This makes me happy! Phoenix and RoR before it have proven to be incredibly productive but the issues with types and runtime errors IMHO made it easy to break stuff.
Types should pass a whole range of runtime errors to build time error, which is a huge plus for reliable software!
Phoenix relies heavily on Elixir macros, so you'd first have to compile and then try to call the generated functions... But I'm not sure that makes a whole lot of sense as you'd be missing out of most of the benefits of Phoenix.
A more pragmatic approach is to use Elixir for the Phoenix specific parts, and from there call the other Erlang based languages for your business logic.
A lot of what Phoenix has to offer comes from Elixir language features and the Mix build tool. To the extent it's possible I can't imagine it would be pleasant or useful.
Yes, but it isn't exactly ergonomic. Elixir tries to namespace most of its stuff within an Elixir namespace, so you have to use that when calling functions out of it.
All the Elixir macro magic, however, will not carry forwards, so you'll have to be using the compiled Phoenix bytecode
I find this this kind of prejudice towards technologies rather reductionist and limiting, to say the least. During interviews I conduct, broad generalizations towards any technology, be it Visual Basic or Rust, for example, are noted as red flags. My rationale is that engineers should be able to understand that tooling is a matter of context + tradeoffs.
There are lessons to be learned in almost any technology. Even more so in battle-tested technologies. And PHP is still a very popular tool that has come a long way over almost three decades since its creation (mid 1994).
Things that come to mind that are worth looking for inspiration in PHP:
- package management (composer is great)
- gradual typing (9 years of use-cases to learn from)
- framework popularity (what makes Laravel so popular?)
- staying power (why is it still so popular after 3 decades?)
- governance and community development model (continuous improvements since inception and relatively little drama)
- backwards compatibility (major changes of internals with very little compatibility breaks. PHP's engine had major core changes while still maintaining friendly levels of backwards compatibility. And JIT is on the way.)
PHP is indeed a good showcase how much improvement can be done, even if the foundation is flawed. In fact, limiting the scope to "how to introduce typing", PHP has done almost everything right. It even rejects nulls by default and I would have wished Python had been just as rigorous in this regard.
Could you please stop posting unsubstantive comments and flamebait? You've unfortunately been doing it repeatedly. It's not what this site is for, and destroys what it is for.
Elixir, and Erlang before it, has always essentially had gradual typing through pattern matching and non-overloaded operators, they just get caught at runtime. This is bringing those checks to compile time.
Very nice. Lack of types in Elixir was off-putting to some people. And it's perfect time to add them, since language is mostly complete. In Ruby, I was not sold on sorbet, because it goes "against philosophy of dynamic language", but I like it as an extra layer of safety in some places (there are rough edges tho) and I'm pretty confident they will get typing right
Very happy about this, switched companies recently and Elixir was the language of choice. Coming from TypeScript, I missed a lot of the benefits of a typed language and having access to the data model from anywhere in the project. Elixir was already great, but this is a huge improvement, congrats to the team that worked on this!
Jose, if you come across this, I just want to express my gratitude to you for Elixir and for everything you've done for the Elixir community. The thoughtfulness and adeptness you bring to the language and the community is so very much appreciated. This change is such an excellent example of the steadiness and diligence you have led with. So rarely does a single individual that I am not interpersonally close to have such an outsized impact on my day-to-day happiness. Thank you, Jose.
Your work has raised the level of projects solo and very small dev teams can build.
Elixir has been one of the nicest things I've discovered since becoming a dev back in 2010. I hope this new feature can improve the experience for larger orgs while keeping it great for indie hackers and new learners!
I agree 100%. I have been using the language professionally everyday for a few years now, having converted our Erlang system to Elixir and Phoenix. The language is a joy to use, the community is great and José is doing a fantastic job in a kind way. I love being part of that community. Thank you for everything.
Elixir is already one of the most static of dynamic langages so I'm not too surprised about the quick progress, the language is almost built for that already.
It's much easier compared to typing Ruby or JavaScript in terms of difficulty.
Exactly! The hard part was to find a type system that we can map most (if not all) Elixir constructs to, and set-theoretic types have been a great fit so far and we were able to develop the theory in the areas where it lacked.
Didn't know nitter. It looks like it allows to search twitter without having an account which is cool since I deleted mine a few months ago because I could not find relevant information anymore (the browsing experience was such a PITA).
It is not a goal of the type system at the moment, as we will focus on software correctness and developer experience.
However, keep in mind the Erlang VM already uses typing information (from patterns, guards, and built-in functions) to optimize code through its JIT compiler.
The examples will be very gradual - no pun intended - as we need to go through each Elixir construct and either add inference (if a pattern or a guard) or type checking rules. Right now we infer and type check one single construct (bitstring) but the foundation is there to build on.
Valim seems to do a really great job stewarding Elixir initiatives with the help of experts, like his collaboration with Sean Moriarty (author of Genetic Algorithms in Elixir) leading to Nx and now his recruiting Dr. Castagna (a PL researcher) and Guillaume Duboc (PhD candidate) to implement gradual typing.
Hejlsberg is an interesting example. Did he do the original typescript type system that was unsound or was that someone else and he cleaned up the mess?
TypeScript's type system is, and it will always be, unsound by design [1].
Working with these constraints and still being able to deliver what they did is nothing short of incredible. They had the maturity and experience to ship something that wasn't Yet Another Copy Of Java's Type System, like Sorbet and Dart, for example.
Yes, AFAIK, Hejlsberg did the original TS type system. He, from what I can tell, wanted to build a usable type system for JS than any theoretical soundness. He seems to have succeeded in that.
Why, exactly? He has no proven experience implementing one as far as I know. I don't think he's a particularly worse choice than anyone else but I don't understand why he'd be a particularly good one.
This is exactly the kind of efforts I wished were built into Ruby. No wonder lots of Ruby devs are now doing Elixir. It's shaping up to be an amazing language!
You need to see the talks of the elixir team and Jose especially about bringing types to elixir. They are being so careful and thoughtful to bring a good type system or none at all.
If I understand the comment you’re replying to, the loss is the competitive advantage.
I.e., if a desire for static types was keeping a large number of devs away, then this development will lead to far more devs using Elixir. This would erode some of throwaway_64000’s competitive edge as those devs also got the productivity gains of using Elixir.
Will gradual typing allow Elixir to implement any new performance optimizations? Adding typing to dynamic languages like Elixir (or Python or Ruby) means developers pay a compile-time cost but still pay a runtime cost for the VM's dynamic type checking.
Note that there are already some of this stuff in the compiler for AOT. Using these new specs for AOT optimisations is going to be a far taller problem than catching some of the errors
I like gradually typed languages but there seems to be no information here. Just a twitter screenshot. I can find a paper from early 2022 https://www.sciencedirect.com/science/article/pii/S259011842... - is there more information about this type system available?
Hmm I’ve started doing some elixir as a side project and coming from typescript honestly I’ve found it a bit “dumb” and slow at the beginning. It didn’t allow you to write precise enough types, and would complain a lot about weird things.
As the project became more complex though, and I learned more about elixir, dialyzer became progressively more useful.
Most of the time when I chase a weird type error in TypeScript it turned out that the types were not correct enough and had to be fixed.
When I’m chasing weird dialyzer complaints, most of the time I discover that it was actually a bug in my code. Honestly sometimes I’m astonished how it can reason about very complex things so well. It’s just the error messages are usually very confusing, but with experience that kinda solves itself.
The difference to me so far has been that dialyzer finds bugs in my code, rather than in my types, which is quite nice.
If they make the built-in type system of elixir faster and with better error messages, I’ll be a very happy camper.
I'm reading chapters in Coders At Work and thinking about who should be interviewed in a 2nd edition. Jose Valim should be at the top of this kind of list. His work on Elixir has been as profound as anyone in Coders.
59 comments
[ 2.9 ms ] story [ 120 ms ] threadTypes should pass a whole range of runtime errors to build time error, which is a huge plus for reliable software!
[0] https://lfe.io/
A more pragmatic approach is to use Elixir for the Phoenix specific parts, and from there call the other Erlang based languages for your business logic.
All the Elixir macro magic, however, will not carry forwards, so you'll have to be using the compiled Phoenix bytecode
> Writing Phoenix controllers, templates, plugs, and live views from Erlang
There are lessons to be learned in almost any technology. Even more so in battle-tested technologies. And PHP is still a very popular tool that has come a long way over almost three decades since its creation (mid 1994).
Things that come to mind that are worth looking for inspiration in PHP:
- package management (composer is great)
- gradual typing (9 years of use-cases to learn from)
- framework popularity (what makes Laravel so popular?)
- staying power (why is it still so popular after 3 decades?)
- governance and community development model (continuous improvements since inception and relatively little drama)
- backwards compatibility (major changes of internals with very little compatibility breaks. PHP's engine had major core changes while still maintaining friendly levels of backwards compatibility. And JIT is on the way.)
https://www.php.net/manual/en/history.php.php
If you wouldn't mind reviewing https://news.ycombinator.com/newsguidelines.html and taking the intended spirit of the site more to heart, we'd be grateful.
(Other examples:
https://news.ycombinator.com/item?id=38918417
https://news.ycombinator.com/item?id=38917440
https://news.ycombinator.com/item?id=38917178 )
nice! i'm jealous.
Elixir has been one of the nicest things I've discovered since becoming a dev back in 2010. I hope this new feature can improve the experience for larger orgs while keeping it great for indie hackers and new learners!
It's much easier compared to typing Ruby or JavaScript in terms of difficulty.
Incidentally, 'nitter.net' isn't the only deployment. See here for a long list of alternative instances: https://github.com/zedeus/nitter/wiki/Instances
Is so, what's a ballpark of perf improvement?
However, keep in mind the Erlang VM already uses typing information (from patterns, guards, and built-in functions) to optimize code through its JIT compiler.
Working with these constraints and still being able to deliver what they did is nothing short of incredible. They had the maturity and experience to ship something that wasn't Yet Another Copy Of Java's Type System, like Sorbet and Dart, for example.
[1] https://www.typescriptlang.org/docs/handbook/type-compatibil...
Yes, AFAIK, Hejlsberg did the original TS type system. He, from what I can tell, wanted to build a usable type system for JS than any theoretical soundness. He seems to have succeeded in that.
This will make it my default choice the next time I need to build a web app or service.
I feel like an anti-evangelist. I actually like reading complaints about the language...that just means there is less competition.
- K8s/microservices are better/standard.
- Elixir is too slow.
- Dynamically typed languages suck.
- There aren't enough libraries/ecosystem. And a lot of the libraries look like they are dead.
- No big companies use it for anything.
- It needs a better auth library, something like devise for ruby.
- It is a functional language, and everyone knows they are too hard to use.
- It is too difficult to hire good people.
But now, one of the standard complaints is gone. :(
This means invariably more people finding out that you can build incredible products with tiny teams. :(
Thank you Jose for your leadership and building an incredible community and ecosystem. Hopefully, as the language grows, this stays the same.
This announcement says it is detecting pattern matching and guards at build time instead of runtime, so nothing is changing.
And the "upcoming" (still in investigation) new types are totally optional. Only enabled if you use fat arrows.
You'll still have the advantages of having the core library and third party libs typed in your IDE, even if you don't use types.
I.e., if a desire for static types was keeping a large number of devs away, then this development will lead to far more devs using Elixir. This would erode some of throwaway_64000’s competitive edge as those devs also got the productivity gains of using Elixir.
Jose's elixirconf keynote this year did a great job providing details on how how this new type system works:
https://youtu.be/giYbq4HmfGA?si=ZYf1g_GXr4LmHQ3H
I used Elixir for a period and loved the language but I got fed up with using Dialyzer so I switched to other typed languages.
When this is done, I'll be coming back as I think Elixir does pretty much everything else better when it comes to writing web apps vs other languages.
As the project became more complex though, and I learned more about elixir, dialyzer became progressively more useful.
Most of the time when I chase a weird type error in TypeScript it turned out that the types were not correct enough and had to be fixed.
When I’m chasing weird dialyzer complaints, most of the time I discover that it was actually a bug in my code. Honestly sometimes I’m astonished how it can reason about very complex things so well. It’s just the error messages are usually very confusing, but with experience that kinda solves itself.
The difference to me so far has been that dialyzer finds bugs in my code, rather than in my types, which is quite nice.
If they make the built-in type system of elixir faster and with better error messages, I’ll be a very happy camper.