Hacker News Reader

Comment by mlevental | Hacker News Reader

>In Golang, dependencies are minimised. Everyone tries to stick to the standard library as much as possible.

i don't understand this. yes net/http is nice but there's no router, there's no orm (yes i know that's not really possible at all because of strong typing), yes database/sql exists but you know most everyone uses at least sqlx and e.g. lib/pq. yes html/template is nice but we're all doing SPAs now anyway. okay there's logging but e.g. no websockets. so in what sense is the stl really complete? what modern web app could you write with just the stl?

View on HN

You don't need a router (especially for an SPA with client-side routing). If you've got less than 20 or so routes, then a big switch statement is all the routing you need. Learning to write "proper" logic around routes instead of parsing them with regexes was a learning milestone for me.

You don't need an ORM. Mapping database tables to objects directly is a really bad idea. Write SQL, it's easy, simple and cheap. Yes, you'll repeat yourself. Golang generally prefers some repetition to some dependency.

You don't need middleware. Once you get used to handling routes as logic, then the need for middleware kinda vanishes.

You can use the semi-official google websockets package [0]. It's borderline whether this could be considered part of the standard library or not (I use the google uuid package all the time for the same reason). I believe there's an intention to include these packages in the standard library at some point, probably Go v2.

Try it out, it's liberating.

0: https://godoc.org/golang.org/x/net/websocket

I still remember the first time I had to pick apart a gigantic chunk of java code and found a big switch statement at the center instead of a web of introspectively discovered implicitly wired reactively chained bean factory factory factories or some such nonsense. The relief was almost palpable.

You don't need a router, but it is much more readable with one.

You don't need middleware, but it is much more readable with r.Use(jwtVerify) then writing a deeply nested half router with logic.

If you're running a single endpoint, then the router is adding zero value.

Readability is, I think, a function of your expectations. If you're used to seeing a particular incantation, it's readable. Personally (emphasis on the personally, which is my point), I don't think it gets much more readable that a great big switch statement.

I see Go community pushing for no framework everywhere. I agree Go std library is good enough but some people prefer the framework magic because they want the stuff done faster.

Community should better be saying try std library if you find it verbose try framework(router+ORM)

> some people prefer the framework magic because they want the stuff done faster.

Those people should try being maintenance programmers sometime /curmudgeonly-grumble

I agree with everything you said except the websocket.. the best websocket lib for golang until now is https://github.com/gorilla/websocket

It's getting to the point where the Gorilla packages could be considered part of the standard library ;)

I agree, but the point is that you can do this with just the "almost official" package. It's not like you have to use the gorilla package because there's no standard library package.

First, I don't think net/http is nice, but you know what's good about Go? It's so modular that you can replace the builtin HTTP server with other one you like (gorilla/http for example) and your code _can_ just continue to work without modification.

And, Go had many third-party Router, ORM, Websocket and Template Engine implementations out there, many of them are also proud to be ... independent, so you're less likely to end up in a Dependency Hell.

Go standard library is not there to force people to use it, rather, it's there to setup a standard so different people can work together more effectively.

So:

> in what sense is the stl really complete? what modern web app could you write with just the stl?

1, It's huge, but not complete;

2, You can actually do it with just Go standard library (No third-party package) if you REALLY wants to. It's not very hygiene though, you will end up wrap many things up for better project structure (And resulting many packages that you might later proudly put on your GitHub page): Use net/http as HTTP server (Request handling and parsing); use net.Dial to connect database and cache (And write their protocol); html/template for outputting rendered HTML landing page; encoding/json for parsing and outputting API request and respond.

> yes net/http is nice but there's no router, there's no orm... yes database/sql exists but you know most everyone uses... yes html/template is nice but we're all...

Three fallacies of dependencies: https://www.youtube.com/watch?v=yi5A3cK1LNA

> what modern web app could you write with just the stl?

Lots. You just have to change your frame of mind.

> i know that's not really possible at all because of strong typing

Once you realise the above, you also realise that most of the time you don't need all that fancy stuff and you just write super boring code that works.

>> you just write super boring code that works.

This. This is the essence of Go to my mind :)

There is totally a router, ServeMux, which while basic is enough for 90% of basic tasks if you structure your requests REST-y.

As for ORM, you should basically never use an ORM, ever. They’re a crutch. Learn SQL, write some basic mappers and your SQL will be so much quicker and more optimized than an ORM can generate. An ORM is literally throwing money out the window.

https://golang.org/pkg/net/http/#ServeMux.Handle

I've seen teams of Go developers shoot themselves in the foot so many times with their hand-crafted SQL. Yes, they should know SQL and be better. And if they think about it, and focus on it, they are.

But it's the mistakes they don't anticipate making that end up causing impacts on their users. Then they create a fix, but the damage is done.

I think the goal behind a good ORM is not to replace making SQL queries entirely. It's to stop stupid mistakes that result from an uneven distribution of expertise. And if they find some area where performance is particularly sensitive and the ORM is not optimal, they can drop down to raw SQL again.

If your team members all have expertise and your test coverage and code review processes are very thorough, it's probably not necessary. An ORM is not a magic bullet for every team. In the same vein, the dogmatic advice to "always handcraft your SQL" is also not a magic bullet.

yeah, but... same for any code. A team of coders who don't know javascript will create buggy, insecure Node apps. There's nothing about SQL as a coding language that creates this.

It's more that modern coders are used to not having to learn SQL and trusting the ORM to do the thing for them. So when they encounter actual SQL (usually because they need to break the object <-> table mapping) it's all scary and weird.

Back in the day, of course, we had DBA's (DataBase Administrators) to make sure that application devs who didn't know enough SQL to not make a mess of things, didn't actually make a mess of things. And yes, we hated the DBA's because they were always telling us our SQL sucked. Great training though.

So, yeah, your point is valid, but there's a better solution to that problem: hire a DBA

Just to mention a third way - I've been happy with a QueryBuilder on recent projects.

The DB access parts of the code look similar to the rest of the codebase, rather than a chunk of text embedded that looks like ransom note (kidding - I know caps aren't mandatory). Enough fine control to closely influence the generated query. No real performance penalty. There's also a good mocking library available for tests.

> An ORM is literally throwing money out the window.

What? Definitely the opposite. First of all, ORMs are just tools, as the name implies they are object-relational-mappers. That's it, you writing SQL to translate to objects is just you writing an ORM, although a very tiny one.

Modern ORM frameworks are incredibly more productive with better security and reliability than hand-crafted SQL and mapping code. Unless you're working at serious scale, doing custom reporting, or other complex scenarios, they handle 99% of CRUD database operations perfectly well. Wasting time reimplementing an existing tool is quite literally throwing money out the window.

> That's it, you writing SQL to translate to objects is just you writing an ORM, although a very tiny one.

No. Absolutely not. This is the fundamental mess that ORMS create.

Objects DO NOT map cleanly to tables.

Eventually your code discovers this, and suddenly it's a world of pain because the ORM can't handle it.

And yet this is reality: again 99% of applications manipulate objects in code, persist the data in relational databases, and map rows to objects just fine. Some mapping is necessary anyway, and from my experience modern ORM frameworks actually handle complex objects and relations much better than fragile hand-written code.

As mentioned, if you're in the 1% scenario where this approach doesn't work then you already know you should be using something else. Not knowing when to use what is a far bigger issue than any problem an ORM can cause.

I think it's more 99% of databases are warped horribly to pretend that objects map to tables.

If you throw away the concept that anything in your database has to map to anything in your code base, then you design better databases. Different databases, more to the point.

Oh ok, so your code just doesn't deal with data in the database then? I dont understand what point you're trying to make here.

You can lay out your tables however you want, and you can then map them to objects using an ORM. That's what mapping means, to translate from domain to another. There's no limit on how things are actually designed and ORMs usually do a much better job of handling all that bulk code for you. Nothing is stop you from writing SQL yourself if you really need something complex but again, you should know when and where that is.

Mapping objects to tables and back is a solved problem.

Let me tell you one thing that isn't solved yet: Serialization of arbitrary objects to json and back.

How do you deal with two refererences pointing to the same object? Duplicate the data?

What about circular references? Give each object an id and save a symbolic reference?

What about polymorphism? Should a json array be turned into an array or an list or a json map turned into an object or a hashmap?

You need a type discriminator attribute.

ORMs have already solved those problems well.

The actual problem is that databases are distributed systems and you still have to write queries with the still 10 times better querybuilder of your ORM. Just try writing raw JDBC queries. It's so painful I'd still use the ORM even if my project only uses native SQL queries via the ORM api.

You will probably now ask what's the point of ORMs if you still have to write queries? Well, it's in the damn name. Object relational mapper. It maps stuff. It's not advertised as WYQFY. "Writes your queries for you"

Reply on news.ycombinator.com