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!
Top comments (106)
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.
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.
Glad to hear that!
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.
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 :).
Thanks for the reply. I'm glad to hear it's a priority and that you're being careful about the language :+thumbsup:
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?
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 are the most practical applications of crystal at the moment? We're a Ruby/Rails shop that could probably jump in and be comfortable using it quickly, but I don't immediately have a practical use case in mind. What's a good place to start making it a worthwhile environment to jump in to?
I've been using Crystal for microservices/REST APIs in the last two months. It actually fits pretty well with those kinds of applications. The code ended up smaller and more well structured than the Node.js version, and it actually runs 2x-5x faster, depending on the endpoint complexity.
AFAIK there are also some folks working on more complex stuff such as machine-learning and game engines using Crystal, so its probably a good contender for those tasks too.
You can use Kemal for building APIs and microservices. It's simple and fast. If you've ever used Sinatra before you'll feel right at home :)
P.S: I'm the author of Kemal.
Kemal is awesome! Makes creating services very easy! :) Kudos good sir!
Cool, that could definitely be interesting.
Compilers and command line tools are for sure good fits. But we have also create bots (slack / twilio / telegram), some other microservices and small web apps with some UI.
Comparing to Rails, there is for sure a lot things to cover. ORM and a smooth assets management would be my top 2 things that I would like to see solved / documented and avoid reinventing the wheel in order to start participating more in the web apps market.
Depending on your interest crystal could be good enough for game dev, using c libs in a more friendly way, and data science.
Could you tell me more about your community, your resources for it look incredibly robust. How did it grow in the many branches it is now ( reddit, chatroom, repo, etc) and how do you manage it? Also in hindsight how would you suggest a newbie go about creating a language and growing a community around a new lang go about it?
Hey community, if you're reading this:
Honestly, it just sort of happened. I think we never imagined Crystal would make it this far. It started as an experiment to see what compiled Ruby would look like. The next thing we knew there were a few people crazy enough to try it and report issues. One day someone posted it to Reddit and people got interested. Step by step.
From time to time someone from the community says "hey, it'd be cool to have X, would you mind if I do it?" and we're like "Yes, go for it!". Then if the thing picks up steam we link it from the official site. I think that happened with reddit, the chatroom, gitter, etc.
For example, recently a group of crystallers joined forces and created github.com/crystal-lang-tools to tackle all editor support projects under the same umbrella. They let us know and we transferred our Sublime plugin. That's great because we currently have more on our plates than we can manage!
Thanks for sharing the organic nature of this sounds super fun. Would you say that Crystal is beginner friendly regarding Open Source Contributions?
It definitely is! We are a small yet really friendly community which is always open for everyone.
Please feel free to join our Gitter channel gitter.im/crystal-lang/crystal :)
Thanks Serdar, will do. I've been looking for a friendly Open Source community to get involved with.
"a few people crazy enough" that makes big difference :)
selamlar abi , abi senin haberin var mi bunlar ne zaman windows ichin bir shey'ler yapacaklar ? ya ELIXIR teamiyle koshushtuk bize buyuk destek oldular ve baazi ishlerimizi elixir'le yaptik amma bunlara ne kadar email falan atiyorum konushuyorum hich bir shey soylemiyorlar ve belli ki 2025 ancak bunlar windows ichin bir sheyler verir ! windows 10 da kullana bilmiyoruz ve bu beni childiriyor ! benim bildighim kadar sen KEMAL'i yazansin sende luft et bunlardan bir konush sor ne zaman bunlar windows ichin bir shey yapacak ?
I don't think a single person could handle all the community channels and communications.
There are some of us (in the whole community) that are more skilled and eager to share, curate and prepare content.
For a management perspective he have some hooks into Slack to avoid constant querying all the channels. But that is the smallest part.
First we have a website, some docs people and read, then we added the mailing list as a way to users reach us and start conversating (besides from github). But then IRC emerged and @jhass created a service to cross post to Gitter. A Slack was also created, but IRC+Gitter worked better for here.
Some people like to spend and collaborate there a lot in the chat <3
I think that constant conversation & sharing in that channel is important to a community. Yet, is not always easy. Serdar and Matias do great job in the community nowadays #Kudos.
An online playground and a share code service (even a basic one) is agreat resource.
As with any open source project there is a challenge (and joy) in trusting people with who you will be able to delegate and even share resposibilities.
Tips: Start little by little. Do something is useful for you. Be patient.
Thank you Brian :) I'm just trying my best to share how awesome Crystal is with everyone!
Thanks that's great advice!
Some things that bugs me a little bit in the language are:
/should always be float division (instead of depending on the receiver/argument types).So I am pretty happy with the language :-)
From the ecosystem, there are some stories to be improved in dependency management and development tools.
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.
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.
There's things I'd like to contribute to the language, but some are some serious undertaking. What's the best way to know what features are more likely to be merged without having to spend weeks implementing something that could potentially just be rejected?
The best you can do is propose the feature or enhancement and allow some time for discussions and points of view to mature. Then we have an "status:accepted" label that we use to signal that if you tackle this as it's been discussed, we'll merge it.
Ah, ok. That makes sense. So make an issue with a proposed feature, wait for it to be accepted, then tackle a PR. I think this would be a great flow for everyone to follow. Thanks!
I think we should document this flow in official Crystal wiki :)
If it's a reported bug/enhancement, I would say to first discuss in the issue how it could be solved.
It might seems like an extra work other than coding, but is a way to avoid doing all the coding work that might not be merged.
We do value the efforts off everybody but sometimes is more that just writing code what is needed. Making decisions sometimes requires other kind of mindset, attention and time.