Shouldn't we just drop the \0 check at the top? Even in non-strict mode the NUL byte should simply be ignored.

That's true, I've updated the PR.

With this change ext/standard/tests/url/bug55273.phpt on Travis fails with

006+ bool(false)
006- string(2) "PH"
009+ bool(false)
009- string(1) "P"

This is probably related to the contents in the ch == base64_pad branch ... which look pretty fishy to me. E.g. it increments current in one loop while not also decrementing the length :/

Whoops. Would you prefer going back to my original fix or revising the whole thing?

The aforementioned padding branch doesn't even handle the length correctly but instead loops through all whitespace, possibly causing past-the-end access.

I opted for the rewrite, which actually revealed various other problems. I'll start a discussion on the mailing list, let's see how it goes.

Submitted another approach as PR 1923.