So, there is a trend to build static type checker for popular managed dynamic languages?
Facebook also built Flow, a type checker for Javascript [1].
Facebook also built Flow, a type checker for Javascript [1].
It's a language that let you start small, and progress a lot with the complexity of your code. You can be productive in Python in 3 days if you know another language. But you can still learn new Python useful things 10 years after you started.
The progress curve is very sane.
And in the same way, your project may start small, then you add docstrings, classes, modules, packaging, unittests, infrastructure... And at some point, you may want types.
I use types on maybe 10% of my code in Python. It's great it's not mandatory. And it's great it's here when I benefit from it.
Without types, if I have the code `a = foo(x=1)` then I have to hunt down the source file for `foo()`, which likely just returns `bar(x)`, so I have to hunt down the source file for `bar()` to figure out what the hell its return type is, and so on and so forth. With types, I just look at the type signature for `foo()` and I'm good to go (and again, editor integration means that I don't even need to look up `foo()` at all!).
YMMV.
It depends of the project. If you write a lot of flask/django code, writing the types is not that worth it except for a few functions/methods.
> Without types, if I have the code `a = foo(x=1)` then I have to hunt down the source file for `foo()`
No, you just hover the function and get the help() out of it in most framework and libs. Again, if your code is mainly using a well define, documented and popular framework/lib/api, that's not a big deal. And it certainly doesn't require YOU to add types.
Or if you write a program that is contained in 1 to 5 files top. Not use for types.
Or if you are writing your program in jupyter.
And you most likely copy a snipet from the doc anyway. After all, if you see that the function you want to use return a AbstractTranscientVectorServiceFactory object, you can't do much with the information without the doc anyway.
But let's be real, most functions in Python are named pretty explicitly, and return things that you expect like "iterable of numbers" or "file like object yielding utf8".
Types are particularly useful in the cases if you are in a big project with a lot of custom code or in a domain either very complex or that you don't master very well. They are a good for of safety net and documentation at the same time.
But they come at a cost and it's a good thing to be able to choose.
There are no editors that can reliably deliver type information. The ones that come closest (PyCharm and YouCompleteMe) are otherwise very poor editors and resource hogs (the former) or nearly impossible to set up correctly (the latter). And neither is of any use when I'm reviewing someone's code on Github or debugging over ssh.
> Or if you write a program that is contained in 1 to 5 files top. Not use for types.
Types are definitely _useful_ for small programs, but you're right that they're not _necessary_ to the degree that they are for larger programs.
> And you most likely copy a snipet from the doc anyway.
This... is not the kind of programming I do. Although it may explain a lot about our difference of opinion on the topic.
> But let's be real, most functions in Python are named pretty explicitly, and return things that you expect like "iterable of numbers" or "file like object yielding utf8".
Ah yes, the get_values_as_iterable_of_numbers() function. :) Let's be real, virtually _no_ functions are named this way, and even if they were, the typing syntax is surely a better way of facilitating this information. More importantly, it still doesn't tell us anything about the types or number of the arguments. Lots of popular Python libraries (Pandas, Matplotlib) take ridiculous combinations of arguments (two strings and an int sometimes, but other times its an int and a float and a file handle, and if it's called on a Tuesday in April then you can omit all int args). If they had to appease a type checker, these libraries would certainly be much more usable.
> But they come at a cost and it's a good thing to be able to choose.
There are tradeoffs in many areas of programming, but typing is a clear win. Specifically, what is the cost you mention? The type checker won't let your type documentation grow stale? Or perhaps it makes it hard to write clumsy signatures like those in Pandas and Matplotlib?
If the help or comments tell you the types of arguments and return values, that is just static typing in the form of comments. Even more verbose than language level type annotations, yet much harder to parse for editors/IDEs/linters.
I program Python for 24 years now. It is my preferred language, but just recently I could go back to develop in it. It is impressive how much new idioms there are to learn.
we love types. They help us ship stuff faster.
I'm rooting for Reason, but it has a few nontrivial hills to climb before it's practical.
Moreover I think that at the time there were no statically typed languages that targeted web.
Well, it depends on the problem! If you’re working with a lot of data structures, or with bytes and serialization, compiled system languages with static typing are going to be productivity boosts over python. Languages aren’t everything, but there are definitely poor language/problem space fits.
Python has fantastic, expressive and productive libraries for those kind of problems. Unless you have some performance issue, I can't see how you are going to be faster than that.
The point of type-checking is that the function needs to work on a duck, then there is no point to ever pass it anything except a duck. In fact,the compiler should not even let you pass it something that's not a duck, because it's so pointless. That's literally the start and end of static typing, and if that's "rigorous" then yes, the whole point of static typing is to introduce this very basic level of rigor into your codebase. Because it's not going to work regardless of whether it passes the compiler.
Pretending interfaces do not exist does not actually make them go away. There is an interface there whether you explicitly enumerate it or not... even duck typing will fail if you try to call duck functions on something that is not a duck. Dynamic typing is not magic, it's the equivalent of passing everything around as Object or String in a static language. And that's an anti-pattern.
http://wiki.c2.com/?StringlyTyped
If you just want something to compile, you can pass in null-values of the appropriate type.
Now: there is a valid complaint that Java in particular really embraces the architecture-astronaut philosophy where everything is an overly-abstracted AbstractSingletonProxyFactoryBean (a convenient superclass for FactoryBean types that produce singleton-scoped proxy objects!). But usually it's fairly simple to wall that badness off from your actual business logic.
I don't always need to check types. I don't need to know that isReady returns a bool. It clearly returns a bool. This idea that type checking has to be absolute is because in statically typed languages it does, not because that's a forgone conclusion.
I'm sorry but you're mistaken, friend. I checked the signature of that function and it actually returns an enumeration of ready states:
enum class Ready { Insert, Process, Clean };
const Ready isReady();
Only if you completely ignore convention (that is prefixes return bools) then, sure, it's not useful. But if you just want to break conventions, then all of this is moot because you are guaranteed to do stupid stuff outside of that.
It won't help if all the arguments have the same type unfortunally but at least it catches something. Just like unit tests, they won't catch everything so the more checks we have and the earlier we run them the less risk there is something slips through to production.
If I know I'm only going to pass ducks, and the software is quick and dirty and non-critical enough that it doesn't matter if I accidentally pass it a goose, then I don't want to have a general contractor standing next to me developing saying "make sure that's a duck!"
There is no situation that is so "quick and dirty and non-critical" that you would want an invalid function invocation that could never, ever possibly succeed.
Again, if you just want a dummy call while you're refactoring so that things compile, pass in nulls in those parameters. But there is literally no reason to ever pass totally invalid but real data to a function.
I'm literally at a loss for any situation where you would ever pass an apple to a function that expects a duck, and think that's not just a valid construct, but a positive one that a language should encourage. It just boggles the mind how stupid that is.
Guess people just hate nulls that much, even as a parameter to a dummy stub, that they're willing to give away type safety.
If you really, really, really want to do a construct like that in a typed language, you can always just do "myFunc((Duck) myApple)" to pass the apple as an instance of type Duck, but if a duck is not an apple then that's guaranteed to fail at runtime, just like in the dynamically typed language (because dynamic typing is not magic).
If am writing something quickly, and it's a small project or test, and I might be testing out whether I want to pass a single value or a list in my workflow, I don't want the rails on constantly. I don't want to guarantee safety by ctrl+H-ing my "str" to "Iterable" constantly. Move quickly, break things, get the product out. At least, when the product is a quick and dirty tool and the options are "spend 30 minutes to get it built" or "don't build it".
You make an excellent point here that I don't see articulated very often. The Landscape of Popular Languages, let's say C/C++/Java/Python/Ruby/JavaScript, has a big gap in the middle. You have good choices between:
1. lower level "systems" oriented languages with static types, usually compiled, like C++. You get lots of flexibility and direct access to primitives. Static types help wrangle big codebases. Generally suited to large projects.
2. higher level "scripting" oriented languages with dynamic types, usually interpreted, like Python. Writing code for most tasks is easier. You give up some stuff you'd want for projects like operating systems or databases. Most projects aren't operating systems or databases, so that's usually a good tradeoff. Generally suited to small projects.
The problem is that lots of projects are medium-ish. You set out to build your web service backend or whatever, it would be a pain in the ass to write in C++, so you use Python. Getting it working is quick and easy. A few months later it's big, complex piece of software and working on it in Python is a pain in the ass. You can't win. What you really wanted was a language that's "easy to write" like Python, but with static or optional types, maybe better thread handling, and at this point the interpreter isn't doing much for you so it might as well be compiled. There are tons of cases where you just want a "better C" or "Python but faster and with static types", and for the longest time the Landscape of Popular Languages just had a giant hole there.
We needed that space filled and Go delivered. I'm usually very critical of Go, but I can't hate on it for being the wrong kind of language. It's definitely the right kind of language for these "goldilocks" problems that aren't too high or too low level, too big or too small. Part of being a good programmer is understanding that languages are tools, and you need to pick one that fits your problem. Go deserves all the success and praise it's gotten for being a language that fits actual problems.
Another thing to keep in mind is that type annotations can be done incrementally. You can't take 1000 Python files and port them to Go one by one, but you can add types a little bit at a time with a completely usable codebase at every intermediate step.
- ruby
- python
- Javascript
I was on a quest for a good, well supported, statically typed scripting language for quite a while, and in the end I defaulted to python.
For me, I was looking for a language that is appropriate for both scripting and slightly larger web application projects.
Scripting / Web application development is NOT a great use case for C/++, and I feel Java is a bit heavy handed to be used for a scripting language.
I tried out Golang but did not end up liking the language.
That being said, while I started out using Flow, TypeScript just has way more community adoption and better tooling. So at this point, I usually recommend TS over Flow to most people. But using either of them is way better than writing just regular JS code.
That being said, Flow is better then no typechecks, and it was half a year ago that I looked, stuff might have improved.
On the other hand, I don't agree with some of the comments here. For example, Flow is not terminal only at all, I never use the terminal to run Flow. The editor integration is totally fine, especially in Atom and VSCode.
# Totoal #
Flow: 2081 open, 2748 closed
Typescript: 2541 open, 14421 closed
# Between 4/11 and 5/11 #
Flow: 88 closed, 93 new
Typescript: 448 closed, 173 new
Note: We are using flow but we are not happy that Flow is not up to speed with answering questions or addressing concerns.