Hacker News Reader

Comment by wokwokwok | Hacker News Reader

> I have been able to implement things in 4 hours that would easily have taken 4 weeks if I had to argue with other developers about the shape of a JSON blob on the wire.

I can’t comment on your specific circumstances because I don’t know the details, but this reads like someone who is drinking the blazer coolaid.

It’s a technology like any other, with specific trade offs.

If you have a front end team who lacks the basic competence to do something that takes four hours in blazor in less than 4 weeks, then maybe it’s a good trade off.

...but in most cases, for most people, that won’t be the case.

It’s also quite slow, hard to integrate with existing js technology (ie bridge) and very difficult to debug production issues with in my experience.

View on HN

The entire point is to avoid javascript as much as possible and to force the few edge cases into an interop shim.

We have a very complex Blazor webapp and the JS interop shim is fewer than 200 lines of code. This is 100% of the custom javascript for the entire application. Front and back. We do things like GetClientRect, Get/Set/Delete cookies, keyboard/mouse event subscription, etc.

There is little-to-no context switching when operating with one of these codebases. The notion of "front-end" and "back-end" is completely meaningless here. The "back-end" is literally an F12 keystroke away from the "front-end".

> and very difficult to debug production issues with in my experience

What sort of production issues have you been encountering that were difficult to track down? I rather enjoy being able to set breakpoints in my view templates to confirm the state of affairs in local testing.

> The notion of "front-end" and "back-end" is completely meaningless here.

So to start with, a disclaimer: I carry some baggage here, from working occasionally on 10+ year old GWT app that's developed into quite a tangled mess.

Having said that -- this sort of thing doesn't seem like an obvious win to me. The notion of "front end" and "back end" represents a real-world distinction that you want to pay attention to. Making it too hard to see, or too easy to communicate across, brings some risk of losing track of which is which, and putting logic that belongs on the server into the client or vice versa. Or having the two communicate with each other much too often, which may not cause any noticeable problems during development, but, in production, can cripple the site for users with higher-latency Internet connections.

Ditto, I've no desire at all for web applications--or heaven forbid clients--to poke directly at a database or cache, or otherwise obviate the need for APIs. One will need to deploy, scale, monitor, and secure those things very differently from one another.

We use SQLite for storage of all things. The entire application lives on a single machine.

   OurApp.exe: 
   SQLite <-> C# Services <-> Razor Components

When you have a stack like this, Blazor goes into additional dimensions of productive. Every F5 of the app is a complete stack running without any frustration in setting up databases, containers, message buses, etc. None of the concerns regarding scalability apply when you plan to go vertical.

I would not currently argue for developing blazor server apps as a replacement for 100% of web applications. But, there are situations where you have a constrained user base and predictable performance characteristics which provides enough confidence to go all-in on a single fast box.

Not familiar with .NET, but am intrigued by the Sqlite all-the-things bit -- can you run your LINQ queries client-side? (e.g. for offline capable apps)

LINQ is built into the core .NET runtime (and the syntax into C#) and available everywhere (I think LINQ was introduced with .NET 4 and C# 3.0 some >10 years ago). So doesn't matter if client/server or blazor. But you need some kind of LINQ-to-your-datasource adapter (i.e. to Sqlite, MongoDB etc.) which might have different availability depending on your environment (i.e. a Sqlite adapter would clearly somehow depend on a native sqlite client - but AFAIK is available in Blazor through WebAssembly).

We use the Blazor Server hosting model, which is inherently incapable of running offline. For our specific use case, this is actually more ideal than any notion of offline capability. In this model, any LINQ would be running on the server.

Now, you certainly can run LINQ client-side if you were on a Blazor WASM (or hypothetically, Desktop) application. This is kind of the entire point of the stack.

Blazor has two modes: Blazor client-side is C# code running on a WASM dotnet runtime. It does have the frontend/backend split. Data still has to be serialized (as HTTP calls since that's all a browser can do) but it's done using the same classes & libraries as on the backend.

Blazor server-side has no distinction though because everything is running on the server, and the UI is just a thin stream of HTML updates.

The model you choose is up to you based on resources, latency requirements, payload size, etc. Blazor just gives you more ways to get things done.

> I rather enjoy being able to set breakpoints in my view templates to confirm the state of affairs in local testing.

That’s a local development environment.

Have you had similar experiences when dealing with deployed applications? How did you do it? It’s rather more difficult when, eg. You can’t have a copy of visual studio there to use on your production deployments.

The fact that you have to interface between different teams/languages is often a big part of the slowdown.

The productivity improvements are real. Not having to design and expose an API for the frontend and instead just reuse the same classes and services is a massive shortcut. Running Blazor on the server is even easier because there's no serialization at all and you can directly call database queries in response to UI events.

I suggest watching some of the Blazor presentations to see just how much complexity can be built in minutes compared to more traditional frontend frameworks: https://www.youtube.com/watch?v=CEjqhTGrqDY

Isn't load balancing still an issue with server side blazor? Because of SignalR for the delta between server and client?

Yes. A single-server can scale to thousands of concurrent users however it's not recommend for massive public sites. Works great for admin backends though where the usage is known.

Then why didn't Meteor take off fully? Because that was the promise of universal JS, about 10 years ago...

Meteor's problems were around performance and features (or the lack of them) IMO, which are not in principle ones that will crop up if you try to unite a stack around a single language.

Javascript is still struggling to be fully capable server-side today, let alone 10 years ago. Going in reverse is much more functional.

> The fact that you have to interface between different teams/languages is often a big part of the slowdown

Different teams is a business organization problem, and if you have that problem, you just aren’t going to choose the kind of server-side-code-with-generated-frontend framework that Blazor is (and of which there are several besides Blazor), so it doesn’t solve any problem there.

Different languages without different teams is a much smaller problem.

I haven't use Blazor but they mentioning about avoiding the json API with all the encoding/deconding that entails sounds nice, plus only doing routing in the backend. These frameworks (liveview, blazor, hotwire, etc...) seem like a nice compromise that can serve well most webapps out there.

Reply on news.ycombinator.com