Attribute value serialization does not take whitespace normalization into account

[bwrrp]

The steps for "to serialize an attribute value" only escape the characters ", &, < and > in the attribute value. White space characters are passed through to the serialization as-is. However, XML processors will replace each space, tab, carriage return or line feed character with a space according to https://www.w3.org/TR/xml11/#AVNormalize unless the character was present as a character reference. It seems therefore that the attribute value serialization algorithm should include a step mapping tab to 	, carriage return to 
 and line feed to 
.

Testing this in various browsers shows that these already apply a similar substitution:

new XMLSerializer().serializeToString(
    new DOMParser().parseFromString('<root attr="&#x20;&#x9;&#xD;&#xA;"/>', 'text/xml')
)
// <root attr=" &#9;&#xD;&#xA;"/> in Firefox
// <root attr=" &#9;&#13;&#10;"/> in Edge / Chrome

The algorithm as described in this specification would generate <root attr=" \t\r\n"/> (where \t \r and \n represent tab, carriage return and line feed respectively). Only Safari seems to follow the specification here. Unfortunately, this serialization does not survive a round-trip, as it is normalized to four spaces by processors such as the DOMParser:

new XMLSerializer().serializeToString(
    new DOMParser().parseFromString('<root attr=" \t\r\n"/>', 'text/xml')
)
// <root attr="   "/>

[cscott]

This also matches the behavior in https://github.com/jsdom/w3c-xmlserializer/blob/master/lib/attributes.js#L29