Anders Hejlsberg Is Right: You Cannot Maintain Large Programs In JavaScript | Hacker News Reader

Anders Hejlsberg Is Right: You Cannot Maintain Large Programs In JavaScript

lostechies.com

23 pointsgspyrou30 comments

I've had this discussion with co-workers before about scripting languages, and in particular Python, as that is what I am working with currently. I feel that once Python -- or indeed any loosey-goosey language that lack strong types -- programs grow beyond a certain size you easily fall into the "inconsistency trap", especially if you have many programmers working on the same code base.

You have to be extra diligent with docstrings; unit tests and variable naming, or invent your own poor, ad hoc type checking with isinstance, or liberal use of exception handling if you insist on ensuring that Python's duck typing continues to work in your program.

Invariably you'll end up with two camps: those who feel <dynamic lang> scales to large codebases, and those who do not; those in the former camp have (almost never) had to maintain a millions of lines of code before so they do not know just how helpful having a strict compiler is, and just how hard it is to refactor or infer the meaning of code due to the dynamic nature of the language. Linting will only get you so far.

It's perfectly possible to write maintainable, large programs in dynamic languages, but it is infinitely harder than it should be.

Python is actually strongly typed. For example, 1+"1" will fail in python. Type errors will (usually) make the program crash. I believe this is a major plus over javascript when writing and maintaining large programs.

Yes, crash at runtime. That's no good in a production environment when you can have hundreds of code paths and no real way of ensuring they are all correct. Mistakes happen, and code coverage tools are in my experience pretty useless at ensuring all code paths have been exercised (a theoretically futile task as well.)

As for strongly typed- yes, that is true, but it is also dynamically typed, and combined with duck typing you're left with a philosophy that encourages interchangeable types provided the interface is similar. Strong typing come in many forms, and you can have strong types and still have a compiler there to help you out. But of course, no system is perfect.

That's all well and good when you still have a complete, mental overview of the code and how it interacts at a macroscopic level. That's much harder to do when you go from 1 to many developers -- especially if you're developing code to be consumed by other developers.

> Yes, crash at runtime. That's no good in a production environments My point is that it's a lot better than trying to continue in an inconsistent state.

I'm not saying dynamic typed languages are better than static typed languages. Right now I use coq. I write programs and their correctness proofs so I definitely get why you would want some static guarantees.

Oh, I know you weren't. It wasn't a criticism of your observation at all, but one of those "sigh, why can't I have my cake and eat it" moments :)

Python has Abstract Base Classes, and also widely used zope.interfaces. These, along with simplicity, convention(pep8, frameworks), a stable base(stdlib), packages, a standard unittest framework, and a strong internet community help python code bases scale.

JavaScript is missing a stable base, packages, ABCs and interfaces, and a shared testing framework. However many of these are available within separate communities like for example the jQuery community. jQuery provides a stable base, plugins, a good community to share code, a standard testing framework, a way to extend code (plugins) to make it reusable, and now it has a module system which can be shared with other JavaScript frameworks.

I've been on 10+ JavaScript developer projects with 100+ developers over all. I've seen JavaScript code reused by one million+ developers. I've seen open source JavaScript projects with 100+ developers. So I have experienced myself that it can scale up to a certain point.

As long as the team has the best practice skills to do it that is. JavaScript is the ultimate spaghetti making machine in the hands of newbies without direction.

They can scale, but not every code base is well written. Operating from the assumption that best practices is always implemented isn't necessarily true. Sometimes you work with sub-par code or code that was written by someone skilled but due to time pressure; misunderstanding the requirements; or a poorly planned architecture means that it does not lend itself to scalability. That's the problem. You can refactor it, but refactoring code in Python can be a nightmare compared with tools like Visual Studio that make it a breeze -- plus you get the benefit of having a compiler catch errors for you.

Well-thought-out code will always trump bad code; but that's a truism.

No, not crash at runtime. Raise exception in runtime. You know, like NullPointerException in Java? Or ClassCastException? Or OutOfMemeoryException.

You'll have more of the runtime errors in Python, but you're not free from them in other languages either, static typing or not.

That's true, as anybody in, say, Delphi would know, when they hard cast from one type to another.

As for the semantics of crashing vs. not crashing, that is a different argument entirely. Do you want your code to suppress a SyntaxError or TypeError exception and "keep going" - probably not? But maybe? It all depends, I suppose. You'll let the unhandled and unexpected exceptions percolate to the top of the stack and then your "main" method will decide what to do next. But putting square pegs into round holes do happen, and the question then becomes: how do you ensure that does not happen as often?

I never made the argument that languages like C#, Ada, Delphi, or Java were immune from runtime errors. Not at all.

You call that strong typing ? IMHO it is natural that it fails - what did you want as a result?

But 1 + 1.0 works in Python - and I just added float to integer (integer was implicitly converted to float (given sufficiently large integer such behavior will result in erroneous calculation)), which is contrary to what strong typing is about.

This is true, but if that means python isn't strongly typed, then neither is Java or c++, since they too will convert for you. So, what widely used language is strongly typed then?

Ada and Haskell from top of my head.

Yes I do. ()+[] fails too. In javascript: 1+"1" --> "11", []+1 --> "1", null+[] --> "null" etc.

Otoh, the scripting languages generate so much less code that many systems just aren't "large" in them...

If I understand your position, the only thing really needed should be optional typing of parameters (and possibly variables) that are expanded to assert statements on compiling?

And editor support for that typing, of course, to get popups with calling profiles?

There is something in Perl [1], I don't know about Python. Write a PEP. :-)

[1] Method::Signatures@CPAN, et al http://search.cpan.org/~barefoot/Method-Signatures/lib/Metho...

Edit: Basic English grammar, sigh.

True, but large systems are just, well, large. I'm not thinking startup land here, but entrenched applications in bigger businesses that grow organically to meet the need of the business.

Python 3 does have type annotations for use by linting tools, but it'll be another 5-6 years before we'll see that filter down into the current Python ecosystem. And even then, it's only as good as the weakest (non-annotated) piece of code. Retrofitting libraries and existing code will bring about its own issues: how do you accurately hint to linters, in a language that encourages duck typing, the type a method parameter is likely to take?

>>how do you accurately hint to linters, in a language that encourages duck typing, the type a method parameter is likely to take?

You didn't check the Method::Signatures link or do I misunderstand your position?

Also check Moose@CPAN, which have typing of object attributes. It isn't like a fully typed language, but it should help a lot.

Edit: Link discussing type checking of attributes in Perl. Should be possible to have something similar in other dynamic languages? http://search.cpan.org/~doy/Moose-2.0602/lib/Moose/Util/Type...

Edit: Please check the links and see if Perl solves your problems? I am curious. (-: Writing a PEP to get something similar into Python is optional. :-)

I didn't look, tbh. Python 3 already has type annotations, but what interests me if how do you nail down types whilst preserving duck typing, or dealing with a method that is already passed duck typed objects.

I'm not going to copy/paste the content/source from links I've given. :-)

(There are easy-to-read abstracts/"synopsis" at the start of both pages.)

I'd have thought that taking a leaf out of Go's book and having passively-defined interfaces would be a useful concept for linter hints.

"I haven’t listened to the interview so I honestly don’t know if the quote is taken out of context or not"

WTF? -1 for not trying.

There's a panel discussion with:

    Erik Meijer (misc, head of Cloud Programmability Team at Microsoft)
    Gilad Bracha (Dart)
    Martin Odersky (most recently Scala)
    Anders Hejlsberg (Turbo Pascal, Delphi, C# etc)
    Peter Alvaro (researcher e.g. datalog)

And some ungooglable twerp doesn't actually listen to it?

Argument from authority seems spot-on when you're replying to something they said.

Make that -2

in all fairness, that sentence is said more like a passing comment and not a discussion topic

I posted the original to reddit/r/javascript so I've got a point to defend ;)

In the video, they do talk about tooling, IDEs, auto-completion, libraries, dynamic types, and so on. Refactoring tools come up repeatedly.

So you can build up a richer context to the reasoning.

There's even discussion about reasoning about systems made from components running on different machines, which is not far from what this Derek chap responds with as his perceived solution.

Its not all tightly packed into a sound-bite.

Link-bait.

The problem of composability has been debated to death and is in general recognized to be the biggest problem of our software industry. Even Unix-derived operating systems come these days with pretty big and monolithic pieces of software. Even Emacs is technically a big and monolithic piece of software, at least when compared to Vi.

     But Derick, You Write Large JavaScript Apps ...
     I write JavaScript applications that appear to be large.

Not trying to make an argument about who Derick is, but he's arguing against Anders Hejlsberg, a very talented software engineer and language designer. Certainly we should be wary about appeals to authority (just like the one I've made), however let's not get ahead of ourselves and forget the shoulders of the giants on which we stand.

I struggle to maintain a large system in Python (which, at least, is strongly typed).

My major banes are:

* performance. You have a hard job retrofitting performance to Python

* typos. Typos that survive my Integration Tests which can never have enough coverage. Stuff that just won't compile in a static language becomes a runtime issue on a dynamic language, and they can hide for months outside the happy path e.g. in some error handing code

* poor libraries. I get nice warm happy feelings about various async server frameworks that last until I have to debug them and stare at their innards and realised they were contributed to and diluted by twits

* pip/virtualenv doing a poor job with native code in modules

At least it has namespaces / modules and such

At least its not common for every 3rd party lib to want to squat on a function called $

And so on.

Static typing is a major plus when you are beyond the prototyping stage and heading heading towards the release and maintenance phase.

Which is why I'm on the look out for implicitly strongly statically typed languages that look like Python :)

Try pyflakes, pylint, etc. They've helped me find a few typos in exception handlers that are rarely run.

Certainly, large programs are harder to maintain than small programs. Certainly, modularity and encapsulation helps, in any language. The implication that, therefore, choice of language is irrelevant to maintainability is spurious.

did you listen to the talk

Definition of large program is "program, which is hard to maintain". So large programs are hard to maintain by definition.

You're referring to a line I never drew. Programs can be larger or smaller in many dimensions - LoC, complexity, features... Other things being equal, the smaller program is easier to deal with. We don't need to say "these programs are large and these are small" to make valid inferences in this area.

The original talk: http://css.dzone.com/articles/you-can-write-large-programs

Martin Odersky keeps a sober view on all of this. His insights on sound/unsound type systems is really good.

Also, Peter Alvaro makes a very good complaint about having a distinction of library creators and every-day programmers, which the others seem to believe not only to be true, but also to be inevitable (I obviously side with Peter on this).

The conversation is much better than this sad remark.

Reply on news.ycombinator.com