Overview

ts-poet is a TypeScript code generator that is a small wrapper around template literals.

It lets you generate code as "just strings" (no need to tediously create a low-level AST), but then also:

1. "Auto imports" only the types/symbols that are actually used in the output

   I.e. you use const Foo = imp("Foo@foo") to define the imports you need in your generated code, use them like function printFoo(foo: ${Foo}): void in your code template strings, and then ts-poet creates the entire import stanza of import { Foo } from foo at the top of your generated output.

   This freedom, to not worry about what symbols you do/do not need make "import ..." lines for, lets you break up/decompose your code generation logic, so that you can have multiple levels of helper methods/etc. that can return code template literals that embed both the code itself and the necessary type imports.

   And when the final file is generated, ts-poet will collect and emit the necessary imports.

2. Automatically avoids import symbol collisions.

   If you're generating types from an external schema, you'll occassionally run into naming conflicts where the external schema has a name that conflicts with your library's own internal name.

   Like the external schema has a Message table/resource, so we want to generate a class Message { ... } type, but Message is also the name of a symbol used by our library itself, i.e. import { Message } from runtime-library for doing like Message.encode or Message.decode calls as part of our implementation.

   With ts-poet, if you declare the symbols in your generated output as class ${def("Message")}, ts-poet will give namespace preference to those definitions, and automatically rewrite the Message import to import { Message as Message1 } from runtime-library, both at the declaration site, as well as the usage sites, i.e. ${Message}.encode(...) in the generated output will automatically become Message1.encode(...) to reflect the rewritten symbol name.

3. Includes any other conditional output (see later), as/if needed.

4. Formats the output with dprint-node, an extremely fast formatter with "basically prettier-ish" output.

   ts-poet originally used prettier directly, but it became the bottleneck for multiple projects that use ts-poet for their code generation, even with caching to only format actually-changed output, so we switched to dprint and saw dramatic speedups.

Example

Here's some example HelloWorld output generated by ts-poet:

import { Observable } from "rxjs/Observable";

export class Greeter {
  private name: string;

  constructor(private name: string) {}

  greet(): Observable<string> {
    return Observable.from(`Hello $name`);
  }
}

And this is the code to generate it with ts-poet:

import { code, imp } from "ts-poet";

// Use `imp` to declare an import that will conditionally auto-imported
const Observable = imp("Observable@rxjs");

// Optionally create helper consts/methods to better organize
// the code generator logic
const greetMethod = code`
  greet(): ${Observable}<string> {
    return ${Observable}.from(\`Hello \${this.name}\`);
  }
`;

// Combine all of the output (note no imports are at the top, they'll be auto-added)
const greeter = code`
  export class Greeter {
    private name: string;
    constructor(name: string) {
      this.name = name;
    }
    ${greetMethod}
  }
`;

// Generate the full output, with imports
const output = greeter.toString();

Formatting Output

By default, we configure dprint-node with "prettier-ish" settings that attempt to match prettier, given that we assume most projects are using prettier for their formatting.

If you'd like to customize the settings, you can either:

• If using ts-poet programmatically, you can pass the dprintOptions key to Code.toString({ ... }), or

• If using ts-poet via another library (like Joist), you can create a .dprint.json file that should get picked up automatically (see the dprint docs)

For either option, the dprint config covers the available formatting options.

Import Specs

Given the primary goal of ts-poet is managing imports, there are several ways of specifying imports via the imp function:

• imp("Observable@rxjs") --> import { Observable } from "rxjs" (named import)
• imp("Observable@./Api") --> import { Observable } from "./Api"
• imp("Observable:Obs@rxjs") --> import { Observable as Obs } from "rxjs" (renamed import)
• imp("t:Observable@rxjs") --> import type { Observable } from "rxjs" (type import)
• imp("t:Observable:Obs@rxjs") --> import type { Observable as Obs } from "rxjs"
• imp("api*./Api") --> import * as api from "./Api" (namespace import)
• imp("api=./Api") --> import api from "./Api" (default import)
• imp("describe+mocha") --> import "mocha" (implicit import)

Avoiding Import Conflicts

Sometimes code generation output may declare a symbol that conflicts with an imported type (usually for generic names like Error).

ts-poet will automatically detect and avoid conflicts if you tell it which symbols you're declaring, i.e.:

const bar = imp('Bar@./bar');
const output = code`
  class ${def("Bar")} extends ${bar} {
     ...
  }
`;

Will result in the imported Bar symbol being remapped to Bar1 in the output:

import { Bar as Bar1 } from "./bar";
class Bar extends Bar1 {}

This is an admittedly contrived example for documentation purposes, but can be really useful when generating code against arbitrary / user-defined input (i.e. a schema that happens to uses a really common term).

saveFiles

If you're generating multiple files, ts-poet provides a saveFiles command that can help with conditional output, i.e.:

const author = {
  name: "Author.ts",
  contents: "class Author extends AuthorCodegen {}",
  overwrite: false,
};
const authorCodegen = {
   name: "AuthorCodegen.ts",
   contents: "class AuthorCodegen {}",
   overwrite: true,
};
await saveFiles({
 directory: "./src/entities",
 files: [author, authorCodegen]
});

Conditional Output

Sometimes when generating larger, intricate output, you want to conditionally include helper methods. I.e. have a convertTimestamps function declared at the top of your module, but only actually include that function if some other part of the output actually uses timestamps (which might depend on the specific input/schema you're generating code against).

ts-poet supports this with a conditionalOutput method:

const convertTimestamps = conditionalOutput(
  // The string to output at the usage site
  "convertTimestamps",
  // The code to conditionally output if convertTimestamps is used
  code`function convertTimestamps() { ...impl... }`,
);

const output = code`
  ${someSchema.map(f => {
    if (f.type === "timestamp") {
      // Using the convertTimestamps const marks it as used in our output
      return code`${convertTimestamps}(f)`;
    }
  })}
  // The .ifUsed result will be empty unless `convertTimestamps` has been marked has used
  ${convertTimestamps.ifUsed}
`;

And your output will have the convertTimestamps declaration only if one of the schema fields had a timestamp type.

This helps cut down on unnecessary output in the code, and compiler/IDE warnings like unused functions.

Literals

If you want to add a literal value, you can use literalOf and arrayOf:

code	output
let a = ${literalOf('foo')}	let a = 'foo';
let a = ${arrayOf(1, 2, 3)}	let a = [1, 2, 3];
let a = ${{foo: 'bar'}}	let a = { foo: 'bar' };
let a = ${{foo: code`bar`}}	let a = { foo: bar };

esModuleInterop Support

Unfortunately some dependencies need different imports based on your project's esModuleInterop setting.

For example, with protobufjs, the Reader symbol is imported differently:

// With esModuleInterop: true, need to use default import
// import m1 from "protobufjs"
// let r1: m1.Reader = ...
const r1 = imp("Reader@protobufjs")

// With esModuleInterop: false, need to use module star import
// import * as m1 from "protobufjs"
// let r1: m1.Reader = ...
const r2 = imp("Reader@protobufjs")

For these scenarios, you can use forceDefaultImport or forceModuleImport:

const Reader = imp("Reader@protobufjs")
const c = code`let r1: ${Reader} = ...`
const esModuleInterop = fromYourConfig();
console.log(c.toString(
  esModuleInterop
    ? { forceDefaultImport: ["protobufjs"] }
    : { forceModuleImport: ["protobufjs"] }
));

This is most useful for frameworks that generate code, and have to support downstream projects that might have either esModuleInterop setting.

Similarly, you can force the import require syntax, to support projects that use the pre-default exports export = typing syntax, to export a single symbol:

const Long = imp("Long=long")
console.log(c.toString({ forceRequireImport: ["long"] }));
// outputs import Long = require("long")

History

ts-poet was originally inspired by Square's JavaPoet code generation DSL, which has a very "Java-esque" builder API of addFunction/addProperty/etc. that ts-poet copied in it's original v1/v2 releases.

JavaPoet's approach worked very well for the Java ecosystem, as it was providing three features:

1. nice formatting (historically code generation output has looked terrible; bad formatting, bad indentation, etc.)
2. nice multi-line string support, via appendLine(...).appendLine(...) style methods.
3. "auto organize imports", of collecting imported symbols across the entire compilation unit of output, and organizing/formatting them at the top of the output file.

However, in the JavaScript/TypeScript world we have prettier for formatting, and nice multi-line string support via template literals, so really the only value add that ts-poet needs to provide is the "auto organize imports", which is what the post-v2/3.0 API has been rewritten (and dramatically simplified as a result) to provide.