Hi there!
I'm Matias Garcia Isaia, and together with Brian Cardiff and Martin Verzilli, we're part of the Crystal Team working at Manas.Tech.
Manas.Tech is a software company from Buenos Aires, Argentina, that develops unconventional tech projects for people around the world. And, after being involved in Crystal's development for a while, I can assure you there are not that many conventional things in developing a programming language!
Crystal is a programming language that aims to be friendly for both humans and computers alike - make developers enjoy writing code, and make code run as efficiently as it can. Statically typed, compiled language with a really heavy type inference to make it feel as scripting - the best of both worlds. And did I say it's open source?
Its first commit has just turned 5, and it has changed from an individual's hobby back then into an amazing language supported by Manas.Tech and an amazing community. To celebrate that, we are doing this AMA. Starting today, at 2PM EST, we'll do our best to share with you everything we know about the project, life, the universe, and everything!
So - don't be shy, ask us anything!
Oldest comments (106)
I imagine adoption isn't just about the technical details and Crystal has a really slick presence. How do you approach the branding/marketing elements of the whole thing? Like, who's job is this, how often do you meet about these kinds of things, etc.
Apart from Crystal, at Manas.Tech we've been working on end-to-end projects for the last ~15 years. So we have some Graphics and Interaction designers we're very proud of in the team :). They designed the logo, fonts, branding, website and, yes, they programmed the famous spinning logo from scratch ;).
And they did a spectacular job
What's the relationship between Crystal and Manas.Tech?
Manas.Tech is the company where Crystal was born and one of the main contributors. We are leading the core team and trying to build a community and supporter network strong enough to carry this effort forward.
Could you give us some insights, visions and facts of the road of Crystal towards v 1.0 ? Such as:
This is not to be critical, we want Crystal to succeed.
We have spent some time putting together an estimate for what’s pending for a 1.0 release that we could be proud of, an it comes to about 1.400 additional hours. At a pace of 2x the monthly hours we are currently covering with donations (~54hs) that means another extra 14 months which is way longer than what we want. We are trying to rally some other sponsors to shorten that time; it's looking positive for now, but we need a few more weeks to provide another update on that.
Parallelization is one of the main efforts for 1.0. We managed to run the multi-threading sieve of Eratosthenes sample with reasonable results on Linux and poor results in OS X.
Windows port is advancing thanks to the extended community, can't comment much else on that.
Glad to hear that!
I wouldn't want Windows support to block 1.0! I think it would be important to get feature complete with 1.0, which, in turn, will expand the interest and base for Windows support.
Agreed. It's not blocking 1.0.
What was the biggest obstacle you had to overcome when creating Crystal?
When creating a new language you have tons of things to think about.
Knowing and being able to focusing on the right ones is the hardest I would say.
From "this feature would be so cool, I would love to use it" to "fix the stylesheet here or there" you have the whole spectrum of ideas, tasks and research. And sometimes the right ones are not the more entretaining ones :-)
A language is easier to toy around at the beginner and when the time goes by the decisions are sometimes harder because consensus is somewhat needed, compatibility is expected, etc.
Brian will post a technical perspective on this, but I wanted to offer a non-technical one. My perception of the team's biggest stumbling situations had to do with our own natural self-criticism. The classic mentality when tackling these sort of problems is "if this could be done, someone would have already done it". The easiest way to disarm those fears is to approach the problem with a humble but curious frame and to just follow the path to see where it leads.
Has crystal approached any larger companies for support / backing ?
Yes, Manas.Tech, as host of Crystal, is approaching a few companies that are interested in seeing a 1.0 release for support. It's part of our plan to speed up that timeline.
Feel free to share the word if you know of any other one that could be interested in supporting!
In the shortest steps possible, how does one go about creating a new programming language?
That's at least for a compiled language. Steps 4 and 5 would be replaced with an interpreter.
There are plenty of tutorials on the web on how to write you own language. What I always like to say is that a language is just a tool like "ls", "cat", or any application you develop: you take some input, analyze it and produce some output.
The tricky parts are:
That's at least the technical part of it. The community part is a whole world of its own, tackling issues and PRs, and I'd say it's even more challenging (but also rewarding!) than developing the language itself :-)
Wow, one of the best answers I've ever heard about creating a programming language :) Thanks Ary!
Yeah, loved it!
In the Manas.Tech team there are around 5 or 6 devs pretty involved, but their dedication fluctuates. We try to spend at least 2 or 3 times the amount of hours funded by through Bountysource, but some months we accumulate budget in order to have longer sprints. On top of that there are 3 other core team members outside Manas.Tech that spend time depending on their personal circumstances. The same goes for the larger community.
The main non-technical challenge right now is finding a sustainable way to fund (at least partially) the effort to complete 1.0. You can help by spreading the word and/or chipping in.
The main technical goals for 1.0 are multi-threading, better debugging tools, modules and generics.
How many people are there in the core team? What does the core team think about the biggest hurdles before 1.0?
dev.to/crystal-lang/we-created-the...
Did the name of the language come from "gee, what's a gemstone that starts with the letter C?"
It was more "gee, what's a gemstone that would make for a good file extension?" cr sounded kind of nice and... we don't like to bike shed on this sorts of things too much :P, so it ended up being Crystal.
Haha nice! Totally a fan of good file extensions.
How important is compilation time to the team? My biggest worry is that speeds will become unbearably slow as a project gets bigger and bigger. What are your thoughts on this. Is it possible to make things faster, maybe via incremental compilation?
I love using Crystal. Thanks for all the hard work!
It is very important, and one of the reasons we slowed down development. Our main concern is to get to a point where we know whether we'll have to make any breaking changes to the language to keep compilation times at a satisfactory level. That question is quite hard to answer without a formal specification of the language, but then again formal specification of the language is by itself a huge endeavor.
The good news is we haven't explored the solution space, we're confident that we can make it.
About incremental compilation, I'm not sure to what extent that will be possible without some "big change" to the language. Right now the "type inference magic" of Crystal depends on knowing all the source code ahead of time, and that makes it difficult to identify proper compilation units.
Thanks for the reply. I'm glad to hear it's a priority and that you're being careful about the language :+thumbsup:
You could look at what Rust is doing with incremental compilation. They do type inference in a similar way (though less insane) and they've been working on getting incremental compilation for a few years now.
We do a look a lot at Rust, not only regarding incremental compilation. We really like how they organized the community, for example. And given they also use LLVM, we sometimes can leverage some of their findings, tooling or advances, so if there's any member of the Rust team seeing this: thank you for being a source of inspiration :).
I wonder if requiring all public api (input params and return types of every public method) of a file to be typed would allow incremental compilation. In that case, you could add a switch to the compiler to warn you of any typing required to allow incremental compilation. That way, if you want to speed up compilation, you could add the required type. Is it feasible?
Go (GC), Rust (ownership model) and Swift (ARC) have different strategies for handling parallelism and memory management. Node and the Javascript community have been promoting immutability and functional programming.
Can you discuss Crystal's approach to solving this problem and some of the advantages/pitfalls that we may encounter using it?
I don't see the GC going away. It could change the which GC is used though.
Regarding parallelism, currently, something that could happen is to go something more like Ruby's Guild and do runtime checks. And it would be great to have some built in constructs for some well-known patterns.
Doing runtime would add some overhead to the operations. That is something to try to minimize.
As I see it Crystal's goals is the type safety, type inference and syntax. I am not confident enough to innovate with something like rust ownership model. I still lack the rust experience I would like to compare that.
As someone who just knows Ruby, how would you recommend learning Crystal? Is there some knowledge of C that would help a lot in order to learn Crystal?
As someone who just knows Ruby you're probably already an intermediate Crystal programmer :). You can actually leverage one of the defects of the official docs: they sometimes assume you know Ruby to some extent. So you should be good to go with: crystal-lang.org/docs/
Recently we published some introductory videos with the folks at Daily Drip: dailydrip.com/topics/crystal
And for more hardcore materials you can download the slides and code examples from the CodeCamp we hosted at San Francisco this past May: codecamp.crystal-lang.org/Crystal-...
As a Rubyist, you're good to go. I wrote a small book titled "Crystal for Rubyists" it's free and you can read it online :)
crystalforrubyists.com/book/index....
Rather than some knowledge of C the biggest initial gap would be to work with value types (stored in stack), using some mandatory type annotations and working with generics for containers like Arrays, Hashes, etc.
Any experience in a statically typed language could be useful: C, C#, Java.
But if you just want to have an experience learning Crystal itself it could help to fill the gaps in some areas to allow others with similar backgrounds.
Cause we haven't fixed that bug!
It's a statically-built site, but we don't rebuild it daily - so that post is from the day before the last build. Just a bug we have to fix.
In the PullRequests for the repo, there's a lot of PRs that have passed the travis build, and been reviewed and accepted. Is there anyone that goes through all of these PRs to either merge them, or reject them? Someone to do housekeeping on the PRs?
We try and review&merge the PRs whenever we can, but we have to decide between investing our energies on that versus developing other features or tasks for the project.
We love having all those PRs and community engagement, but it's our responsibility to prevent lots of little steps in the form of PRs from diverting the path we want the project to pursue.
So there are PRs hanging there because we didn't find time to merge them, but also because we want to have a more general view of the underlaying issues, and come up with a global solution to that - instead of adding a bunch of patches for special cases.
There's a great talk called Hammock Driven Development, by Clojure's Rich Hickey that explains this concept way better than me.
Re-reading the whole thing, if there's a PR with green CI + two core members' approvals, it should be merged. So, if you happen to find one that isn't - please ping us, because it's a mistake :)
Ok, that makes sense. I'll keep an eye out for any of those. Maybe that will help clean up some PRs.
What language/s do you consider the one to beat?
What language do you consider the most similar to Crystal?
Do you consider the current GC algorithm good enough for v.1.0 ?
Do you consider web dev as the bigger use case or do you have other scenarios like native app-dev, games, IoT, etc as main use cases?
Manas.Tech is full of language geeks, and as such we don't like to frame Crystal's growth in terms of beating X language. We prefer to take inspiration from them, and think there's room in the world for everyone. Do The Beatles beat The Rolling Stones or vice-versa? I don't want to pick, I want to listen to them both!
Crystal will be appropriate in some contexts, and its syntax and philosophy might resonate better with you than other languages in those cases, then you'll use it.
As far as similarities go, I see it being most comparable to Go, Ruby, Python in different aspects or contexts.
The current GC is out of the box, and does a surprisingly good job for now. We might need to tweak it or write one from scratch if we stumble upon any limitations while getting to support parallelism; we still don't know.
Excellent questions!
I don't think we want to beat other languages. We actually use other languages and love them all, like Ruby, Elixir and Go. With Crystal we try to provide an alternative that's fun to work with and performant when possible, but that of course depends on everyone's preferences (some like Ruby's syntax, others don't). We actually encourage others to combine Crystal with other languages.
It's hard to find other languages that are similar to Crystal. I'd say the syntax is very close to Ruby, the runtime maybe a mix of Ruby and Go (but just a bit), and since we have value types we are maybe similar to C#, D and Swift. And we have multiple dispatch, which Julia also has, so there's a similarity too. We use LLVM as a backend, which Rust and Julia do too. As you can see, we get inspiration from many, many languages.
I think the GC algorithm might be good enough for v1.0. Versions past 1.0 will be all about improving the performance and fixing potential bugs, while retaining backwards compatibility. For example Go improves their GC each time in every version, so maybe the one in 1.0 wasn't that good (if you compare it to the current one ;-)).
I'd personally love to see Crystal being used in other contexts than web apps. For example I'm a big fan of videogames, I play them every day, and I'm super happy when I see others creating games in Crystal (BlaXpirit is probably the one leading this effort).
I've some high scale APIs running for more than 1 year without any restarts and there's no memory leak at all. I can easily say that the current GC (Boehm GC) is pretty good enough.