Dynja

Jinja pseudo-engine focused on DevEx and Performance

Why Dynja?

Let's look at two of the alternatives:

• Askama: extremely fast on benchmarks, but doesn't have a very fun development experience, since you have to recompile your webserver each time you modify a template
• MiniJinja: decent performance on benchmarks, but has an awesome developer experience, with hot reloading and possibly even live reloading

So let's mix both: use MiniJinja for debug mode (better DevEx), and Askama for release mode (better performance)

And that's what Dynja essentially is

How to use?

Add the dynja dependency to your Cargo.toml, along with the askama dependency. The minijinja dependency isn't necessary, because it is only used internally, whereas askama needs to be exported on release builds.

[dependencies]
dynja = "0.1"
askama = "0.12"

Now you can import dynja and use it as if it were askama. Nice huh?

use dynja::Template;

#[derive(Template)]
#[template(path = "index.html")]
struct MyTemplate {
    name: &'static str,
}

fn main() {
    let template = MyTemplate { name: "Test" };
    println!("Template Render: {}", template.render().unwrap());
}

It will automatically choose between minijinja on debug, and askama on release, so you don't have to worry about it.

Have fun!

Benchmarks

NOTE: These benchmarks are not done properly, so they don't represent a real world scenario. They do let you see the difference between switching the engines though.

Tested on: https://github.com/rdbo/axum-htmx-dynja-test

Command: rewrk -c 100 -t 3 -h "http://127.0.0.1:8000" -d 10s

Dynja 0.3.0 (Debug)

Beginning round 1...
Benchmarking 100 connections @ http://127.0.0.1:8000 for 10 second(s)
  Latencies:
    Avg      Stdev    Min      Max      
    26.99ms  10.18ms  0.71ms   67.77ms  
  Requests:
    Total:  36861  Req/Sec: 3690.25
  Transfer:
    Total: 31.92 MB Transfer Rate: 3.20 MB/Sec  

Dynja 0.3.0 (Release)

Beginning round 1...
Benchmarking 100 connections @ http://127.0.0.1:8000 for 10 second(s)
  Latencies:
    Avg      Stdev    Min      Max      
    2.73ms   1.12ms   0.06ms   29.69ms  
  Requests:
    Total: 364482  Req/Sec: 36478.85
  Transfer:
    Total: 315.62 MB Transfer Rate: 31.59 MB/Sec

The release build got about 10 times the amount of requests per second.

You may think this is due to the web server and other packages also being compiled in release vs in debug, and that does play a role in the results.

It's important to say, though, that I've spent some time testing each individual engine on both debug and release, and this big diffence between the engines is expected.

On a side note, this benchmark also doesn't say that minijinja is slow by any means. In other to achieve hot reloading of the templates, we have to clear the cached templates of minijinja for every render(), which means we add a severe bottleneck to its performance to get a better development experience. Here are the results of a test done before hot reload was introduced:

Dynja 0.2.0 (Debug)

Beginning round 1...
Benchmarking 100 connections @ http://127.0.0.1:8000 for 10 second(s)
  Latencies:
    Avg      Stdev    Min      Max      
    8.41ms   3.08ms   0.30ms   24.74ms  
  Requests:
    Total: 118376  Req/Sec: 11851.72
  Transfer:
    Total: 102.51 MB Transfer Rate: 10.26 MB/Sec

License

This project is licensed under the GNU AGPL-3.0. No later versions allowed.

Read the LICENSE file in the root directory of the project for more information.

Considerations

Even though MiniJinja and Askama are both related to Jinja, they are not 100% compatible with each other. So be wary of inconsistencies!