Concatenative selectors

[keithamus]

Hi @jonathantneal, thanks for making such a cool postcss plugin!

I feel like I've found a potential bug with this. The spec doesn't mention this behaviour specifically, but I'm quite sure it is undesired.

Input:

.foo {
  color: blue;
  &bar {
    color: red;
  }
}

This plugins output:

.foo {
  color: blue
}
.foobar {
  color: red;
}

I would expect to see an error, I think, as the spec suggests & is equivalent of :matches() - which, to my understanding, means they shouldn't concatenate like this.

Perhaps @tabatkins could clarify this?

[tabatkins]

It's wrong, but only because type selectors don't have an identifying prefix. For all other selectors, you can just naively concatenate them - .foo { &.bar {...}} goes to just .foo.bar {...}, for example. Same for cases where the & is immediately followed by a combinator - naive concatenation works just fine.

So this fails only in two cases:

1. The inner selector puts the & at the beginning of a compound selector whose non-& part starts with a type selector, like .foo { &div {...}}.
2. The inner selector puts the & inside of the compound selector, where the parent selector starts with a type selector, like div { .foo&.bar {...}}.

In either case, you just have to do some minor rewriting to make sure the type selector goes first - the first case should result in div.foo, the second in div.foo.bar.

as the spec suggests & is equivalent of :matches()

You're misreading a bit - if the parent selector uses commas (it's a selector list of length >1), you have to enclose it in a :matches() before substituting it into the child, if you're directly synthesizing an equivalent selector. So a, b { & c {...}} becomes :matches(a, b) c {...}. Same if the parent selector has >1 compound selectors and & isn't in the very first compound selector - a b { c & {...}} becomes c :matches(a b) {...}.

So & isn't equivalent to :matches(), but constructing equivalent selectors without & often requires :matches().

[taion]

Is it possible to add some sort of check for this? Right now this is prone to accidental use for cases like &_title, which is probably not great.

[tbremer]

@taion, that's a good point. Could be a pretty quick RegEx check, using something like:

/^&(?: *[a-zA-Z\.#:>\[*+~]|(?:\|\|))/

However, there are pieces that (I think, maybe @tabatkins can confirm) this plugin and the spec diverge, the plugin allows for:

a {
    color: blue;
    & & {
        color: orange;
    }
    & b & {
        color: red;
    }
}

which compiles to:

a {
    color: blue;
}

a a {
    color: orange;
}

a b a {
    color: red;
}

But the Spec doesn't treat the & as selector replacement.

[tbremer]

The @nest rules are a little less terse, so checking validity would have to be thought out a little more, however the spec does call this out specifically as invalid:

.foo {
  color: red;
  @nest & .bar, .baz {
    color: blue;
  }
}
/* Invalid because not all selectors in the list
   contain a nesting selector */

But the plugin compiles it to:

.foo {
    color: red;
}

.foo .bar, .baz {
    color: blue;
}

[tabatkins]

However, there are pieces that (I think, maybe @tabatkins can confirm) this plugin and the spec diverge, the plugin allows for:

Correct, the spec treats & as a real selector that refers to a specific element (whatever's matched by the parent rule), so & & never matches anything; an element can't be its own descendant.

[taion]

FWIW, I'm a little bit less worried about those edge cases – more as a defensive thing, preventing people from using the nesting syntax for concatenation in "obviously wrong" cases.

[jonathantneal]

It seems like this plugin cannot emulate what you are saying the specification defines.

For instance, are we saying that class selectors cannot be extended? I imagine almost anyone writing BEM-like CSS is probably using this plugin (alone or within cssnext) to do things like &-button, etc.

[taion]

👍 to that – I can't imagine many people are writing & &, selectors, but supporting something like &-button that people might actually use incorrectly would be bad.

[tbremer]

Yeah, I believe this (see below) is invalid as the spec is written.

.button {
    &__primary {
        &--toggled {}
    }
}

I think @taion brings up a really valid point though if this plugin is to respect the spec then it should conform. Obviously the concern is that this type of change would be a major breaking type of change.

[jonathantneal]

Okay, I can update the plugin to effectively work more like :matches, allowing .foo { &div {} to be like .foo:matches(div) or div.foo. I can dismiss or warn out any selector concatenating more than one type selector, causing div { &span {} to break like div:matches(span) . Then, to work around this newly understood meaning of &, I can dismiss or warn out any selector using more than one &. Do I understand the spec a bit better? Would this make the plugin spec compliant?

This would definitely be a major release change, and, sadly, I’m certain many developers will, at some point, blindly pull the major release and discover that all of their BEM concatenations are broken.

[tbremer]

There could be a "loose" option that allows selector concatenation.

[oalberdi]

Missing semicolon on nesting selectors, as shown on the first example. It´s not a rendering issue but would be great to add or have it.

Input

.foo {
color: blue;
&bar {
color: red;
}
}

Output

.foo {
color: blue
}
.foobar {
color: red;
}

[jquense]

bit confused does this mean that & + & is also invalid per the spec, its certainly allowed by the plugin?

[tabatkins]

It's not invalid in the spec, but it's meaningless and will never match anything, because & actually refers to an element, it's not just an alias for some selector text. An element can't be a sibling of itself.

[jquense]

Hmm OK, so it's that this isn't actually polyfillable. If you don't mind me asking why not have it be the selector? that approach seems more useful am I just looking at it the wrong way?