Refactoring with Codemods to Automate API Adjustments

Clear a Stale Function Toggle

Let’s begin with a easy but sensible instance to show the
energy of codemods. Think about you’re utilizing a characteristic
toggle in your
codebase to manage the discharge of unfinished or experimental options.
As soon as the characteristic is dwell in manufacturing and dealing as anticipated, the subsequent
logical step is to scrub up the toggle and any associated logic.

As an illustration, think about the next code:

const information = featureToggle('feature-new-product-list') ? { identify: 'Product' } : undefined;

As soon as the characteristic is totally launched and not wants a toggle, this
could be simplified to:

const information = { identify: 'Product' };

The duty entails discovering all cases of featureToggle within the
codebase, checking whether or not the toggle refers to
feature-new-product-list, and eradicating the conditional logic surrounding
it. On the identical time, different characteristic toggles (like
feature-search-result-refinement, which can nonetheless be in improvement)
ought to stay untouched. The codemod must perceive the construction
of the code to use modifications selectively.

Understanding the AST

Earlier than we dive into writing the codemod, let’s break down how this
particular code snippet seems to be in an AST. You should use instruments like AST
Explorer to visualise how supply code and AST
are mapped. It’s useful to grasp the node sorts you are interacting
with earlier than making use of any modifications.

The picture under reveals the syntax tree when it comes to ECMAScript syntax. It
incorporates nodes like Identifier (for variables), StringLiteral (for the
toggle identify), and extra summary nodes like CallExpression and
ConditionalExpression.

Determine 2: The Summary Syntax Tree illustration of the characteristic toggle test

On this AST illustration, the variable information is assigned utilizing a
ConditionalExpression. The check a part of the expression calls
featureToggle('feature-new-product-list'). If the check returns true,
the consequent department assigns { identify: 'Product' } to information. If
false, the alternate department assigns undefined.

For a job with clear enter and output, I want writing exams first,
then implementing the codemod. I begin by defining a destructive case to
guarantee we don’t by chance change issues we wish to go away untouched,
adopted by an actual case that performs the precise conversion. I start with
a easy state of affairs, implement it, then add a variation (like checking if
featureToggle known as inside an if assertion), implement that case, and
guarantee all exams move.

This method aligns properly with Take a look at-Pushed Improvement (TDD), even
for those who don’t apply TDD usually. Understanding precisely what the
transformation’s inputs and outputs are earlier than coding improves security and
effectivity, particularly when tweaking codemods.

With jscodeshift, you’ll be able to write exams to confirm how the codemod
behaves:

const remodel = require("../remove-feature-new-product-list");

defineInlineTest(
  remodel,
  {},
  `
  const information = featureToggle('feature-new-product-list') ? { identify: 'Product' } : undefined;
  `,
  `
  const information = { identify: 'Product' };
  `,
  "delete the toggle feature-new-product-list in conditional operator"
);

The defineInlineTest perform from jscodeshift lets you outline
the enter, anticipated output, and a string describing the check’s intent.
Now, working the check with a traditional jest command will fail as a result of the
codemod isn’t written but.

The corresponding destructive case would make sure the code stays unchanged
for different characteristic toggles:

defineInlineTest(
  remodel,
  {},
  `
  const information = featureToggle('feature-search-result-refinement') ? { identify: 'Product' } : undefined;
  `,
  `
  const information = featureToggle('feature-search-result-refinement') ? { identify: 'Product' } : undefined;
  `,
  "don't change different characteristic toggles"
);

Writing the Codemod

Let’s begin by defining a easy remodel perform. Create a file
referred to as remodel.js with the next code construction:

module.exports = perform(fileInfo, api, choices) {
  const j = api.jscodeshift;
  const root = j(fileInfo.supply);

  // manipulate the tree nodes right here

  return root.toSource();
};

This perform reads the file right into a tree and makes use of jscodeshift’s API to
question, modify, and replace the nodes. Lastly, it converts the AST again to
supply code with .toSource().

Now we are able to begin implementing the remodel steps:

1. Discover all cases of featureToggle.
2. Confirm that the argument handed is 'feature-new-product-list'.
3. Change the complete conditional expression with the consequent half,
   successfully eradicating the toggle.

Right here’s how we obtain this utilizing jscodeshift:

module.exports = perform (fileInfo, api, choices) {
  const j = api.jscodeshift;
  const root = j(fileInfo.supply);

  // Discover ConditionalExpression the place the check is featureToggle('feature-new-product-list')
  root
    .discover(j.ConditionalExpression, {
      check: {
        callee: { identify: "featureToggle" },
        arguments: [{ value: "feature-new-product-list" }],
      },
    })
    .forEach((path) => {
      // Change the ConditionalExpression with the 'consequent'
      j(path).replaceWith(path.node.consequent);
    });

  return root.toSource();
};

The codemod above:

• Finds ConditionalExpression nodes the place the check calls
  featureToggle('feature-new-product-list').
• Replaces the complete conditional expression with the resultant (i.e., {
identify: 'Product' }), eradicating the toggle logic and leaving simplified code
  behind.

This instance demonstrates how straightforward it’s to create a helpful
transformation and apply it to a big codebase, considerably lowering
handbook effort.

You’ll want to write down extra check instances to deal with variations like
if-else statements, logical expressions (e.g.,
!featureToggle('feature-new-product-list')), and so forth to make the
codemod strong in real-world eventualities.

As soon as the codemod is prepared, you’ll be able to check it out on a goal codebase,
such because the one you are engaged on. jscodeshift supplies a command-line
instrument that you should use to use the codemod and report the outcomes.

$ jscodeshift -t transform-name src/

After validating the outcomes, test that each one purposeful exams nonetheless
move and that nothing breaks—even for those who’re introducing a breaking change.
As soon as glad, you’ll be able to commit the modifications and lift a pull request as
a part of your regular workflow.

Codemods Enhance Code High quality and Maintainability

Codemods aren’t simply helpful for managing breaking API modifications—they’ll
considerably enhance code high quality and maintainability. As codebases
evolve, they typically accumulate technical debt, together with outdated characteristic
toggles, deprecated strategies, or tightly coupled elements. Manually
refactoring these areas could be time-consuming and error-prone.

By automating refactoring duties, codemods assist maintain your codebase clear
and freed from legacy patterns. Repeatedly making use of codemods lets you
implement new coding requirements, take away unused code, and modernize your
codebase with out having to manually modify each file.

Refactoring an Avatar Part

Now, let’s have a look at a extra advanced instance. Suppose you’re working with
a design system that features an Avatar part tightly coupled with a
Tooltip. Every time a consumer passes a identify prop into the Avatar, it
robotically wraps the avatar with a tooltip.

Determine 3: A avatar part with a tooltip

Right here’s the present Avatar implementation:

import { Tooltip } from "@design-system/tooltip";

const Avatar = ({ identify, picture }: AvatarProps) => {
  if (identify) {
    return (
      <Tooltip content material={identify}>
        <CircleImage picture={picture} />
      </Tooltip>
    );
  }

  return <CircleImage picture={picture} />;
};

The aim is to decouple the Tooltip from the Avatar part,
giving builders extra flexibility. Builders ought to have the ability to determine
whether or not to wrap the Avatar in a Tooltip. Within the refactored model,
Avatar will merely render the picture, and customers can apply a Tooltip
manually if wanted.

Right here’s the refactored model of Avatar:

const Avatar = ({ picture }: AvatarProps) => {
  return <CircleImage picture={picture} />;
};

Now, customers can manually wrap the Avatar with a Tooltip as
wanted:

import { Tooltip } from "@design-system/tooltip";
import { Avatar } from "@design-system/avatar";

const UserProfile = () => {
  return (
    <Tooltip content material="Juntao Qiu">
      <Avatar picture="/juntao.qiu.avatar.png" />
    </Tooltip>
  );
};

The problem arises when there are a whole bunch of Avatar usages unfold
throughout the codebase. Manually refactoring every occasion could be extremely
inefficient, so we are able to use a codemod to automate this course of.

Utilizing instruments like AST Explorer, we are able to
examine the part and see which nodes symbolize the Avatar utilization
we’re concentrating on. An Avatar part with each identify and picture props
is parsed into an summary syntax tree as proven under:

Determine 4: AST of the Avatar part utilization

Writing the Codemod

Let’s break down the transformation into smaller duties:

• Discover Avatar utilization within the part tree.
• Examine if the identify prop is current.
• If not, do nothing.
• If current:
• Create a Tooltip node.
• Add the identify to the Tooltip.
• Take away the identify from Avatar.
• Add Avatar as a baby of the Tooltip.
• Change the unique Avatar node with the brand new Tooltip.

To start, we’ll discover all cases of Avatar (I’ll omit a few of the
exams, however it is best to write comparability exams first).

defineInlineTest(
    { default: remodel, parser: "tsx" },
    {},
    `
    <Avatar identify="Juntao Qiu" picture="/juntao.qiu.avatar.png" />
    `,
    `
    <Tooltip content material="Juntao Qiu">
      <Avatar picture="/juntao.qiu.avatar.png" />
    </Tooltip>
    `,
    "wrap avatar with tooltip when identify is supplied"
  );

Much like the featureToggle instance, we are able to use root.discover with
search standards to find all Avatar nodes:

root
  .discover(j.JSXElement, {
    openingElement: { identify: { identify: "Avatar" } },
  })
  .forEach((path) => {
    // now we are able to deal with every Avatar occasion
  });

Subsequent, we test if the identify prop is current:

root
  .discover(j.JSXElement, {
    openingElement: { identify: { identify: "Avatar" } },
  })
  .forEach((path) => {
    const avatarNode = path.node;

    const nameAttr = avatarNode.openingElement.attributes.discover(
      (attr) => attr.identify.identify === "identify"
    );

    if (nameAttr) {
      const tooltipElement = createTooltipElement(
        nameAttr.worth.worth,
        avatarNode
      );
      j(path).replaceWith(tooltipElement);
    }
  });

For the createTooltipElement perform, we use the
jscodeshift API to create a brand new JSX node, with the identify
prop utilized to the Tooltip and the Avatar
part as a baby. Lastly, we name replaceWith to
change the present path.

Right here’s a preview of the way it seems to be in
Hypermod, the place the codemod is written on
the left. The highest half on the correct is the unique code, and the underside
half is the reworked consequence:

Determine 5: Run checks inside hypermod earlier than apply it to your codebase

This codemod searches for all cases of Avatar. If a
identify prop is discovered, it removes the identify prop
from Avatar, wraps the Avatar inside a
Tooltip, and passes the identify prop to the
Tooltip.

By now, I hope it’s clear that codemods are extremely helpful and
that the workflow is intuitive, particularly for large-scale modifications the place
handbook updates could be an enormous burden. Nonetheless, that is not the entire
image. Within the subsequent part, I’ll make clear a few of the challenges
and the way we are able to tackle these less-than-ideal elements.

Leveraging Supply Graphs and Take a look at-Pushed Codemods

To deal with these complexities, codemods must be used alongside different
methods. As an illustration, just a few years in the past, I participated in a design
system elements rewrite undertaking at Atlassian. We addressed this concern by
first looking the supply graph, which contained nearly all of inner
part utilization. This allowed us to grasp how elements had been used,
whether or not they had been imported below totally different names, or whether or not sure
public props had been continuously used. After this search section, we wrote our
check instances upfront, making certain we lined nearly all of use instances, and
then developed the codemod.

In conditions the place we could not confidently automate the improve, we
inserted feedback or “TODOs” on the name websites. This allowed the
builders working the script to deal with particular instances manually. Often,
there have been solely a handful of such cases, so this method nonetheless proved
helpful for upgrading variations.

Using Current Code Standardization Instruments

As you’ll be able to see, there are many edge instances to deal with, particularly in
codebases past your management—reminiscent of exterior dependencies. This
complexity signifies that utilizing codemods requires cautious supervision and a
evaluate of the outcomes.

Nonetheless, in case your codebase has standardization instruments in place, reminiscent of a
linter that enforces a specific coding type, you’ll be able to leverage these
instruments to cut back edge instances. By implementing a constant construction, instruments
like linters assist slim down the variations in code, making the
transformation simpler and minimizing sudden points.

As an illustration, you possibly can use linting guidelines to limit sure patterns,
reminiscent of avoiding nested conditional (ternary) operators or implementing named
exports over default exports. These guidelines assist streamline the codebase,
making codemods extra predictable and efficient.

Moreover, breaking down advanced transformations into smaller, extra
manageable ones lets you sort out particular person points extra exactly. As
we’ll quickly see, composing smaller codemods could make dealing with advanced
modifications extra possible.

Codemod Composition

Let’s revisit the characteristic toggle removing instance mentioned earlier. Within the code snippet
we’ve a toggle referred to as feature-convert-new must be eliminated:

import { featureToggle } from "./utils/featureToggle";

const convertOld = (enter: string) => {
  return enter.toLowerCase();
};

const convertNew = (enter: string) => {
  return enter.toUpperCase();
};

const consequence = featureToggle("feature-convert-new")
  ? convertNew("Howdy, world")
  : convertOld("Howdy, world");

console.log(consequence);

The codemod for take away a given toggle works effective, and after working the codemod,
we wish the supply to seem like this:

const convertNew = (enter: string) => {
  return enter.toUpperCase();
};

const consequence = convertNew("Howdy, world");

console.log(consequence);

Nonetheless, past eradicating the characteristic toggle logic, there are further duties to
deal with:

• Take away the unused convertOld perform.
• Clear up the unused featureToggle import.

After all, you possibly can write one huge codemod to deal with every little thing in a
single move and check it collectively. Nonetheless, a extra maintainable method is
to deal with codemod logic like product code: break the duty into smaller,
unbiased items—similar to how you’d usually refactor manufacturing
code.

Breaking It Down

We are able to break the large transformation down into smaller codemods and
compose them. The benefit of this method is that every transformation
could be examined individually, protecting totally different instances with out interference.
Furthermore, it lets you reuse and compose them for various
functions.

As an illustration, you would possibly break it down like this:

• A metamorphosis to take away a selected characteristic toggle.
• One other transformation to scrub up unused imports.
• A metamorphosis to take away unused perform declarations.

By composing these, you’ll be able to create a pipeline of transformations:

import { removeFeatureToggle } from "./remove-feature-toggle";
import { removeUnusedImport } from "./remove-unused-import";
import { removeUnusedFunction } from "./remove-unused-function";

import { createTransformer } from "./utils";

const removeFeatureConvertNew = removeFeatureToggle("feature-convert-new");

const remodel = createTransformer([
  removeFeatureConvertNew,
  removeUnusedImport,
  removeUnusedFunction,
]);

export default remodel;

On this pipeline, the transformations work as follows:

1. Take away the feature-convert-new toggle.
2. Clear up the unused import assertion.
3. Take away the convertOld perform because it’s not used.

Determine 6: Compose transforms into a brand new remodel

You may also extract further codemods as wanted, combining them in
varied orders relying on the specified final result.

Determine 7: Put totally different transforms right into a pipepline to kind one other remodel

The createTransformer Operate

The implementation of the createTransformer perform is comparatively
simple. It acts as a higher-order perform that takes an inventory of
smaller remodel features, iterates by way of the checklist to use them to
the foundation AST, and at last converts the modified AST again into supply
code.

import { API, Assortment, FileInfo, JSCodeshift, Choices } from "jscodeshift";

kind TransformFunction = { (j: JSCodeshift, root: Assortment): void };

const createTransformer =
  (transforms: TransformFunction[]) =>
  (fileInfo: FileInfo, api: API, choices: Choices) => {
    const j = api.jscodeshift;
    const root = j(fileInfo.supply);

    transforms.forEach((remodel) => remodel(j, root));
    return root.toSource(choices.printOptions || { quote: "single" });
  };

export { createTransformer };

For instance, you possibly can have a remodel perform that inlines
expressions assigning the characteristic toggle name to a variable, so in later
transforms you don’t have to fret about these instances anymore:

const shouldEnableNewFeature = featureToggle('feature-convert-new');

if (!shouldEnableNewFeature && someOtherLogic) {
  //...
}

Turns into this:

if (!featureToggle('feature-convert-new') && someOtherLogic) {
  //...
}

Over time, you would possibly construct up a group of reusable, smaller
transforms, which might enormously ease the method of dealing with tough edge
instances. This method proved extremely efficient in our work refining design
system elements. As soon as we transformed one bundle—such because the button
part—we had just a few reusable transforms outlined, like including feedback
in the beginning of features, eradicating deprecated props, or creating aliases
when a bundle is already imported above.

Every of those smaller transforms could be examined and used independently
or mixed for extra advanced transformations, which quickens subsequent
conversions considerably. Because of this, our refinement work turned extra
environment friendly, and these generic codemods are actually relevant to different inner
and even exterior React codebases.

Since every remodel is comparatively standalone, you’ll be able to fine-tune them
with out affecting different transforms or the extra advanced, composed ones. For
occasion, you would possibly re-implement a remodel to enhance efficiency—like
lowering the variety of node-finding rounds—and with complete check
protection, you are able to do this confidently and safely.