've learned that it is usually not the best answer. In this case, it took one hour end to end to fix the issue. It would have taken a month+, a lot of money and time and disrupted Philippe flow to bring a new website for no obvious gains.


Looking back at my code, I can't help seeing many good patterns such as use of components, wrapping mysql calls into a helper, centralizing routing code... But, they were not being used consistently and there was also a lot of terrible patterns. So the fact that they were there is probably a mix of luck, intuition and part of trial and error.

After 10 years, I am now able to pick out every single pattern in this code and talk about the pros and cons and decide whether it would be a good or bad idea to use there. Experience made me a lot more confident and intentional when I code.

Practice makes perfect

If you were to give a similar task to my 14-years old self, I would probably have been able to figure out that I needed to instrument the mysql_query wrapper and add an index. But I bet it would have taken me multiple days to do that.

The reason is because it required me to execute a lot of auxiliary tasks such as

  • Setup the dev environment to update files on a remote FTP on save.
  • Figure out how to find out the code responsible for outputting an element on the screen.
  • Write PHP code that can be run on a 10 years old version (we are spoiled with Hack at Facebook!).
  • Implement a print function with variadic arguments that outputs console.log using a script tag and know that I didn't need to implement escaping.
  • Connect to a mysql instance and select the right database in order to be able to run the query.
  • ...

I know that I struggled with each of those for many days/weeks/months in the past, but at this point, I've done variations of those so many times that I have muscle memory for all those tasks. I also used Google extensively during this hour, not to learn, but to remind me how to do those tasks.

Final thoughts

It will be interesting to read this article in 10 years and figure out what I was not seeing while writing this article. It'll certainly be more meta 🙂