Processing instruction attribute map follow-ups #1504

What is the issue with the DOM Standard?

While reviewing WebKit's implementation of processing instruction attributes (WebKit/WebKit#73326) I ran into three problems.

1. Copies don't get an attribute map

Clone a single node sets a ProcessingInstruction copy's target and data, but the copy's attribute map stays empty:

const pi = document.createProcessingInstruction("t", 'a="1"');
pi.cloneNode().getAttribute("a"); // null

Extract and clone the contents then set the clone's data to a substring directly rather than through replace data. So even with cloning fixed, a processing instruction cut by a range keeps a map that doesn't match its shortened data.

I suggest running update attributes from data on the copy in all three places. The WebKit patch re-parses copies from their data, which amounts to the same thing. Copying the map instead would keep entries that setAttribute() added under names the pseudo-attribute grammar can't parse, such as $, but the range steps would still need to re-parse.

2. removeAttribute() rewrites data when the name isn't there

removeAttribute(name) always runs update data from attributes, even when the map has no entry for name. That re-serializes the data and queues a characterData mutation record, and data that doesn't parse as pseudo-attributes is lost:

const pi = document.createProcessingInstruction("php", "echo 1;");
pi.removeAttribute("x");
pi.data; // ""

Element's remove an attribute by name does nothing when the attribute is missing, and neither does toggleAttribute(name, false) on a processing instruction. I suggest returning early when this's attribute map[name] doesn't exist, which is what the WebKit patch does.

I suppose I could be convinced the other way too, but it seems like we need test coverage for this at a minimum.

3. Typo

In update data from attributes, "Replace any U+003D (>)" should be U+003E.

cc @foolip @noamr