Hacker News Reader

Go inevitably results in a very large codebase for a given piece of functionality, especially if it’s used alongside Google’s companion projects such as protobufs and grpc. Lack of generics means that it is impossible to generalize over related types, leading to duplicated code, and in some cases, crazy hacks like serializing a struct, passing it through the generic part of the code and then deserializing it again when you need to use it.

Other choices such as structs being initialized with “zero values” in their fields (instead of having a compile error when you forget a field), and the complicated sometimes equivalence of zero values and null create a lot of footguns when refactoring.

Go’s strength is a batteries included stdlib and its looseness can make prototyping faster. But it feels very much like an old school scripting language like php or js, where you have to use countless trivial unit tests to have any confidence about what your code is doing.

> [Go]'s the only language I'm aware of that makes large-scale engineering its north star.

Ada, Eiffel, Java, Pascal and derivatives, C#, Rust. Probably many others but I'd argue outside of tooling (which Go is particularly good at and has done an incredible job of standardizing, Rust being a peer in that regard with cargo) these languages are as good or better than Go for software engineering.

By "north star" I mean the guiding principle that other aspects are sacrificed for. A heuristic might be: "[Lang] would not be [lang] without [aspect]."

I'm sure that the Rust team wants Rust to be a good language for large-scale engineering, but it's not their north star; "fearless concurrency" is probably the closest. For example, Rust is willing to sacrifice simple syntax and fast compile times to achieve more safety; Go is not.

I don't know enough about the other languages to speak confidently about them, but if you can point me to any source materials (i.e. articles or talks by the language creators) that discuss large-scale engineering, I'd be interested to read them!

> it's not their north star; "fearless concurrency" is probably the closest.

That's just a more recent slogan to succinctly express what their real north star means. Their real north star is "safety and correctness", and like all languages, they push to make the most important thing to them easy and approachable.

Here's the thing, I'm not sure as a project saying your focus is large scale engineering and pushing towards that actually gets you closer to the desired result in the end than pushing for safety and correctness instead. I could fully see a language that pushes for safety and correctness eventually being a much better and easier to use choice when evaluating with respect to large scale engineering, because those are extremely important to large scale engineering.

Safety and correctness certainly are important -- but they're also orthogonal to maintainability (among other values)!

It's good to have most of the core values in common, but a language's guiding principle manifests in many small ways that can be very compelling in aggregate. Go's weird capital-letter namespace visibility rule is a great example of this: it has nothing to do with safety or correctness, but it's enormously helpful for getting your bearings in unfamiliar code, because it localizes an important piece of information, saving you a context-switch. Sure, another language might hit upon this feature incidentally (Rust has conventions around capitalization, and IIRC Haskell enforces it to some degree), but how many languages will chance upon all of the little things that make for a great engineering language?

I keep writing poor tutorials (due to space) of Ada here. Check out https://learn.adacore.com/ for more.

Overall I like Go, but for software engineering in the large Ada, in contrast to Go, has:

Explicit separation of specification and implementation of packages and types making it easier to shard the work out to numerous individuals and teams.

What a package exports is explicit, not the result of case conventions.

  package Foo is
    procedure Bar(N : in Integer); -- exported
  private
    procedure Baz(N : out Integer); -- unexported
  end Foo;

  package body Foo is
    -- implementations of the above
  end Foo;

Generics from day 1, which enabled a set of collection packages similar to C++'s STL in Ada 2005 among other things.

Tasks, similar to Go's goroutines, but have a lot of additional guarantees and behaviors that have to be made explicit in Go.

Protected objects and protected types which permit you to write code without explicitly acquiring/releases mutexes and using condition variables.

Design-by-contract with SPARK/Ada becoming a proper subset of Ada in Ada 2012.

> but they're also orthogonal to maintainability

Not entirely. Some amount of maintainability is fixing mistakes that result in less correct or safe programs. By reducing occurrences of these classes of problems later, you're reducing the maintainability burden.

It's the same thing that makes a language with a type system more maintainable in the long run. Eliminating classes of errors pays dividends later. The decision to make is whether the extra effort at the time of creation is worth that reduction. Depending on the type of software, it's expected lifetime, and how many people are expected to use it, that may differ. For large scale engineering, I imagine the extra effort required up front is almost always worth it.

https://www.amazon.com/Eiffel-Language-PRENTICE-HALL-OBJECT-...

> Quickly emerging as the language of choice for developers of quality software, this guide explains how the Eiffel language combines rigorous software engineering principles with advanced object-oriented techniques. It reveals the simplicity, consistency, and blend of various powerful mechanisms in Eiffel, including: an object-oriented structure based on classes ; multiple and repeated inheritance information hiding assertions to guarantee, document, and test the correctness of software components; strong typing backed by support for generic classes; dynamic binding a highly dynamic run-time model supporting garbage collection renaming and redefinition facilities . . . disciplined exception handling and support for persistent objects. As both an implementation language and a high-level notation for analysis and design, Eiffel emphasizes the industrial production of reusable software components. Written for software designers, analysts, and programmers.

https://www.eiffel.com/

Rust was explicitly designed from the beginning for creating maintainable large-scale software.

Here's the original elevator pitch from 2010:

https://github.com/rust-lang/rust/blob/d6b7c96c3eb29b9244ece...

Note the text about "programming in the large".

Eiffel might be another.

I don't know much about Eiffel; do you know of any articles or talks that discuss its design tradeoffs?

Eiffel's bigger contributions are probably:

Design-by-contract. The contracts aren't just comments, they aren't haphazardly enforced by adding asserts to the code. They're explicit parts of the language which allows for a lot of interesting consequences when utilized. Testing can be semi-automated directly from the code, some elements of the program design can be proved to be satisfied statically without the need for compilation and execution. The ability to encode your specification into the language is incredibly useful.

Command-query separation. If methods return a value (queries) they should be referentially transparent, having no side effects. This is a principle so not an explicit language feature, but it can be enforced using DbC.

> Go is explicitly optimized for software engineering;

There's nothing to back up this claim. Using golang in practice shows otherwise. The mess resulting in large code bases is much more difficult to deal with compared to Java for instance.