Issue 6034 - :-abp-contains() accept a regular expression

lib/content/elemHideEmulation.js

Index: lib/content/elemHideEmulation.js
===================================================================
--- a/lib/content/elemHideEmulation.js
+++ b/lib/content/elemHideEmulation.js
@@ -117,16 +117,34 @@
   }
   styles.sort();
   return {
     style: styles.join(" "),
     subSelectors: splitSelector(rule.selectorText)
   };
 }
 
+/**
+ * Check if a text argument is a regexp.
+ * @param {string} text argument.

[Manish Jethani, 2018/01/09 07:53:02]

This should be:

  @param {string} text the text argument

Right?

[hub, 2018/01/09 18:33:06]

Done.

+ * @return {?string} an unescaped RegExp string. null if it is not a
+ * regexp, ie not surrounded by '/'.
+ */
+function checkRegExpParameter(text)
+{
+  let regexpString = null;
+  if (text.length >= 2 && text[0] == "/" &&
+      text[text.length - 1] == "/")
+  {
+    regexpString = text.slice(1, -1)
+      .replace("\\7B ", "{").replace("\\7D ", "}");

[Manish Jethani, 2018/01/09 07:53:02]

Why is this needed? If it's for legacy reasons, maybe we can drop it now for
-abp-contains

[hub, 2018/01/09 18:33:06]

We still want to have a valid CSS selector syntax.

[Manish Jethani, 2018/01/10 11:22:14]

Acknowledged.

[Manish Jethani, 2018/01/16 12:00:22]

Why does this have to be valid CSS selector syntax, since we're not actually
generating a CSS selector here? Unlike -abp-has (which does take a selector),
this only takes a string.

Imagine CSS had a contains pseudo-class [1]:

  div:contains("Hello {}") {
    display: none !important;
  }

Even if the quotes were optional, the parser should have no trouble parsing the
"{" correctly. The escaping is only required for CSS identifiers, not for plain
strings. [2]

If we do support escape sequences, we should support them properly [3], though
like I said it shouldn't be required in this case. We can simply accept the
string as a JavaScript regular expression without any CSS escaping rules applied
to it.

For what it's worth, a backslash in a regular expression would have to be
escaped with another backslash, so we would be looking at "\\\\7B " rather than
"\\7B "

[1]:
https://www.w3.org/Style/CSS/Test/CSS3/Selectors/20011105/html/full/flat/css3...
[2]: https://www.w3.org/TR/css3-values/#string-value
[3]: https://mathiasbynens.be/notes/css-escapes

+  }
+  return regexpString;
+}
+
 function* evaluate(chain, index, prefix, subtree, styles)
 {
   if (index >= chain.length)
   {
     yield prefix;
     return;
   }
   for (let [selector, element] of
@@ -219,54 +237,60 @@
       }
       yield null;
     }
   }
 };
 
 function ContainsSelector(textContent)
 {
-  this._text = textContent;
+  let regexpString = checkRegExpParameter(textContent);
+  if (regexpString)
+    this._regexp = new RegExp(regexpString);

[Manish Jethani, 2018/01/09 07:53:02]

We have to consider what happens if the regular expression is invalid. For
example /(foo/ which contains an unterminated group. There are some options for
how to deal with this, but I think we should just catch the error here and set
this._regexp to null. Since this._text is null too, it'll evaluate to false in
this.match.

[lainverse, 2018/01/09 10:49:22]

I think it should be either "RegExp(regexpString, 'i')" to be case-insensitive
or checkRegExpParameter should support regexp flags defined right in the filter
and return new RegExp instead of just a string.

+  else
+    this._text = textContent;
 }
 
 ContainsSelector.prototype = {
   requiresHiding: true,
 
+  match(element)
+  {
+    if (this._regexp)
+      return this._regexp.test(element.textContent);
+
+    return element.textContent.includes(this._text);

[lainverse, 2018/01/09 10:49:22]

I wasn't quite sure that "string.includes()" actually better than
"regexp.test()" and made this test to check:
https://jsperf.com/compare-regexp-test-and-string-include/1

...and it looks like "string.includes()" 30~40 percent slower both in Google
Chrome 63 and Firefox 58 DE on my system.

So, it might be a good idea to construct new RegExp even from a simple string
and use .test() in all cases. And drop this function.

[hub, 2018/01/09 18:33:06]

I'll rewrite this part taking the comments above into account aswell. Looks like
I'll use regexp matching even for plain text.

+  },
+
   *getSelectors(prefix, subtree, stylesheet)
   {
     for (let element of this.getElements(prefix, subtree, stylesheet))
       yield [makeSelector(element, ""), subtree];
   },
 
   *getElements(prefix, subtree, stylesheet)
   {
     let actualPrefix = (!prefix || incompletePrefixRegexp.test(prefix)) ?
         prefix + "*" : prefix;
     let elements = subtree.querySelectorAll(actualPrefix);
 
     for (let element of elements)
     {
-      if (element.textContent.includes(this._text))
+      if (this.match(element))
         yield element;
       else
         yield null;
     }
   }
 };
 
 function PropsSelector(propertyExpression)
 {
-  let regexpString;
-  if (propertyExpression.length >= 2 && propertyExpression[0] == "/" &&
-      propertyExpression[propertyExpression.length - 1] == "/")
-  {
-    regexpString = propertyExpression.slice(1, -1)
-      .replace("\\7B ", "{").replace("\\7D ", "}");
-  }
-  else
+  let regexpString = checkRegExpParameter(propertyExpression);
+  if (!regexpString)
     regexpString = filterToRegExp(propertyExpression);
 
   this._regexp = new RegExp(regexpString, "i");
 }
 
 PropsSelector.prototype = {
   preferHideWithSelector: true,
   dependsOnStyles: true,