As I mentioned previously, non-SSL sites are heading toward being marked as security issues by browsers. If you want SSL and
full control over your website, including a top-notch CDN, AWS provides the tools you need, and it’s not too hairy to get going.
Considering what a nightmare anything SSL usually is, AWS makes it as painless as can be, as long as you are willing to get your
feet wet with some infrastructure, and do a bit of manual scripting.
Since Chrome is marking non-SSL sites as insecure soon, I finally got around to moving my site’s hosting from GitHub Pages to AWS. Setting up a static website using CloudFront and S3 is surprisingly easy. I’ll write up how to do it in a future post, but for now, enjoy the sweet security of SSL :)
Python famously holds as a value “Explicit is better than implicit”. This is not
something Rubyists (or many functional programmers) value, instead favoring compactness or “elegance”. I believe senior
developers undervalue explicit code that uses fewer features than it could. Compact code that uses many language, library, and
framework features excludes people from understanding and contributing to it, and I believe this is a serious consideration when
writing code. What better judge of a codebase’s maintainability could there be than “fewest concepts needed to understand it”?
I’ve been traveling to San Francisco every two months (more or less) for the last five years as part of my job. The hardest part
used to be the traveling and staying in hotels (now it’s that I pack so much activity into such a small amount of time, I barely see any of my San Francisco friends any more). This is about how to make the travel and hotels suck less, and if you are currently
working remotely, or planning to, reducing travel anxiety and hassle should be something you actively address.
Engineers often deride anything that’s not writing code as being not “real” work. While this diminishes the contributions of
others, and is generally an unhealthy bravado, it is critical to understand the difference between activities that directly
solve a problem and those that organize said activities. The later, I call “meta-work” and engineers should be very conscious of
spending too much time on it.
I’ve added another “problem” to “What problem does it solve?” about serverless architectures, which I think are
the future of distributed systems (though that future isn’t here yet):
The less code you are responsible for running in production, the better. Running an actual server means it’s all on you. Using virtual servers delegates some to your hosting provider. Using a platform-as-a-service like Heroku delegates yet more. A serverless architecture reduces this to what might be the absolute minimum - you are responsible for your code, and everything else is handled by the provider.
There are three big challenges: deployment, vendor lock-in, and testing.
In this month’s issue of PragPub, I have an article about Rails, and how it’s still as relevant today as it was when
it came out:
What Rails taught us about agility in 2005 is still applicable today. Rails has also kept up with the times, baking in more conventions around common problems, but also adding more technology to help solve them.
And there is testing support for all of it, from unit tests to browser-based system tests, so you can confidently ship your entire app, fully tested, whenever you need to make a change.
A few people on Twitter were talking about developers that run their entire test suite only on their continuous integration (CI) server. The idea was that this was a sign of low quality tests, low quality code, or an otherwise bad process. In the past, I
felt this way, and “breaking the build” (by checking in code that had failing tests) was viewed as bad. I don’t believe this any
longer, and now feel like it’s a critical ability for a high-functioning team to have. Not that it isn’t important to identify low
quality tests or code, but what exactly is the value of having a fast test suite?
One of my biggest learnings creating Webpack from Nothing was just
systems that only do half of what you need and are extensible only through dense plugin mechanisms.
The last chapter is an attempt to describe this problem
in detail and offer some solutions.
I was surprised at how counter-intuitive each step ended up being—I can totally see why there are so many blog posts describing how to set it up. From a test runner that can’t run tests, to a system that supports presets, but includes none of them, I was faced with many choices along the way, but also faced with completely opaque systems whose behavior and failure modes were unpredictable.
It focuses on Webpack and Karma, which I found very difficult to make work together:
Webpack and Karma’s interoperability is completely opaque to the user. This means that you have no way to know how the tools are working or even observe them working together without debugging into the source code.
If you didn’t make it all the way through the entire Webpack site, this last chapter is worth a read.