Archive for the ‘JavaScript’ Category

Don’t use JSON for configuration files

Monday, April 25th, 2016

For quite some time, I wondered about this: “why the hell are comments forbidden in json files?”.

The short answer is: Douglas Crockford cared about interoperability (https://plus.google.com/+DouglasCrockfordEsq/posts/RK8qyGVaGSr).

The problem is that nowadays, many CLI tools make us of json files to store their configuration. It’s nice because the syntax is pretty lightweight and because it’s really easy to parse, but that’s where it ends because you know what? Comments are pretty darn useful in configuration files..

Unfortunately, as it stands, many of those tools (or at least the parsers they rely upon) choose not to accept comments. As Douglas states, nothing prevents us from sending json files through a minifier to get a comments-free version but… but it’s just a pain to have to do that before passing json files around; worse so when you need to have the file available on disk for some tool and even worse when that file needs to have a certain name (e.g., tsconfig.json).

Some tools do add support for comments, but then you realize that any surrounding tools must also accept that, which is often not the case or takes a while to get there. So that’s that, and IDEs which will complain if you start adding comments to json files (and rightly so..).

All in all, my opinion about this matter now is that json is just not the answer for configuration files. Since json does not support comments, then don’t use json, use something else, don’t try to hack your way around.

What should we use instead? Who cares, as long as it supports comments and doesn’t force you into hacks just to be able to comment things that need be!

YAML is one option, TOML is another, XML is yet another (though way too verbose) and I’m sure there are a gazillion other ones.

If you’re in the JS world then why not simply JS modules? There you get the benefit of directly supporting more advanced use cases (e.g., configuration composition, logic, etc).


Static sites? Let’s double that!

Monday, March 14th, 2016

Now that I’ve spent a good deal of time learning about what’s hot in the front-end area, I can go back to my initial goal: renew this Website.. or maybe I can fool around some more? :) In this post, I’ll describe the idea that I’ve got in mind.

One thing that’s been bothering me for a while is the dependency that I currently have on WordPress, PHP and a MySQL database. Of course there are pros and cons to consider, but currently I’m inclined to ditch WordPress, PHP and MySQL in favor of a static site.

Static site generators like Hugo (one of the most popular options at the moment) let you edit your content using flat files (e.g., using Markdown) with a specific folder structure. Once your content is ready for publication, you have to use a CLI/build tool that takes your content (e.g., posts, pages, …) and mixes it with a template.

Once the build is completed, you can upload the output on your Web host; no need for a database, no need for a server-side language, no need for anything more than a good old Apache Web server (or any Web server flavor you like). Neat!

Now what I’m wondering is: can we go further? What if we could create doubly static static sites? :)

Here’s the gist of my idea:
First, we can edit/maintain the content in the same way as with Hugo: through a specific folder structure with flat files. Of course we can add any feature we’d like around that: front matter, variables & interpolation, content editor, … For all that a build/CLI should be useful.. more on that later.

Note that the content could be hosted on GitHub or a similar platform to make the editing/publishing workflow simpler/nicer.

So, we’ve got static content, cool. What else? Well now what if we added a modern client-side Web application able to directly load those flat files and render them nicely?

If we have that then we could upload the static content to any Web host and have that modern Web app load the content directly from the client’s Web browser. The flow would thus be:

  • go to https://www.dsebastien.net
  • receive the modern Web app files (HTML, CSS, JS)
  • the modern Web app initializes in my Web browser
  • the modern Web app fetches the static content (pages, posts, …)
  • the modern Web app renders the content

Ok, not bad but performance could be an issue! (let’s ignore security for a moment ok? :p).
To work around that, we could imagine loading multiple posts at once and caching them.
If we have a build/CLI could also pack everything together so that the Web app only needs to load a single file (let’s ignore the HTTP 1.1 vs HTTP 2.0 debate for now).

In addition, we could also apply the ‘offline-first’ idea: put pages/posts in local storage on first load; the benefit would be that the application could continue to serve the content offline (we could combine this with service workers).

The ideas above partially mitigate the performance issue, but first render would still take long and SEO would remain a major problem since search engines are not necessarily great with modern client-side Web apps (are they now?). To fix that, we could add server-side rendering (e.g., using Angular Universal).

Server-side rendering is indeed nice, but it requires a specific back-end (let’s assume node). Personally I consider this to be a step back from the initial vision above (i.e., need for a server-side language), but the user experience is more important. Note that since dedicated servers are still so pricey with OVH, it would be a good excuse to go for DigitalOcean.. :)

Another important issue to think about is that without a database, we don’t have any way to make queries for content (e.g., search a keyword in all posts, find the last n posts, …). Again, if we have a build/CLI, then it could help work around the issue; it could generate an index of the static content you throw at it.

The index could contain results for important queries, post order, … By loading/caching that index file, the client-side Web app could act more intelligently and provide advanced features such as those provided by WordPress and WordPress widgets (e.g., full text search, top n posts, last n posts, tag cloud, …).

Note that for search though, one alternative might be Google Search (or Duck Duck Go, whatever), depending on how well it can handle client-side Web apps :)

In addition, the build/CLI could also generate content hashes. Content hashes could be used to quickly detect which bits of the content are out of date or new and need to be synchronized locally.

There you have it, the gist of my next OSS project :)

I’ll stop this post here as it describes the high level idea and I’ll publish some additional posts to go more in depth over some of the concepts presented above.


Modern Web Development – Part one

Wednesday, February 17th, 2016

Since April last year, I’ve been plunging again in the world of Web development.. and what fun it has been! In this series of posts, I’m going to summarize the stuff I’ve done last year in order to hop back on the train and I’ll describe what I’ve learned along the way.

At the time, I published two blog posts, which were my way of condensing my vision for an important project at work aiming to modernize the way we create Web applications by going towards a client-side architecture combined with RESTful Web Services on the back-end.

When I started looking back at how the Web platform had evolved during the 2012-2015 period, the main things I had on my mind were:

  • mobile first & responsive web design
  • client side Web application architecture (which I personally consider to be the part of Web 3.0 — seriously, why not?)
  • the new specs that had reached broad support in modern Web browsers and were gaining a lot of traction
  • the offline first idea that these specs made more realistic

I wanted to learn more about AngularJSnode.jsnpm and sass but that was about it. I remember that at first, I had no precise idea yet about the build tool and the build steps that I wanted/needed… I hadn’t even heard about ES6 yet!

Since then, I’ve learned a ton about ES2015, TypeScript, module systems, module loaders, JS frameworks & the tooling around, front-end state management solutions, front-end build systems, project boilerplates, css style guides, quality assurance for front-end apps, unit & e2e testing libraries, … and the integration of it all…

The funny thing is that… I failed to deliver.

Initially, my personal goal was to create a responsive client-side Web app exploiting the RESTful API of my WordPress installation to replace my current theme, but I changed my mind along the way… So far, my site hasn’t changed one bit. I did improve some things though, but that was more around security than anything else.

So what made me change my mind and where did I spend my time?

At first, I was concentrated on the task at hand and I looked at how the HTML5 boilerplate had evolved as I knew that it was one of the best starting points around for creating modern Web apps. My idea was simple: use HTML5 boilerplate or InitializR to get ModernizR… and add some good old script tags… :p

I started with HTML5 boilerplate, but shortly after, I stumbled upon Web Starter Kit which was fresh out of Google’s oven, was based on HTML5 boilerplate and had some really cool features.

It came out of the box with a nice build which included support for JSCS (JS code style), JSHint (JS code quality), autoprefixing, BrowserSync (if you don’t know that one, DO check it out!), sass and ES6 (that was still the name at that point) with the help of Babel, …

 I really liked their setup and decided to use it as basis for my project; and that’s where my trajectory deviated :)

Given that I’m quite curious, I spent a while deconstructing Web Starter Kit’s build so that I could really understand what made it tick. That made me discover npm, gulp and the whole ecosystem of gulp plugins.

I really enjoyed doing so as it has helped me better grasp the necessary build steps for modern Web apps:

  • transpile code (ts->js, sass->css, …)
  • check quality
  • check style
  • create a production build (bundle, minify, mangle, …)
  • execute unit tests
  • execute end to end tests

At that moment, I was happy with the build as it stood so I continued to focus on developing my app. I took a good look at what ES6 was, what it meant for JavaScript, its ecosystem and how Babel helped (was it still called 6to5 then?). Learning about ES6 features took me a long while and I’m still far from done, but it was well worth it. ES2015 is such an huuuuuuuuuuuge step forward for the language.

I also took a glance at Angular 2 which was still in alpha state. It looked interesting but I believed that it would never be ready in time for our project at work (and it wasn’t). Still, I did spend a few days toying around with the alpha just to get to understand the basic principles.. and I must say that I really really loved what I saw!

That quick research spike also made me discover TypeScript.

Having a strong Java & OO background, TypeScript (TS) directly got me excited. I’m a strong believer in strong (heh) typing, and the fact that TS already supported many ES6 features that weren’t natively supported by Web browsers yet was very appealing to me.

Moreover, having dozens of Java developers in our development teams at work, TypeScript seemed really ideal for us as it supports many features and idioms that our developers are very familiar with (classes, interfaces, generics, strong typing, decorators, …).

If you want to learn more about TypeScript, I definitely recommend the Typescript Deep Dive.

At that point, tsconfig.json wasn’t there yet and the most evident choice to integrate the necessary build step was gulp, as advertised by Dan Walhin’s excellent blog post. If I had read more about npm I might have gone a completely different path (i.e., used npm scripts only).. ^^.

At that point, I had to deviate from what Web Starter Kit offered me in order to add build tasks for TypeScript, tslint, etc. Fiddling with the build made me realize that it was quite brittle, so I refactored it quite a lot and tried to improve things (e.g., separate the build tasks in different files, extract the configuration settings, ensure that it would not break the build on each error, etc). I remember that I wanted to contribute back to Web Starter Kit but realized too late that I had made too many changes at once for them to be able to integrate easily (silly me, bummer).

I went pretty far with actually as at some point, I was using TypeScript to output ES6 code that I then sent through Babel, just so that I could use async/await and other things that TypeScript wasn’t able to transpile to ES5… :)

The exercise helped me see how “immature” and “fragile” the whole JavaScript ecosystem was. What I mean by that is that there seems to be only moving parts and each of those parts don’t necessarily keep happy with each other. Not only do too few people really understand what semver actually means and respect it, but everything that shines bright gets replaced faster than the speed of light :)

As a technologist, I love the pace it imposes for the fun and innovation it brings to the table, but it’s also frustrating for many reasons and (should be) quite scary for enterprises (to some degree). People talk about JavaScript fatigue, which is quite a fun way to put it and I can certainly understand the idea now.

One example that I thought a lot about is the fact that each and every front-end project seems to have its own build chain and build configuration that lives within the project, in complete isolation and has to be maintained.

Of course each and every project has its specificities so there really can’t be ONE rigid & reusable solution to rule them all, but the idea of duplicating so much effort needlessly across a whole community of developers violates the DRY principle as much as anything ever could.

Just try and imagine how many people must have used some Yeoman generator to scaffold projects, which now all have separate builds with tasks that all do the same things but are all defined 20.000 times in a gazillion different ways using variable and unreliable dependency versions… :)

When you scaffold a project using a generator, you end up with a snapshot of the template and of the build provided by the generator at that point in time and then it’s up to you to keep your version up to date and to integrate all improvements and bug fixes, assuming you have time to follow that… you poor thing!

Being part of a core software development team at work, my focus is most often on finding reusable solutions to common problems, limiting effort duplication and what not and thus, the front-end universe’s situation seems quite sad in that regard.

Another point that struck me was how limited the main package management solution was. npm is nice and all, but not being able to define some parent/generic/reusable configuration (e.g., like parent pom files in Maven) is kind of surprising. Again, the DRY principle probably corresponds to DO Repeat Yourself in the frontend universe. I’m sure that front-end experts will tell me that you can work around all that in countless ways, but that’s exactly the issue: I shouldn’t have to invent my solution for a general issue people should be concerned about.

To conclude on a positive note though, I do believe that all the tooling DOES bring added value because it makes it possible to manage dependencies correctly, define build steps which execute tests, generate coverage reports (e.g., using Istanbul), generate production builds etc.

This piece is getting a bit long, so I’ll continue my little story in part two!

 


Installing node and npm on Ubutun 15+

Friday, December 18th, 2015

In case you would want to use one of my recent projects (e.g., ModernWebDevGenerator or ModernWebDevBuild) on Ubuntu (or any other OS btw), you’ll need nodejs and npm.

If you’re using ubuntu and go the usual way (i.e., sudo apt-get install…) then you’re in for a bad surprise; you’ll get node 0.1x.y and also a very old npm release.

Actually, the best way to get nodejs and npm on Ubuntu is to use the node version manager (nvm).

nvm can be used to install and keep multiple versions of node in parallel, which is very useful, especially when you have to test your node-based project on multiple versions.

The installation is very straightforward:

curl -o- https://raw.githubusercontent.com/creationix/nvm/v0.29.0/install.sh | bash

After that, close and reopen your terminal. You now have ‘nvm’ at your disposal.

nvm install 4.0
nvm install 5.0
nvm use 5.0

Just with the above, you get two versions of node (along with npm) installed. As you can see, you can use ‘nvm use’ to change the active version easily.

That’s it!


My current global npm packages

Sunday, July 26th, 2015
If you’re familiar with nodejs & npm you already know this (just skip this part), but newcomers should realize that npm packages can not only be installed locally in a project’s folder, but also globally. Packages that are installed globally are.. globally accessible, which is really cool because using npm you can install many CLI tools to streamline your workflow and boost your productivity.
 
To install a package globally, you simply need to use the –global (-g) flag. For example:
 
npm install --global gulp
Note that you can customize where npm stores globally installed packages by creating a .npmrc file in your home folder and adding the following to it:

 
prefix="/path/to/your/global/npm/packages"
You then simply have to add that same path to your system’s path to get all the tools available at your fingertips.
 
Here’s a small list of npm packages that I currently install globally
  • gulp: Streaming build system for the web!
  • babel: JS transpiler. Because we all want ES2015/2016/20xy right now!
  • jspm: JavaScript package manager: one package manager to rule them all. Let’s just forget about npm vs bower vs git vs whatever, just use jspm and be done with it
  • typescript: Ahhh TypeScript, worth explaining in its own post because strongly typed JS is the future and the future is now
  • tsd: TypeScript type definitions downloader. Because TypeScript without type definitions isn’t very useful
  • brower-sync: Easy to use web server that will make your developer life easier: automatically refresh/sync across all connected devices
  • http-server: Minimalist web server (zero-configuration). Using this you can easily serve local content
  • sass: CSS preprocessor (also deserves its own post). Until I free up some time to learn more about PostCSS, I’ll continue to use this
  • node-sass: SASS without Ruby, weeee
  • yo: CLI to run Yeoman generators
  • slush: Another scaffolding CLI (based on Gulp)
  • caniuse-cmd: CLI to easily check browser compatibility of certain features using data from caniuse.com
  • reveal-md: Quickly generate a reveal.js presentation from markdown content
  • superstatic: Nice web server for SPAs
  • bower: Package manager for the web. I install it for older projects
  • grunt: Task runner for the web. Same as above
  • node-inspector: Blink-based debugger for NodeJS apps
  • node-debug: Wrapper for node-inspector
  • speed-test: Test the speed of your Internet connection
  • semantic-release-cli: Ease your life when creating new releases
  • npm-bump: Another alternative for easy releases
  • center-code: a simple way to show file contents in the console
  • npm-check: check for outdated dependencies
  • rimraf: easily delete files (even paths that are too long on Windows)

Quick NPM tip and a little rant about node-gyp

Wednesday, July 1st, 2015

Before I start explaining why I’m writing this, here’s my NPM tip of the day: if you encounter errors pertaining to node-gyp “rebuild”, while trying to install an NPM package, then before wasting precious hours of your life, just try to install using the –no-optional flag; if you’re in luck, that’ll just work (as it did for me in most cases).

Now what the heck is node-gyp? That’s a fair question to ask. As they put it in their readme it’s a “cross-platform command-line tool written in Node.js for compiling native addon modules for Node.js … and takes away the pain of dealing with various differences in build platforms”.

Well the way I now see it, it might just do what they say.. for people who need/care about that, but for the rest of the world and especially people like me who just want to install an npm package and get on with their life.. it’s just trouble and needless time waste.

Sometimes when you try to install an NPM package, there will be some dependency in the tree that requires to be built specifically for your platform and at that point, node-gyp (which is one of the dependencies of NPM itself) might come into play. The issue is that to be able to do its job, node-gyp has some prereqs that vary from OS to OS and those prereqs are not part of node/NPM (you’ll soon understand why :p). If you’re one of the good guys and use Linux (you should… and I should too but can’t) then you’ll be alright: python + make will make your day (you’ll also be fine with OSX).

Unfortunately, if you’re a sad panda working on a Windows box just like me, then tough luck!

Here’s a little overview of the ‘light/small’ requirements that node-gyp has on Windows 8+:

  • Python (2.7.x and NOT 3.x+): ok just with this one I already dislike node-gyp
  • Microsoft Visual Studio C++ 2013: ok, now I hate it. Do I really need 7GB just to get an npm dependency on my machine? Wtf (pre-compiled binaries FTW, if I wanted to compile everything myself on my machine, I’d still be using gentoo..)
  • and last but not least, for 64-bit builds… Windows SDK: are you kidding me?!!

Assuming that you’re motivated, then you’ll go ahead and install these.. try again and… still get the same error?! Gee… Well the thing is that quite some people have encountered this problem and have hopped through all kinds of hoops to finally get it to work. Some have had success by uninstalling all Visual C++ redistributable packages (any gamers around here?), reinstalling node-gyp’s dependencies in a specific order, adding environment variables and whatnot..

In my case I was pretty happy to discover that in all cases, the dependencies that needed node-gyp were optional (e.g., for babel, browserify and some others), so simply avoiding them was fine. If you really do need node-gyp to work then I pity you and your disk space ^^. Just take a look at some of these links and may the force be with you.

What also sucks is that npm install rolls back on error even for optional dependencies although it’s not supposed to..


SyntaxHighlighter brush autoloading

Wednesday, June 15th, 2011

SyntaxHighlighter is a very cool syntax highlighting JavaScript library created by Alex Gorbatchev (if you don’t know about it yet, check out the demo page). As you’ll notice, I actually use it on this very blog.

Using it is pretty straightforward:

  • load the XRegexP library required by SyntaxHighlighter: http://xregexp.com/

  • load the main script:

    <script src="js/shCore.js" type="text/javascript"></script>
  • load one or more of the available brushes (syntax highlighting scripts):

    <script src="css/shBrushJScript.js" type="text/javascript"></script>
  • include at least one of the available theme stylesheets:

    <link href="css/shCore.css" rel="stylesheet" type="text/css" />
    <link href="css/shThemeDefault.css" rel="stylesheet" type="text/css"/> 
  • wrap the code to highlight in a <pre> block and specify the brush that should be used:

                 <pre class="brush: js">function foo(){ ... }</pre>
               
  • finally, execute the highlighter using the following:

                 <script type="text/javascript">
                   SyntaxHighlighter.all();
                 </script>
               

The main issue with this way of configuring SyntaxHighlighter is that you’ll either have to load exactly those brushes you need or load them all even if they’re not needed. This problem remained until version 3 introduced the Autoloader script.

You can find all the details about its usage on that page, but basically you have to load the autoloader script and tell it about all the brushes aliases that it should recognize along with the path to the corresponding scripts. Much better, yay!

One problem I’ve had with the autoloader script is that it doesn’t seem to appreciate URIs for the location of the brush scripts when using the default syntax:

Array: [ 'alias1 alias2 /full/path/to/brush.js', ... ]

Although, it works perfectly fine when using the alternate syntax (which is alas a little bit more verbose):

Array: [ [ 'alias1', 'alias2', '/full/path/to/brush.js' ], ... ]

Finally, I’ve come up with the following script, largely inspired by the example on that page but using the alternate syntax:

	<script type="text/javascript"> 
		var baseSyntaxHighlighterScriptsPath = "http://base-path-to-scripts-folder/";
		function getSyntaxHighlighterScriptPath(name){
			return name.replace('@', baseSyntaxHighlighterScriptsPath);
		}
		
		SyntaxHighlighter.autoloader(
			[ 'applescript', getSyntaxHighlighterScriptPath('@shBrushAppleScript.js') ],
			[ 'actionscript3', 'as3', getSyntaxHighlighterScriptPath('@shBrushAS3.js') ],
			[ 'bash', 'shell', getSyntaxHighlighterScriptPath('@shBrushBash.js') ],
			[ 'coldfusion', 'cf', getSyntaxHighlighterScriptPath('@shBrushColdFusion.js') ],
			[ 'cpp', 'c', getSyntaxHighlighterScriptPath('@shBrushCpp.js') ],
			[ 'c#', 'c-sharp', 'csharp', getSyntaxHighlighterScriptPath('@shBrushCSharp.js') ],
			[ 'css', getSyntaxHighlighterScriptPath('@shBrushCss.js') ],
			[ 'diff', 'patch', 'pas', getSyntaxHighlighterScriptPath('@shBrushDiff.js') ],
			[ 'erl', 'erlang', getSyntaxHighlighterScriptPath('@shBrushErlang.js') ],
			[ 'groovy', getSyntaxHighlighterScriptPath('@shBrushGroovy.js') ],
			[ 'java', getSyntaxHighlighterScriptPath('@shBrushJava.js') ],
			[ 'jfx', 'javafx', getSyntaxHighlighterScriptPath('@shBrushJavaFX.js') ],
			[ 'js', 'javascript', 'jscript', getSyntaxHighlighterScriptPath('@shBrushJScript.js') ],
			[ 'perl', 'pl', getSyntaxHighlighterScriptPath('@shBrushPerl.js') ],
			[ 'php', getSyntaxHighlighterScriptPath('@shBrushPhp.js') ],
			[ 'text', 'plain', getSyntaxHighlighterScriptPath('@shBrushPlain.js') ],
			[ 'py', 'python', getSyntaxHighlighterScriptPath('@shBrushPython.js') ],
			[ 'ruby', 'rails', 'ror', 'rb', getSyntaxHighlighterScriptPath('@shBrushRuby.js') ],
			[ 'sass', 'scss', getSyntaxHighlighterScriptPath('@shBrushSass.js') ],
			[ 'scala', getSyntaxHighlighterScriptPath('@shBrushScala.js') ],
			[ 'sql', getSyntaxHighlighterScriptPath('@shBrushSql.js') ],
			[ 'vb', 'vbnet', getSyntaxHighlighterScriptPath('@shBrushVb.js') ],
			[ 'xml', 'xslt', 'html', 'htm', getSyntaxHighlighterScriptPath('@shBrushXml.js') ]
		);
		
		SyntaxHighlighter.all();
	</script>