Adventures in working with JavaScript, Dart and Emojis

I spent the whole day today working with Strings being sent between a JavaScript serverside app and a Dart clientside app. As an industry we have been doing this forever, so you’d think it’d be easy but then along came emojis to muck up my day 🤬

Instead of writing my own primer on Strings here (and doing a bad job), let me just link to Joel Spolsky’s excellent post on the subject

This really old post still does a great job of bringing us up to speed to the Unicode world we live in today. And then came Emojis

There are numerous posts of the pain of dealing with Emojis whenever you have to because it does screwy things like combining neighboring characters to form a single emoji. This means that the length of a string, if it is just a measure of the unicode CodePoints used is different from what you would count on the screen.

This gives you whacky results like “💩”.length == 2 and generally makes working with strings just a pain even to the extent of crashing your iPhone. On the flip side some things like being able to delete family members from the 4 member family emoji with every backspace are kinda amusing, since it is it’s actually 7 characters: 4 ‘normal’ people characters and 3 invisible ‘joining’ characters in between.

Which brings me to today. I am playing around with moving a client/server app from JavaScript everywhere to JavaScript server and a Dart client app. In the previous iteration, strings that needed to be sent had special characters that needed to be escaped and sent across: no problem. JavaScript’s escape/unescape worked pretty well.

Moving to Dart though was a challenge, because there is no escape/unescape method. Turns out escape/unescape is best avoided anyway, and encodeURI/decodeURI is a better option. Dart has a decodeFull method on the Uri class that does the job pretty well.

Except that the characters in the list also included emojis and Dart’s Uri class doesn’t work with anything more than UTF-8 characters and crashes when encountering strings with emojis that are just ‘escaped’. This, as it turns out, is as per spec and all those fancy emoji domains that I thought used Unicode in the URI, use a different idea around Internationalized Resource Identifiers and Punycode. Thankfully passing in a URI encoded string with emojis seems to work fine and emojis come out 👍on the other side of the decode process.

While this seemed to work at that point, passing the decoded string to my Yaml loader crashes the app again (is Yaml supposed to be restricted to Ascii/Utf-8 ? ). But that is a problem for a different day.

For now, I have decided to just convert emojis to shortcodes for the transit and remap them to emojis on the other side. Its not pretty but it works.

Oh and in the meanwhile, if you want to know how to loop through a String with emojis in Dart, you can do that by looking through the Runes in a String:

A Quick 2018 Retrospective

I am a little behind on my 2018 review, seeing that its almost mid Jan already. But better late than never, so here goes

VR

2018 continued to be a year of tremendous education professionally. For a big part of the year I continued to work in the VR space, specifically using A-Frame and WebVR. I am definitely a fan of WebVR and I really do believe that the app model that is prevalent in the mobile ecosystems is a bad idea for VR. I completely agree with the vision of VR being a web of experiences, and the web technology stack has matured to a point where deploying a VR experience is trivial. Hopefully more people will take up building VR experiences in JavaScript and WebVR. What the community really needs is a diversity of ideas and to grow VR beyond its early base of gamers.

While it was a lot of fun, I am looking at other things beyond VR this year and am excited for certain new ideas I am playing with. Will share more on that later.

Blockchains

I did a fair bit of work on Blockchains in 2018, mostly at the Dapp level. Its early days for this space but I do believe they present a once in a generation opportunity for a step function change in how we use technology. There is a lot of pessimism about the space right now, after the unrealistic craziness that was the 2018 bubble when Bitcoin hit $19,000 but I am excited about where the tech is going.

I spoke at a panel at the Coinvention Conference on the Philly Blockchain scene (Thanks Mike) as well as at the inaugural session of the Drexel Blockchain club (Thanks Adit)

That’s me, that pink blob there!

Besides these, I did work on a LOT of JavaScript and Rails which has been pretty rewarding and learned a lot about Google’s cloud infrastructure, specially Firebase. I need to distill a lot of that into a future blogpost

Reading

I did a fair bit of reading this year but I did abandon a lot of books halfway. I am trying to be okay with that rather than pushing through a bad book, just to complete it. I wish there was a better app than Goodreads for books though

Misc

Some other things that happened this year:

  • Philly Google Developer Group (GDG) continues to go strong in its 8th (!) year since its start as AndroidPhilly in 2011. Its a great community that I look forward to meeting at least once a month and have made some great friends there.
  • I didn’t travel as much for work this year, which was good. My favorite event though was the MIT Media Lab’s Fall Member Event. I do like all the demos that the Media Labs groups present but the best part is the talks with other sponsors from different organizations.
  • I worked with a lot of interns and co-ops this year, mostly from Drexel, and I loved it. These guys and girls are smart, enthusiastic and I find conversations with them refreshing since they question a lot of assumptions I often have. Maybe I should consider some work in academia 😉

2019 is guaranteed to be a year of many changes and I am excited for most of them. Stay tuned.

On Rich Text Editors

In the last couple of years I have found myself in a couple of projects using web based rich text editors. Since these projects were written using React, I primarily focused on libraries that worked with that framework but have been keeping an eye on other projects as well. This post is a braindump of my thoughts on the space

Many years prior to that a few friends of mine and I had tried building a web-based Evernote clone that would do smart things based on patterns we could define in Regular Expressions. So for example, if I typed in an email address in my document, the JavaScript would recognize that and convert it to a mailto: link. With the advent of contentEditable, we thought this would be easy to do

We were totally wrong. Turned out contentEditable is a terrible API as documented in this article by the Medium engineering team. Modifying the underlying document while the user was editing it was just impossible to get right even on one browser. And for bonus pain points, every browser wrote different underlying HTML when the user edited a contentEditable element.

DraftJS

About 3 years ago I was assigned to work on an online CMS for an internal portal at work and started looking at RTEs again. Since the rest of the project was going to be in React, I started looking closely at React libraries. React had one advantage that didn’t exist when we had tried our earlier adventure: Since React keeps the document structure in its Virtual DOM, it didn’t have to fight with the browser specific implementations of contentEditable as well as fight the browser for things like cursor position and selections.

DraftJS seemed like the best choice at the time. It was (and still is) actively developed by Facebook and is used by them for text editors on Facebook.com.

Generally it worked well enough. We could have used more documentation but were able to get a default experience working. Draft comes with very few extensions and really tries to sell its “toolkit” nature by having you code behavior I’d have expected as default. There is a different project called DraftJS-Plugins that gives you a lot of that. Weirdly, DraftJS-Plugins requires you to use its Editor which seems to be a modified version of the DraftJS editor. Ideally Draft should support plugins out of the box.

We did make a mistake with storing the document though. Our thinking was that the document in the database should be saved as HTML and Draft should just re-render the HTML when the document needs to be edited. Turns out Draft really prefers saving and loading a serialized state of the document. This does make some sense but makes any future migrations to a different editor much harder. This brings me to the next library I looked at

Ghost

Ghost is a pretty interesting CMS. Visually its very polished and the editor interface there is pretty amazing. Earlier versions of Ghost used an editor that really was centered around Markdown, which is awesome but not a fit for the target audience of our system. Ghost also had very limited database support: basically SQLite for development and MySQL for production. Since we didn’t want to use MySQL it was a deal breaker.

But from the point of view of Text Editors, Ghost 2.0 that was recently released shipped with a new and much nicer editor. The editor is now part of the main Ghost repo but I imagine can/should be extracted to its own project.

Whats really interesting about the Ghost Editor is that they also released a library for building WYSIWYG editors supporting rich content via cards called MobileDoc-Kit and an open source spec for these documents called MobileDoc.  This addresses my issue with storing a proprietary format for a document in the database. So yay!

WordPress Gutenburg

Meanwhile WordPress has also been working on their own new editor that just shipped as part of WordPress 5.0 today. In a big departure from previous PHP centric WordPress editors, Gutenburg is written in ReactJS. Additionally, unlike Draft (as well as Ghost 2.0 editor I think), Gutenburg works on mobile devices. My biggest issue is that I really hate WordPress’s “HTML comments as a data structure” approach. And I am not the only one. Using Gutenburg has so far been pretty okay. If you’d like to learn more about it, read Matt Mullenweg’s post on it today

CKEditor

Other projects also crossed my path during this project. CKEditor for example looks promising but I have’t dug into it. What makes it really interesting is its newly added support for collaborative editing out of the box, something I’d love to add to my project but its not a big ask (yet)

Trix

If Rails is your thing, Rails 6.0 will include a Rich Text Editor out of the box which is really cool. The editor itself, Trix, is out right now and can be added to your project if you need it. 

Final Thoughts: Blocks

One interesting thing is that almost all the editors are starting to move into the concept of a rich document as a group of content blocks. This is a departure from earlier architectures that gave you the entire document as a rich document with all the formatting. One reason for this is the ability to export the content blocks in a variety of formats besides simple HTML (like Markdown). 

Blocks also let you use non-text elements within the document. These can be things like media or even rich widgets like photo galleries etc. Given a well defined document data-structure, these can also be supplied by third-parties.

For example, Elementor has released a bunch of widgets for WordPress’s Gutenburg

Modern word processor centric startups are taking these ideas even further. For example,Notion takes the same block based approach to a person notes app

AirTable, which takes a similar block centric approach to spreadsheets, was recently valued at $1 billion 

Both Notion and AirTable measure usage and price their plans not on the basis of number of documents but rather the number of blocks used.

Trying out Google’s Cloud Endpoints vs Amazon API Gateway

I was recently looking to scaling up an API currently hosted on Heroku.com. While adding dynos to Heroku was an option, I also thought it was a good excuse to get more familiar with Google Cloud Platform (GCP), which I have been curious about for a while and have had some really good conversations on at previous events at GDG Philly.

GCP’s Cloud Endpoints seemed to be the right answer, especially since they do claim to support hybrid cloud deployments.

After setting up the Google Cloud SDK   and enabling billing on my active project, I clicked on the Cloud Endpoints left nav item to get this screen…

Turns out, the Cloud Endpoints service has no user interface by default. To get started, you have to deploy your api spec (as a Swagger2 / OpenAPI document) to see anything at all. Maybe that works for folks that live and breathe the command line but I was kinda put off, specially since I was aware of how AWS API Gateway worked (we’ll look at that later)

Anyway, so clicking on “Learn more” and then a further link to the documentation takes you to this page with the following table:

So based on their documentation, it seems that the only JSON apis that Cloud Endpoints can front are those hosted on Google App Engine or Compute Engine.

Argh! (Though gRPC does look kinda interesting, just not for now)

At this point I actually debating deploying the Rails app on the App Engine. Ok, let me try out how hard that is. After all they have a blog post on how to do it.

After an hour of configuring, I was able to deploy my Rails app to Google Cloud, but it took me too much effort to connect it to the CloudSQL backend and even more to try to get migrations running (never did, and I finally gave up). Google does have a video on how to get rails migrations working using their Cloud SQL proxy   but I was kinda done by then. This was taking me down a path I wasn’t really planning to go down.

Getting it work on Amazon’s API Gateway was so much easier.

The AWS console for API Gateway actually has a fairly intuitive user interface for fronting any third party api with API Gateway. At the end, you end up with a nice UI that describes the flow from the client through the gateway to your api base URL (Screenshot below)

I generally was surprised how easy AWS made their service and hopefully Google Cloud Endpoints reach the same level of simplicity. I am sure given how much Google is investing in their cloud services, this post will be out of date in a couple of months but for now AWS API Gateway does seem to have the edge

Using ActiveStorage with DropZone in a React app

The last year or so I have been working on an app for teaching programming that I hope to release in a few weeks. The CMS for the app is a React based web app powered by a Rails server. The Rails server is running in an API-only mode which has been working out reasonably well.

One of the recent additions to Rails has been the ActiveStorage system that purports to make attaching media to Rails models really simple. Unfortunately most of the documentation around it involves using Rails’ View tier as well which the API-only mode strips out. In fact, getting there is an open bug on ActiveStorage to actually make it work out of the box in API-only projects.

Oh and one more thing: ActiveStorage doesn’t support uploading to multiple buckets, just in case thats a dealbreaker for you.

After a couple of days of struggling through it, I finally have it working in an API only mode using the React Dropzone Component instead of a boring filepicker.

Also note, there is an open source React component for interacting with ActiveStorage which I did not try, mostly cause I was already halfway done with the Dropzone implementation by the time I discovered it.

Issue #1: ActiveStorage CSRF errors in API only mode

ActiveStorage ships with a JavaScript library that does a lot of the work, but one of the issues with using ActiveStorage from a react app was that it tries to do CSRF token validation on the requests but cannot. To solve it, I added a initializer in the initializers folder that skipped CSRF checks for it

ActiveStorage::DirectUploadsController.instance_eval { skip_forgery_protection }

Issue #2: Cors challenges

So this isn’t a bug but when working on a React app using the CreateReactApp project, the React app runs on port 3000 while the Rails server runs on port 3001. To prevent CORS issues when uploading images, I had to add the Ract::Cors gem and a cors initializer that allowed requests from localhost:3000 during development

Rails.application.config.middleware.insert_before 0, Rack::Cors do
 allow do
   origins 'localhost:3000'
   resource '*',
    headers: :any,
    methods: [:get, :post, :put, :patch, :delete, :options, :head]
 end
end

Also, since the React app was using DirectUpload to S3, you have to add the right Cors settings for your S3 buckets as well (this blog post helped figure out the CORS issue)

Uploading an Image using React and Dropzone:

The core part of the image uploader component is below:

const url = '/rails/active_storage/direct_uploads'
const upload = new DirectUpload(file, url, this)

upload.create((error, blob) => {
 if (error) {
    console.log("Image Error:", error)
 }
 else {
    this.dropzone.removeAllFiles(true);
    this.props.onImageUploaded(blob)
   }
})

The DirectUpload class comes from the ActiveStorage JavaScript file. When this code is run, the image is uploaded to the storage provider and you get back a blob (just a JavaScript hash)  with a bunch of metadata (that is saved in the `active_storage_blobs` database table.

One of the keys in the hash is called signed_id. To assign the uploaded image as an attachment to any of the models, the model has to have a has_one_attached or a has_many_attached field. In my case its something like

class Book < ApplicationRecord
  has_many_attached :images
end

In my book controller I can then accept a PUT request that updates the record like so:

@book.images.attach( params[:signed_id] )

or better yet just let strong_params do it automatically.

[Update] Another good post on using pre-signed POST requests for S3 Uploads

Using Symlinked Node libraries with React Native 0.55

I recently updated the React Native app I have been working on for a while from RN 0.47 to 0.55. I’ll admit I was a bit callous about the update and hadn’t really looked at the change log, but hey, version control does give one a foolish sense of bravado.

Anyway, needless to say there were issues. As of RN 0.55.4, the `setJSMainModuleName` has been renamed to `setJSMainModulePath` and it took me a bit of sleuthing to figure that out (Find the Github commit here)

However a bigger issue came up when I tried to package the app after resolving the compile errors.

Screen Shot 2018-05-16 at 1.19.27 AM

Turns out the new Metro packager cannot follow symlinks, like those created by npm-link

This was a total fail for me, since my app uses local npm modules to hold pieces of common code for the web and mobile clients.

Thankfully someone did come up with a bit of a hack that generates absolute paths for all symlinked libraries and launches the cli.js file of the packager with a config file with the list of absolute paths.

It works for now, but hopefully this bug will get fixed soon.

Building Smart Contracts on the Ethereum Blockchain

Last week’s Comcast Lab Week gave me another opportunity to dig deeper into Blockchains. In my previous writeup on CodeCoin, I had used Ethereum to create a bounty system for Github issues. However under the hood, we had cloud servers managing various wallets belonging to the different issues. Smart Contracts offer a better way to handle this.

What Are Smart Contracts?

Smart Contracts are pieces of code that execute on the Blockchain. Think of them as classes in an Object Oriented Programming model. Once deployed, you can invoke methods on the Contract from any wallet on the Blockchain. The method, when called, gets not only the parameters that you explicitly sent as part of the method call but also the caller’s wallet address and any value (Ether or its smaller fraction Wei) that the contract was sent. The contract can then keep part or all of the value sent in return for executing the code.

The world’s simplest Smart Contract.

Note that the above contract is a “hello world” contract. Ours was a bit more complicated.

Tools and Setup

Similar to the last time, we used the TestRPC program, now rebranded as Ganache, or specifically it’s CLI  version, the Ganache-CLI) to develop and test the application. The app itself was pretty simple: It allowed users to rent an asset in our store by sending a particular amount of ether to our smart contract.

ganache setup
Configuring Ganache

In addition to Ganache, we also used the Truffle framework to build the application as well as MetaMask to run the transaction.

The Smart Contract itself was written in Solidity. Even though there are may editor plugins you can add to your favorite editor, I found using the the cloud hosted Remix IDE the best to get started. It’s already configured with linters and debug tools that make developing your contract much simpler.

The Truffle framework can be thought of as an equivalent to Ruby on Rails, just for Ethereum projects. When you start a new project, it creates a project structure with folders for contracts, migrations, tests and a truffle.js configuration file that lets you deploy your code to various dev/test/prod environments (think of truffle.js as a config.yml from Rails)

Once you place your contract in the contracts/ folder, running truffle compile compiles the Solidity code to the bytecode that can be deployed to the Ethereum Blockchain. Running truffle migrate deploys the contract to the chain. Truffle also provides a handy REPL console that you can use to interact with your contract from the command line. Very convenient to find your contract’s deployed address for example by calling

Truffle Console
Find the address of the SimpleStorage Contract

Running the Application

Our client application ran a small React app that would send a little bit of Ether to a deployed smart contract. If the transaction was successful, we’d send the successful Transaction ID to a middleware server that would validate it and authorize the user to a piece of content

The client code communicates with the Blockchain via the Web3 library injected into the browser by MetaMask. The code to use the contract looks something like this:

That’s pretty much it.

Random Learnings

  • We had issues with MetaMask and Ganache seeing each other’s wallets. This might be by design but to get anything done, I had to write another small server script that funded Metamask accounts with Ether from Ganache accounts
  • All transactions are done in Wei (10^-18 Ether). During development we’d send small values like 10 Wei across and were perplexed that Metamask or other wallets didn’t change their displays till we realized that the number was too small for MetaMask to show in its UI
  • We assumed a successful transaction on the contract assumed successful payment. We did not wait on the transaction to be mined to declare payment success. We should be waiting on that using the Web3’s filter API but we just ran out of time on the project.
  • To connect with a contract using Web3, you need to point it to its address and the JSON interface of the contract. Truffle’s console has a `toJSON()` method but that is not the JSON you are looking for. The right JSON file is located in the `/build/contracts/` directory. Once you have that JSON file you can create the Contract object in Web3 by using
    var myContract = new web3.eth.Contract(SimpleStorageABI, address)
  • We found an interesting project called OpenZeppelin that seems to be a public repo/tool for often used Contracts. I need to try that next.

Final Thoughts

This was a fun 5 day project that demystified a lot of things around Smart Contracts that I wasn’t sure about. And as always, it was great to work with a bunch of smart engineers I usually don’t get the opportunity to work with otherwise 🙂