Comment by saberience | Hacker News Reader

Comment by saberience | Hacker News Reader

Parent Full thread

Well, for a start... having a single file called "server.rs" which is 2256 lines long is already disqualifying. This file, code structure would be incredibly poor for developer efficiency. It just grows and grows and does all the things while having a vague sounding name like "server.rs" Why is server.rs doing "EditSite." What does the name EditSite mean anyway?

Poor code reuse, same bits of code copy and pasted all over instead of being refactored out into their own methods. Methods which are doing too much different crap. The single responsibility principle is broken everywhere.

Imagine trying to write unit tests for "Perform for EditPost", this method is doing far too much and would be a pain in the ass to fully test. If you're making a method called "EditPost" it should only be doing the actual post editing, not making connections, or user validation or checking for a ban. These should happen in a method which only handles connections, or a method which is called "ValidateUser" which is properly re-used.

Generally speaking, the structure as a whole isn't very readable, poor names, methods and files which are doing too much. The project wouldn't scale well with a bigger team or if it grew in terms of codebase size.

What is the max lines in a single file for a developer "to get fired" at your company?

573

The company has a shell script running in Cron. If it finds a source file with more than 1000 lines it deletes the author from the payroll.

Sadly this cron has fired the shell script author, and no one dares to touch it.

No, it is written in APL for extra conciseness. When they wrote it in shell it deleted the authors from the payroll.

The author is entire team.

It's a ridiculous metric.

I do have to say, 2k lines for a single file is scary sounding to me. It would probably make me consider splitting it up.

BUT, sometimes things to just logically belong together and having one file to find them is just easier.

It's a rediculous metric to use.

Depends a lot on tooling as well. If you have tools that give you effective jump-to-definition, find-usages, etc, then the file structure sort of fades away, because you jump from fragment to fragment without really worrying about what files they are in.

Ever had the pleasure of opening a 20k+ line file in IntelliJ? (Not 2k, but 20k!). That happened one time, it wasn't a fun experience.

I consider IJ and most jetbrains products to be top-notch but that was a crippling experience. And I can't really say I blame them :)

EDIT: I agree with you by the way* just saying that even good tools can fail when the sizes get crazy!

Small note here: it's obviously not it's an "absolute" rule that X lines of codes = fired. But for me, it's just a major code smell when people are writing very large single files with generalized names like "server.rs" or "processor.java."

It tends to mean that someone isn't following the single responsibility principle and the code is harder to understand and less maintainable. Obviously these aren't hard and fast rules but are combined with many other code smells and metrics.

For example, I really really dislike long methods with generic names. Usually this means there are too many things happening in a single method. The method name, it's input and output parameters should give the reader a fairly exact idea of what's happening in the method.

So if I wrote a method with this signature: public List<bool> ProcessData(List<string> input)

If you tried to predict what this might contain in the body, you would have zero idea.

Now this method:

public List<bool> ValidateVipCodes(List<string> input)

Just seeing this signature, you could easily imagine what the body would look like without much thought at all.

Great code should be structured in such a way that someone can view the project, ONLY LOOKING at file names and method names and have a good idea about what is going on and how it might work.

There are a few opportunities for code generation here.

The server.rs looks like an implementation of RPC over websockets. So a refactoring might be to model the interface using something like https://grpc.io/ or any other IDL then code generate Rust server stubs and Typescript client stubs.

Cargo has a mechanism for this. https://doc.rust-lang.org/cargo/reference/build-scripts.html

The other area where there is some boiler plate is the Diesel ORM. Diesel does generate some code ie. schema.rs but you are left to write the structs that implement these as well as Form structs for passing data to Diesel.

> 2256 lines long is already disqualifying

I'd love to hear a justification for this idea.

Reply on news.ycombinator.com