[selectors] Rejigger the absolutizing algorithm to be more correct.

--HG--
extra : rebase_source : 20ac2350bed9ed3383c41d0cc0d59a4753d15024

Author: tabatkins

selectors/Overview.html

@@ -209,7 +209,7 @@ <h2 class="no-num no-toc" id=contents>Table of Contents</h2>
       Selectors</a>
       <ul class=toc>
        <li><a href="#absolutizing"><span class=secno>3.4.1. </span>
-        Absolutizing a Scope-relative Selector</a>
+        Absolutizing a Relative Selector</a>
       </ul>
 
      <li><a href="#pseudo-classes"><span class=secno>3.5. </span>
@@ -1407,12 +1407,12 @@ <h3 id=relative><span class=secno>3.4. </span> Relative Selectors</h3>
    href="#scope-filtered"><i>scope-filtered</i></a>.
 
   <h4 id=absolutizing><span class=secno>3.4.1. </span> Absolutizing a
-   Scope-relative Selector</h4>
+   Relative Selector</h4>
 
   <p> To <dfn id=absolutize>absolutize a relative selector</dfn>:
 
   <p> If the <a href="#reference-element-set"><i>reference element
-   set</i></a> is non-empty,
+   set</i></a> is empty or consists of one or more elements:
 
   <ol>
    <li> If the selector starts with a <a
@@ -1421,9 +1421,6 @@ <h4 id=absolutizing><span class=secno>3.4.1. </span> Absolutizing a
     <a href="#scope-pseudo"><code>:scope</code></a> as the initial <a
     href="#compound"><i>compound selector</i></a>.
 
-   <li> Otherwise, if the <a href="#reference-element-set"><i>reference
-    element set</i></a> is empty, the selector is already absolute.
-
    <li> Otherwise, if the selector does not contain any instance of the <a
     href="#scope-pseudo"><code>:scope</code></a> pseudo-class (either at the
     top-level or as an argument to a functional pseudo-class), prepend <a
@@ -1433,22 +1430,37 @@ <h4 id=absolutizing><span class=secno>3.4.1. </span> Absolutizing a
    <li> Otherwise, the selector is already absolute.
   </ol>
 
+  <p class=issue> The above algorithm is different from Selectors API: its
+   step 2, which bailed if the reference element set was empty, has been
+   removed. This is because handling an explicitly empty list should return
+   nothing; if a reference element set was not given at all, Selectors API 2
+   should be amended to say that the selector must be absolute already.
+   (Broken example: doing <code>document.find('img', [list-of-links]))</code>
+   finds images that are descendants of links normally, but instead finds all
+   images in the document if the list-of-links is empty.)
+
   <p> Otherwise, if the <a href="#reference-element-set"><i>reference element
-   set</i></a> is empty,
+   set</i></a> consists of only a <a href="#virtual"><i>virtual scoping
+   root</i></a>:
 
   <ol>
    <li> If the selector starts with a <a href="#child-combinator"><i>child
     combinator</i></a>, remove the child combinator. The selector is now
     absolute, with the additional constraint that the first compound selector
     in the selector only matches elements without a parent.
 
-   <li> If the selectors starts with any other combinator other than the <a
-    href="#descendant-combinator"><i>descendant combinator</i></a>, the
-    selector represents nothing.
+   <li> Otherwise, if the selector starts with any combinator other than the
+    <a href="#descendant-combinator"><i>descendant combinator</i></a>, change
+    the selector to ‘<a href="#negation-pseudo"><code
+    class=css>:not(*)</code></a>’. <span class=note>This is the shortest
+    selector that is valid, but guaranteed to match nothing.</span>
 
    <li> Otherwise, the selector is already absolute.
   </ol>
 
+  <p> Otherwise, this is a spec error. Please report it to the relevant
+   standards body.
+
   <p> To <dfn id=absolutize-list>absolutize a relative selector list</dfn>,
    absolutize each relative selector in the list.
 

selectors/Overview.src.html

@@ -700,23 +700,19 @@ <h3 id="relative">
     or <i>scope-filtered</i>.
 
 <h4 id='absolutizing'>
-Absolutizing a Scope-relative Selector</h4>
+Absolutizing a Relative Selector</h4>
 
   <p>
     To <dfn id='absolutize'>absolutize a relative selector</dfn>:
 
   <p>
-    If the <i>reference element set</i> is non-empty,
+    If the <i>reference element set</i> is empty or consists of one or more elements:
 
   <ol>
     <li>
       If the selector starts with a <i>combinator</i> other than the <i>descendant combinator</i>,
       prepend <code>:scope</code> as the initial <i>compound selector</i>.
 
-    <li>
-      Otherwise, if the <i>reference element set</i> is empty,
-      the selector is already absolute.
-
     <li>
       Otherwise, if the selector does not contain any instance of the <code>:scope</code> pseudo-class
       (either at the top-level or as an argument to a functional pseudo-class),
@@ -726,8 +722,16 @@ <h4 id='absolutizing'>
       Otherwise, the selector is already absolute.
   </ol>
 
+  <p class='issue'>
+    The above algorithm is different from Selectors API: its step 2, which bailed if the reference element set was empty, has been removed.
+    This is because handling an explicitly empty list should return nothing;
+    if a reference element set was not given at all,
+    Selectors API 2 should be amended to say that the selector must be absolute already.
+    (Broken example: doing <code>document.find('img', [list-of-links]))</code> finds images that are descendants of links normally,
+    but instead finds all images in the document if the list-of-links is empty.)
+
   <p>
-    Otherwise, if the <i>reference element set</i> is empty,
+    Otherwise, if the <i>reference element set</i> consists of only a <i>virtual scoping root</i>:
 
   <ol>
     <li>
@@ -738,13 +742,17 @@ <h4 id='absolutizing'>
       only matches elements without a parent.
 
     <li>
-      If the selectors starts with any other combinator other than the <i>descendant combinator</i>,
-      the selector represents nothing.
+      Otherwise, if the selector starts with any combinator other than the <i>descendant combinator</i>,
+      change the selector to '':not(*)''.
+      <span class='note'>This is the shortest selector that is valid, but guaranteed to match nothing.</span>
 
     <li>
       Otherwise, the selector is already absolute.
   </ol>
 
+  <p>
+    Otherwise, this is a spec error.  Please report it to the relevant standards body.
+
   <p>
     To <dfn id='absolutize-list'>absolutize a relative selector list</dfn>,
     absolutize each relative selector in the list.