avajs / ava-codemods Goto Github PK

From avajs/ava#644

@spudly has a start here: https://astexplorer.net/#/Uwl4f415pa

For completeness, here's a list of everything that I think would be needed for a complete mocha codemod:

• transform it() to test()
• add t parameter to a ArrowFunctionExpression passed to test()
• don't de-sync async functions (the codemod above will mess those up currently)
• add t parameter to a FunctionExpression passed to test()
• transform beforeEach() to test.beforeEach()
• transform afterEach() to test.afterEach()
• transform after() to test.after()
• transform before() to test.before()
• remove describe() blocks, prepending their text to each test block's description
• intelligently handle nested describe blocks if possible. how do we handle nested describe blocks with their own lifecycle hooks?
• if the test body FunctionExpression or ArrowFunctionExpression is async, it should use test.serial instead of test

Also, it would be nice to have a separate codemod to transform chai assertions to ava assertions.

@spudly - avajs/ava#644 (comment)

• Transform tests containing a callback ( it('foo', function (done) { ... }) ) to test.cb(t => ... t.end();

@jfmengels - avajs/ava#644 (comment)

from avajs/ava#644

We use jasmine extensively here at Redfin, and we're starting the long and tedious process of migrating to ava to simplify our babelification / es6, and to get our tests to run faster. It would significantly speed up our migration if we could use a codemod.

avajs/ava#644

It looks like #24 didn't add account for how the cli would be consumed by users doing npm install -g ava-codemods.

That PR was commited to master a month after the last release, so the version in the npm registry is still ok. But, releasing a new version from master would not work in it's current state.

How would you like me to address this @sindresorhus? Options being:

1. Revert changes to use ES6
2. Compile in prepublish to a dist directory and ignore src files for publish
3. Run babel-register when users run the CLI

3 doesn't seem like the best idea for DX.

Follow-up to #28.

Commented by @skovhus:

• if t.end is in the scope of the test function, we could remove it and not use cb style
• rename first param in test callback function, if it is not t (and replace usage of identifier in block). Note: currently we just fail with a nice error message instead.

My observations:

• Would be nice to convert some of the syntax to ES2015 too. Might be able to use an existing transform for that.

If you've tried it out and have some feedback. Comment below.

For codemods like same => deepEquals, we should just examine the AVA config in package.json to determine which files are test files, which are sources, etc.

We could do similar things for mocha/tap/etc by parsing their configs. (not sure we want to go down that road).

The jscodeshift CLI automatically loads babel/register (it's still on babel@5).

No reason we shouldn't be able to do the same for our tests, etc.

There exist a conversion from AVA to Jest, so would be nice to have the reverse too.

@skovhus Interested? Asking since you have some experience with the Jest API now.

Side question: I wonder if it would make more sense to have a separate project for each conversion? Not a big fan of bundling many binaries in one module.

We haven't maintained these codemods at all. This tool also tries to do a lot more than is necessary. For the latest throwsAsync change I wrote a simple transform that can be applied by using jscodeshift directly: https://gist.github.com/novemberborn/c2cdc94020083a1cafe3f41e8276f983

Perhaps we should instead host such transform files and explain how to use them, rather than trying to maintain a separate tool?

@sindresorhus what do you think?

Right now the CLI requires a clean git working directory. The only issue with that is, if you've just done npm i -D ava@latest, your package.json is modified, meaning you need to either commit the bump in package.json first, or use the --force flag every time.

IMO a --force flag I'm deciding to use every time is kinda stupid. So I vote we make an exception for package.json.

I just tried using tape-to-ava and got this error on terminal.

➜  serializeJSON tape-to-ava serializeJSON.test.js -f
fatal: Not a git repository (or any of the parent directories): .git
WARNING: Unable to determine if git directory is clean. Forcibly continuing. Ensure you have a backup of your tests or commit the latest changes before continuing.
Executing command: jscodeshift -t /usr/local/lib/node_modules/ava-codemods/lib/tape.js serializeJSON.test.js
Processing 1 files...
Spawning 1 workers...
Sending 1 files to free worker...
/usr/local/lib/node_modules/ava-codemods/lib/tape.js:115
export default function tapeToAva(fileInfo, api) {
^^^^^^
SyntaxError: Unexpected token export
    at Object.exports.runInThisContext (vm.js:78:16)
    at Module._compile (module.js:543:28)
    at Module._extensions..js (module.js:580:10)
    at Object.require.extensions.(anonymous function) [as .js] (/usr/local/lib/node_modules/ava-codemods/node_modules/babel-register/lib/node.js:152:7)
    at Module.load (module.js:488:32)
    at tryModuleLoad (module.js:447:12)
    at Function.Module._load (module.js:439:3)
    at Module.require (module.js:498:17)
    at require (internal/module.js:20:19)
    at setup (/usr/local/lib/node_modules/ava-codemods/node_modules/jscodeshift/dist/Worker.js:67:18)
All done.
Results:
0 errors
0 unmodified
0 skipped
0 ok
Time elapsed: 2.567seconds

I'm running on node 7.2.1.

We need to have docs in place in time for the next AVA release so users can easily codemod their tests.

I want to reuse the command in many projects and don't want to go through the interactive interface each time.

Automatically detect which test framework a user is currently using.

Scanning the require statements should work for tap and tape.

Scanning for describe / it, etc. Should work for mocha.

When using ava-codemods on:

test(t => {
    t.same(fn.sync('1.tmp', {cwd: t.context.tmp}), [path.join(t.context.tmp, '1.tmp')]);
});

I get:

test(t => {
    t.deepEqual(
        fn.sync('1.tmp', {cwd: t.context.tmp}),
        [path.join(t.context.tmp, '1.tmp')]
    );
});

It should not change unrelated code.

And it doesn't even use the correct indentation. Source was tab indented, now it's 4 spaces.

We're gonna change the CWD in tests to the package root: avajs/ava#32

Should we have a codemod that ensures there's a {cwd: __dirname} in childProcess and not relative paths in fs methods?