I should not see /REGEXP/ is considerably slower than the non-RegExp variant

[Ichaelus]

It surely does not come as a surprise, but matching DOM elements based on a regular expression is way slower than by compared to simple text matching. Nevertheless, I observed that especially the "not" combination can make a testsuite painfully slow and would propose looking into it.

Reason why

• In Capybara 3, Node#text no longer fully normalizes whitespace, instead returning text as close as possible to "as rendered".

This means non-blanking spaces will no longer be collapsed, carriage returns may be returned in text strings, etc. This affects not only the Node#text method but also methods that take a :text option as well as the text related predicates, assertions, and matchers (has_text?, has_content?, assert_text, have_text, etc.).

• This seems to be a little impractical to us, since it differs the way the HTML and Browser correlate: While text might be be split across multiple lines in the HTML, the user might see it in a single line in the browser.
• Ultimately, it led to an increased usage of steps like I should see /my text\s+split across multiple lines/ for me.

Well, and since the step I should not see // "patiently" traverses the entire DOM with .//descendant-or-self::*, executing it takes some precious time on large HTML pages.

[makmic]

The root cause of my problen has been sepearately inspected and solved in #95. As I do no longer see an immediate need to look into this, I'll close this issue for now.