ts-morph

Performance

There's a lot of opportunity for performance improvements. The library originally started off favouring correctness, but it's now starting to switch to improving performance.

View Issues

Manipulations are slow...

Right now with every manipulation the following occurs:

  1. The file text is updated.
  2. The new text is parsed and a new AST is created using the compiler API.
  3. The previously wrapped nodes are backfilled with new compiler nodes.

This might not be too bad when working with small to medium sized files, but large files may take a bit of time. If you find it's too slow, then I recommend reading the performance tips below.

Note: I'm working on eliminating the need to do a complete parse of the source file between manipulations. Once implemented these performance tips won't be necessary.

Performance Tip: Work With Structures Instead

Structures are simplified ASTs. You can get a huge performance improvement by working with structures as much as possible. This is especially useful to do if you are code generating.

Example:

// this code will get a structure from declarations.ts, make all the descendants not be exported,
// then create a new file called private.ts from that structure
import { forEachStructureChild, SourceFileStructure, StructureKind, Structures, Structures } from "ts-morph";

const project = new Project({ tsConfigFilePath: "tsconfig.json" });
const classesFile = project.getSourceFileOrThrow("declarations.ts");
const classesFileStructure = classesFile.getStructure();

removeExports(classesFileStructure);

project.createSourceFile("private.ts", classesFileStructure);

function removeExports(structure: Structures) {
  forEachStructureChild(structure, removeExports);

  if (Structure.isExportable(structure))
    structure.isExported = false;
}

Read more in structures.

Performance Tip: Batch operations

You can reduce the amount of parsing that needs to happen by batching operations.

For example, instead of writing code like this:

for (const classStructure of classStructures)
  sourceFile.addClass(classStructure);

Write this instead:

sourceFile.addClasses(classStructures);

Performance Tip: Analyze then Manipulate

If the code analysis is using types, symbols type checker, or program, then a large performance improvement can be gained by doing an initial analysis of the code first, then afterwards carrying out the manipulations.

For example, given the following code:

for (const sourceFile of sourceFiles) {
  for (const classDec of sourceFile.getClasses()) {
    if (someCheckOnSymbol(classDec.getSymbolOrThrow()))
      classDec.remove();
  }
}

Write it this way instead:

for (const classDec of getClassesToRemove())
  classDec.remove();

function getClassesToRemove() {
  const classesToRemove: ClassDeclaration[] = [];

  for (const sourceFile of sourceFiles) {
    for (const classDec of sourceFile.getClasses()) {
      if (someCheckOnSymbol(classDec.getSymbolOrThrow()))
        classesToRemove.push(classDec);
    }
  }

  return classesToRemove;
}

This is because the program is reset between manipulations.

Tracking Nodes - Overview

This library makes manipulations easy for you by keeping track of how the underlying syntax tree changes between manipulations.

Behind the scenes, when you manipulate the AST:

  1. A new source file is created using the TypeScript compiler.
  2. The previously navigated nodes have their underlying compiler nodes replaced with the new compiler nodes.

It's why you can do this:

// sourcefile contains: interface Person { name: string; }
const personInterface = sourceFile.getInterfaceOrThrow("Person");
const nameProperty = personInterface.getPropertyOrThrow("name");
nameProperty.setType("number");
nameProperty.getText(); // "name: number;"

Instead of having to renavigate the tree after each manipulation:

// thankfully the library does not work this way
let personInterface = sourceFile.getInterfaceOrThrow("Person");
let nameProperty = personInterface.getPropertyOrThrow("name");
nameProperty.setType("number");
nameProperty.getText(); // "name: string;"
personInterface = sourceFile.getInterfaceOrThrow("Person");
nameProperty = personInterface.getPropertyOrThrow("name");
nameProperty.getText(); // "name: number;"

When thinking about performance, the key point here is that if you have a lot of previously navigated nodes and a very large file, then manipulation might start to become sluggish.

Forgetting Nodes (Advanced)

The main way to improve performance when manipulating, is to "forget" a node when you're done with it.

personInterface.forget();

// or to only forget a node's descendants that are currently in the wrapped cache
sourceFile.forgetDescendants();

That will stop tracking the node and all its previously navigated descendants (ex. in this case, nameProperty as well). It won't be updated when manipulation happens again. Note that after doing this, the node will throw an error if one of its properties or methods is accessed.

Forget Blocks (Advanced)

It's possible to make sure all created nodes within a block are forgotten:

import { ClassDeclaration, InterfaceDeclaration, ModuleDeclaration, Project } from "ts-morph";

const project = new Project();
const text = "namespace Namespace { interface Interface {} class Class {} }";
const sourceFile = project.createSourceFile("file.ts", text);

let moduleDeclaration: ModuleDeclaration;
let interfaceDeclaration: InterfaceDeclaration;
let classDeclaration: ClassDeclaration;

project.forgetNodesCreatedInBlock(remember => {
  moduleDeclaration = sourceFile.getModuleOrThrow("Namespace");
  interfaceDeclaration = moduleDeclaration.getInterfaceOrThrow("Interface");
  classDeclaration = moduleDeclaration.getClassOrThrow("Class");

  // you can mark nodes to remember outside the scope of this block...
  // this will remember the specified node and all its ancestors
  remember(interfaceDeclaration); // or pass in multiple nodes
});

moduleDeclaration.getText(); // ok, child was implicitly marked to remember
interfaceDeclaration.getText(); // ok, was explicitly marked to remember
classDeclaration.getText(); // throws, was forgotten

// alternatively, return the node to remember it
const node = project.forgetNodesCreatedInBlock(() => {
  const classDec = sourceFile.getClassOrThrow("MyClass");
  // ...do a lot of stuff...
  return classDec;
});

node.getText(); // ok

Also, do not be concerned about nesting forget blocks. That is perfectly fine to do:

project.forgetNodesCreatedInBlock(() => {
  moduleDeclaration = sourceFile.getModuleOrThrow("Namespace");
  interfaceDeclaration = moduleDeclaration.getInterfaceOrThrow("Interface");

  project.forgetNodesCreatedInBlock(remember => {
    classDeclaration = moduleDeclaration.getClassOrThrow("Class");
    remember(moduleDeclaration);
  });

  classDeclaration.getText(); // throws, was forgotten outside the block above
  interfaceDeclaration.getText(); // ok, hasn't been forgotten yet
});

moduleDeclaration.getText(); // ok, was marked to remember in one of the blocks
interfaceDeclaration.getText(); // throws, was forgotten
classDeclaration.getText(); // throws, was forgotten
Async

This method supports async and await:

await project.forgetNodesCreatedInBlock(async remember => {
  // do stuff
});