Consider always escaping U+0020

[zcorpan]

In ad3660f U+0020 changed from being escaped to be not escaped for schemes like data javascript mailto etc. Rationale seems to be that it's what some browsers did back then, per http://krijnhoetmer.nl/irc-logs/whatwg/20121026#l-573

Test http://software.hixie.ch/utilities/js/live-dom-viewer/saved/4238

It appears that Gecko still escapes space in all of these cases. Are there known Web-compat problems for Gecko because of this?

Round-tripping the space can be problematic because space is like the only character that really separates a URL from the next or from the following text. Features like ping="" and srcset="" will get bogus values if someone assumes that a parse-serialization roundtrip will eliminate spaces from a URL (and it typically does).

[annevk]

Can you or @sleevi fix Chromium? Need to have at least two browsers for such a change I think.

[zcorpan]

cc @tyoshino

[annevk]

@achristensen07 any thoughts on this issue?

[achristensen07]

alert(new URL("foo:a\u001F\u0020b")); shows that Chrome and Safari do not escape spaces according to the spec. I hope to keep this to minimize change.

[annevk]

Okay, so per the URL standard we use the "simple encode set" (which doesn't include U+0020) for hosts (doesn't matter since hosts forbid spaces anyway), cannot-be-a-base-URL paths, and fragments.

For both cannot-be-a-base-URL paths and fragments the results are identical: Chrome, Edge, and Safari do not encode U+0020, Firefox does (although Firefox does not for data URLs). I tested various schemes, including https, test, data, and x.

Given this I wonder if @zcorpan has made any progress convincing a non-Firefox browser that this is a problem. Otherwise I think we should consider this a lost cause.

[zcorpan]

I have not. Unless @achristensen07 has a change of heart, let's close. 🙁

[annevk]

I tested Firefox unfairly for data URLs. Firefox has data URLs implemented in the URL parser and therefore it would reject input such as data:test test and echo that back out. If you instead give data:,test test#test test as input, it will encode the spaces as %20.