fomkin/levsha

Fast Scala eDSL for HTML

Source code at GitHub https://github.com/fomkin/levsha

{
  "createdAt": "2017-04-28T17:15:38Z",
  "defaultBranch": "master",
  "description": "Fast Scala eDSL for HTML ",
  "fullName": "fomkin/levsha",
  "homepage": "",
  "language": "Scala",
  "name": "levsha",
  "pushedAt": "2022-09-13T04:16:07Z",
  "stargazersCount": 56,
  "topics": [],
  "updatedAt": "2025-08-28T07:18:39Z",
  "url": "https://github.com/fomkin/levsha"
}

Levsha

Levsha is a fast HTML template engine and eDSL for Scala 2.12, 2.13 and Scala 3. Optimized templates works without additional memory allocation. Levsha supports changeset inference, which allows to use it as virtual-dom-like middleware.

Static rendering

You can use Levsha as a static HTML renderer.

libraryDependencies += "com.github.fomkin" %% "levsha-core" % "1.2.0"

// In your code
import levsha.text.renderHtml
import levsha.dsl._
import html._

val features = Seq("Very fast", "Memory-effective")

val html = renderHtml {
  optimize {
    body(
      div(
        clazz := "title",
        backgroundColor @= "red",
        "Hello, I'm Levsha!"
      ),
      ul(clazz := "list",
        features map { feature =>
          li(class := "item", feature)
        }
      )
    )
  }
}

println(html)

<body>
  <div style="background-color: red" class="title">Hello, I'm Levsha!</div>
  <ul class="list">
    <li class="item">Very fast</li>
    <li class="item">Memory-effective</li>
  </ul>
</body>

Benchmarks

Benchmarks show that Levsha is really fast. Unlike Twirl, Levsha’s performance does not depend on template complexity.

Test	Engine	Ops/s
simpleHtml	levsha	1336693,499
simpleHtml	scalatags	533740,566
simpleHtml	twirl	5950436,854
withConditionAndLoop	levsha	1299646,768
withConditionAndLoop	scalatags	531345,430
withConditionAndLoop	twirl	239537,158
withVariables	levsha	1140298,804
withVariables	scalatags	483508,457
withVariables	twirl	2146419,329

In your sbt shell.

bench/jmh:run .StaticRenderingComparision

As a virtual DOM

Levsha can be used as virtual-DOM-like middleware. Unlike other popular virtual DOM solutions, Levsha doesn’t allocate additional memory for construction of a new virtual DOM copy. Also it does not allocate memory in changes inferring phase. Levsha’s memory usage is constant.

libraryDependencies += "com.github.fomkin" %%% "levsha-dom" % "1.2.0"

// In your code
import org.scalajs.dom._
import levsha.dom.render
import levsha.dom.event
import levsha.dsl._
import html._

case class Todo(id: String, text: String, done: Boolean)

def onSubmitClick() = {
  val input = document
    .getElementById("todo-input")
    .asInstanceOf[html.Input]
  val inputText = input.value
  // Reset input
  input.value = ""
  val newTodo = Todo(
    id = Random.alphanumeric.take(5).mkString,
    text = inputText,
    done = false
  )
  renderTodos(todos :+ newTodo)
}

def onTodoClick(todo: Todo) = {
  renderTodos(
    todos.updated(
      todos.indexOf(todo),
      todo.copy(done = !todo.done)
    )
  )
}

def renderTodos(todos: Seq[Todo]): Unit = render(document.body) {
  optimize {
    body(
      div(clazz := "title", "Todos"),
      ul(clazz := "list",
        todos map { todo =>
          li(
            todo match {
              case Todo(_, text, true) => strike(text)
              case Todo(_, text, false) => span(text)
            },
            event("click")(onTodoClick(todo))
          )
        }
      ),
      input(id := "todo-input", placeholder := "New ToDo"),
      button("Submit", event("click")(onSubmitClick()))
    )
  }
}

val todos = Seq(
  Todo("1", "Start use Levsha", done = false),
  Todo("2", "Lean back and have rest", done = false)
)

renderTodos(todos)

Memory allocation model explanation

As noted below Levsha does not make additional memory allocations if template optimized. It is possible because optimized template, in compile-time rewrites into calls of RenderContext methods (unlike other template engines which represent their templates as AST on-heap).

For example,

optimize {
  div(clazz := "content",
    h1("Hello world"),
    p("Lorem ipsum dolor")
  )
}

Will be rewritten to

Node { renderContext =>
  renderContext.openNode(XmlNs.html, "div")
  renderContext.setAttr(XmlNs.html, "class", "content")
  renderContext.openNode(XmlNs.html, "h1")
  renderContext.addTextNode("Hello world")
  renderContext.closeNode("h1")
  renderContext.openNode(XmlNs.html, "p")
  renderContext.addTextNode("Lorem ipsum dolor")
  renderContext.closeNode("p")
  renderContext.closeNode("div")
}

In turn, RenderContext (namely DiffRenderContext implementation) saves instructions in ByteBuffer to infer changes in the future.

Of course, Levsha optimizer does not cover all cases. When optimization can’t be performed Levsha just applies current RenderContext to the unoptimized node.

optimize {
  ul(
    Seq(1, 2, 3, 4, 5, 6, 7).collect {
      case x if x % 2 == 0 => li(x.toString)
    }
  )
}

// ==>

Node { renderContext =>
  renderContext.openNode(XmlNs.html, "ul")
  Seq(1, 2, 3, 4, 5, 6, 7)
    .collect {
      case x if x % 2 == 0 =>
        Node { renderContext =>
          renderContext.openNode(XmlNs.html, "li")
          renderContext.addTextNode(x.toString)
          renderContext.closeNode("li")
        }
    }
    .foreach { childNode =>
      childNode.apply(renderContext)
    }
  renderContext.closeNode("ul")
}

When you write your Levsha templates, keep in your mind this list of optimizations:

Nodes and attrs in branches of if expression will be moved to current RenderContext
Same for cases of pattern matching
xs.map(x => div(x)) will be rewritten into a while loop
maybeX.map(x => div(x)) will be rewritten into an if expression
void will be removed

The third item of this list shows us how to rewrite previous example so that optimization could be performed.

optimize {
  ul(
    Seq(1, 2, 3, 4, 5, 6, 7)
      .filter(x => x % 2 == 0)
      .map { x => li(x.toString) }
  )
}

// ==>

Node { renderContext =>
  renderContext.openNode(XmlNs.html, "div")
  val iterator = Seq(1, 2, 3, 4, 5, 6, 7)
    .filter(x => x % 2 == 0)
    .iterator
  while (iterator.hasNext) {
    val x = iterator.next()
    renderContext.openNode(XmlNs.html, "li")
    renderContext.addTextNode(x.toString)
    renderContext.closeNode("li")
  }
  renderContext.closeNode("div")
}

Optimizer options

You can pass this options to SBT.

$ sbt -Doption=value

Option	Description	Possible values	Default
`levsha.optimizer.logUnableToOptimize`	Write positions of unoptimized parts of code to a file.	true/false or file name	false
`levsha.optimizer.unableToSort.forceOptimization`	If tag node content couldn’t be sorted in compile time, optimizer will keep the code unoptimized (so content will be sorted in runtime). You can force optimizer to ignore unspecified Documents (that couldn’t be sorted) and optimize anyway, but keep in mind that node content should be ordered (styles, attrs, nodes).	true/false	false
`levsha.optimizer.unableToSort.warnings`	Warns that optimizer can’t sort tag content when optimization is forced.	true/false	true

Worthy to note

The Tale of Cross-eyed Lefty from Tula and the Steel Flea