No useless expressions

[dcousens]

I recently had something bite me in some of my own code, and I wonder if we can't avoid this happening:

var o = {}

// ...

if (o.value === undefined) {
  o.value === 3
}

That should have been o.value = 3, but, alas, something went wrong that day.
Is there an existing eslint rule for this?
Any reason not to have it enabled?

[feross]

I would open an issue on ESLint about adding a rule for this, but I don't think it's possible for them to reliably detect that o.value === 3 is useless, since o.value could be a getter that mutates global state. Though, that's a terrible idea. :)

[dcousens]

eslint/eslint#7642

[feross]

Nice find, going to re-open until we've released v9.

http://eslint.org/docs/rules/no-unused-expressions

[feross]

This rule has a lot of false positives with silly testing frameworks that do: expect(err).to.exist and other pointless sugary APIs.

# tests 422
# pass  401
# fail  21

People need to stop being so clever. All this fancy stuff just causes more problems than it solves.

How is this:

expect(err.message).to.equal('You had one job')

better than this?

assert.equal(err.message, 'You had one job')

Grrr...

[dcousens]

Oh dear, does expect(err).to.exist really happen?
The grammar-like chaining (to.equal()) could make sense I suppose, and at least doesn't break any odd rules... but please not that ^

[feross]

And guess what happens when you write expect(err).to.exists (note the extra s). It's undefined, so it silently does nothing.

[LinusU]

We could add this rule and encourage users to move away from that horrible api though?

[feross]

Yeah, maybe we should just ship this. I don't want to scare too many people off of standard, but honestly, people need to use better modules.

[dcousens]

What were your final thoughts @feross ?

[feross]

I think we should ship this in the beta for v9 and see how it goes.

[dcousens]

@feross, I added this to the v9 milestone

[feross]

Question: This rule has two options that relax parts of the rule. Should we enable them?

Without allowShortCircuit enabled, this is an error:

trySelectWire(false) || trySelectWire(true)

Without allowTernary enabled, this is an error:

isWebSeed ? piece.cancelRemaining(reservation) : piece.cancel(reservation)

More examples of both:

!valid ? reject(new HARError(validate.errors)) : resolve(data)

opts.alwaysOnTop ? emitBlur() : hideWindow()

oldRelease && oldRelease()

I would argue that all these should be if statements, since those are less clever. But if it's going to cause too much pain upgrading to v9, better to just allow these cases for now, so we can get the rest of the benefit of this rule.

[timoxley]

My vote is to enable, I use both of those patterns often. If I had to pick only one to enable I'd pick the ternary. I've flip-flopped on preference for if vs && a few times, no strong feelings either way, though I do agree that using if is a little less "clever". I think ecosystem impact will be revealing.

[feross]

With no exceptions:

1..422
# tests 422
# pass  396
# fail  26

With both exceptions:

1..422
# tests 422
# pass  408
# fail  14

Of the 14 failures:

• silly test framework (11)
• bugs (3)

Here are the bugs:

if (Array.isArray(objects)) {
  return objects.some(function (object) { object.type && object.type !== type })
}

if (/use strict/.test(line)) line += '\n' + preamble
else line + preamble + '\n' + line

self.resetTimer

[timoxley]

@feross anything interesting in the 12 that only passed with both exceptions enabled?

[feross]

@timoxley No -- the 3 examples in my previous comment were representative (#690 (comment)).

[timoxley]

@feross i.e. no bugs getting through due to relaxed rule is good

[feross]

For users of Chai, here's an explanation of why Chai's behavior is unsafe and an anti-pattern: Beware of libraries that assert on property access.

Fortunately, there is a workaround that doesn't require giving up Chai. Install dirty-chai, a project that "Extends Chai with lint-friendly terminating assertions."

In short, your tests will go from:

expect(true).to.be.true
foo.should.be.ok

To this:

expect(true).to.be.true()
foo.should.be.ok()

And just like that, your tests will be much more robust! (Many thanks to @erinspice for this tip.)

---

If you cannot change your tests now, feel free to continue using standard v8 until you're ready to upgrade.

If you wish to upgrade and temporarily disable this rule, add a comment to the top of each test file:

/* eslint-disable no-unused-expressions */

[dcousens]

if (Array.isArray(objects)) {
  return objects.some(function (object) { object.type && object.type !== type })
}

Haha, this gets me way too often.

Looking forward to it.

[feross]

This will be part of standard v9. I enabled both exceptions (short circuiting and ternary expressions), so it should only catch bugs and unsafe test frameworks.

[billiegoose]

I love that one of the main decision factors for standard style is preventing buggy code. <3

[bcoe]

this rule breaks one of the common approaches that folks use when invoking yargs:

 yargs.argv

this would require that I need to update 100s of tests to continue using standard:

 /Users/benjamincoe/bcoe/yargs/test/command.js:33:7: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:78:7: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:104:7: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:129:7: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:509:7: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:527:7: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:549:7: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:586:7: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:602:7: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:619:7: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:713:5: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:719:9: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:729:5: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:792:9: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:835:5: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:850:7: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:873:9: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:884:9: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:896:9: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:907:9: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:918:9: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:930:9: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:944:9: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:955:9: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:972:13: Expected an assignment or function call and instead saw an expression.
  /Users/benjamincoe/bcoe/yargs/test/command.js:974:9: Expected an assignment or function call and instead saw an expression.

I'm less concerned about this, and more concerned that I'd need to update all of the existing documentation -- to point folks towards code that plays nice with standard.

I'm a huge advocate of standard, and have been using it for everything I write for the past couple years, but this rule specifically is making me a bit gun shy about upgrading.

[dcousens]

@bcoe, but, you understand that yargs.argv, on its own, is completely weird... right?

[bcoe]

@dcousens .argv is the convention we've been using in optimist and yargs for years at this point, to indicate that we're done building up configuration, and the parser should run, e.g.,

#!/usr/bin/env node
var argv = require('yargs')
    .count('verbose')
    .alias('v', 'verbose')
    .argv

Just had yargs.argv as a minimal example -- I think there are probably literally thousands of repos using this approach in the wild, because it's what we've always documented in our examples.

[dcousens]

If assigned as above, there is no issue.
It is the abuse of property side effects that is being discouraged; as it is completely unclear.

For standard users, you could promote a different approach, or, just add an ESLint ignore?

[bcoe]

@dcousens this comes up a lot when you're writing a CLI using yargs, e.g.:

require('yargs') // eslint-disable-line
  .usage('$0 <command>')
  .commandDir('../lib/commands')
  .help()
  .alias('h', 'help')
  .version()
  .strict()
  .demandCommand(1)
  .argv

or,

require('yargs') // eslint-disable-line
  .command(['serve', '*'], 'start the pkg-info-cache server', () => {}, (argv) => {
    const logger = bole(serviceName)
    bootstrapLogger()
    logger.info(argv)
    server(argv)
  })
  .option('port', {
    describe: 'port to run server on',
    default: process.env.PORT || 9999
  })
  .option('host', {
    describe: 'host to bind on',
    default: '0.0.0.0'
  })
  .argv

because yargs handles the flow control; one of the first things I do now when creating a bin in a project is to cargo cult the // eslint-disable-line this feels suboptimal though.

[knksmith57]

@bcoe fwiw just got bit hard by this; my google foo wasn't strong enough to find this thread (and not for lack of trying).

resorting to eslint-config-standard w/ all this boilerplate was the best I could do without forcing inline comments: https://gist.github.com/knksmith57/e433fa7416e8530427a4054cfa22b5d3

also a huge proponent of standard; literally forced it on every project I've ever had influence over.

have to say tho, requiring all that ^^ just to prevent a very idiomatic pattern from failing the build w/out becoming intrusive on teammates leaves a bad taste in my mouth. for all the time and debate standard has saved (and aims to save!) devs, this one was a big step back.

hope y'all reconsider your position on this one.

/cc @feross

[LinusU]

As of Yargs 9.0.0 this is no longer a problem 🎉 👏

Instead of using the getter .argv, you make a call to .parse().

  require('yargs')
    .command('hello <name>', 'say hello to <name>', {}, yargs => {
      console.log(`hello, ${yargs.name}!`)
    })
    .demandCommand(1)
    .help()
-   .argv
+   .parse()

/cc @knksmith57

[searls]

1. Add Cypress to my project
2. Get a boatload of these errors from the example tests it ships with
3. Google to see what's up, find this thread
4. Remember, "oh, right, chai is a thing and this is a bad API"
5. Delete example tests

This is the system working, I think