为什么通用兄弟组合器允许切换伪元素的内容，而不是相邻的兄弟？

Question

In this question " CSS3 Selector That Works like jQuery's .click()? " I posted an answer using the :checked state of an input , of type="checkbox" to toggle the display of an element.

This is the HTML of the demo I posted in that answer:

<input type="checkbox" id="switch" />
<nav>
    <h2>This would be the 'navigation' element.</h2>
</nav>
<label for="switch">Toggle navigation</label>

And the CSS (with transitions stripped for brevity):

#switch {
    display: none;
}
#switch + nav {
    height: 0;
    overflow: hidden;
    /* transitions followed */
}
#switch:checked + nav {
    height: 4em;
    color: #000;
    background-color: #ffa;
    /* transitions followed */
}

label {
    cursor: pointer;
}

JS Fiddle demo .

Once I'd posted the answer it occurred to me that we could also toggle the text of the label used to trigger the state-change of that checkbox, using the following selectors (having amended the label 's text to 'navigation'):

label {
    display: inline-block;
    cursor: pointer;
}

#switch + nav + label::before {
    content: 'Show ';
}

#switch:checked + nav + label::before {
    content: 'Hide ';
}

Simplified/basic JS Fiddle demo .

This did not work, in that while the selector matched while the input was in its unchecked state (and the label showed Show navigation ), the selector failed to match when the state of the input changed. Note that the transitions were still effected on the nav element, and the original matching selector indicates that the next-sibling combinator matched originally. The above link shows a simplified demo of the not-working (in Chrome 27/Windows XP) selectors.

It then occurred to me to try the general-sibling combinator, to reduce the selector-chain. which resulted in the following CSS (with transitions again stripped for brevity):

#switch:checked + nav {
    background-color: #ffa;
}

label {
    display: inline-block;
    cursor: pointer;
}

#switch ~ label::before {
    content: 'Show ';
}

#switch:checked ~ label::before {
    content: 'Hide ';
}

JS Fiddle demo .

Somewhat to my surprise, this worked (the content of the label changed in response to the changed-state of the input ).

So, the question: why does the general-sibling combinator allow for updating of a later-sibling while chained next-sibling combinators (which match the elements and the structure of the DOM) does not?

Further, this does seem to work in Firefox (21, on Windows XP); so I guess the question is altered slightly to include: is this a bug in Chrome/Webkit, or an expected behaviour?

And, even further , it seems that while this is a bug in Chrome (thanks @Boltclock), there's a somewhat ludicrous 'do-nothing' animation that fixes the non-working demo (though other, perhaps better, alternatives exist, as Scott's answer shows):

body {
    -webkit-animation: bugfix infinite 1s;
}
@-webkit-keyframes bugfix {
    from {
        padding: 0;
    }
    to {
        padding: 0;
    }
}
#switch {
}
#switch + nav {
    -moz-transition: all 1s linear;
    -ms-transition: all 1s linear;
    -o-transition: all 1s linear;
    -webkit-transition: all 1s linear;
    transition: all 1s linear;
}
#switch:checked + nav {
    background-color: #ffa;
    -moz-transition: all 1s linear;
    -ms-transition: all 1s linear;
    -o-transition: all 1s linear;
    -webkit-transition: all 1s linear;
    transition: all 1s linear;
}
label {
    display: inline-block;
    cursor: pointer;
}
#switch + nav + label::before {
    content:'Show ';
}
#switch:checked + nav + label::before {
    content:'Hide ';
}

JS Fiddle demo .

Note: the reason I'm updating the question with this 'fix,' rather than posting it as an answer, is simply because the question wasn't "how can I fix this?" but (basically) "why doesn't it work?"

Answer 1

Bug Work Around

Apparently, certain valid pseudo-classes chained together allows it to work.

These work (see Fiddle #1 , Fiddle #2 , Fiddle #3 ):

#switch:checked + nav:only-of-type + label::before
#switch:checked + nav:nth-of-type(1) + label::before
#switch:checked + nav:nth-child(2) + label::before

This did not (see Fiddle #4 ):

#switch:checked + nav:not([class]) + label::before

I tried some other :not() combinations, none of which allowed it to work.

** Best choice **

#switch:checked + nav:nth-child(n) + label::before

Answer 2

This is a long-standing bug in WebKit browsers related to the use of certain dynamic pseudo-classes with next-sibling combinators. This happens whether you're applying styles to the sibling element itself or a pseudo-element of that sibling element.

I don't know if anybody has filed a bug report yet, but this has been seen rather frequently on the site:

• Webkit bug with `:hover` and multiple adjacent-sibling selectors
• CSS adjacent sibling selectors, Safari and <nav> elements

Strangely it was also reported that Chrome had issues with the general sibling combinator, but as you note it works in your given scenario:

• Why doesn't this CSS selector work: a:hover ~ span?

So either that was fixed, or something else triggers/triggered it.