This causes "request.formData() with input: test=" and "response.formData() with input: test=" to fail in LayoutTests/external/wpt/url/urlencoded-parser.html:

    FAIL request.formData() with input: test= assert_array_equals: property 0, expected "test" but got "test"
    FAIL response.formData() with input: test= assert_array_equals: property 0, expected "test" but got "test"

The strings above are misleading, as test's actually creating Request and Response objects like this:

    let init = new Request("about:blank", { body: "\uFEFFtest=\uFEFF", method: "LADIDA", headers: {"Content-Type": "application/x-www-form-urlencoded;charset=windows-1252"} }).formData()

"\uFEFF" is the UTF-16 BOM.

When |body| is being extracted (https://fetch.spec.whatwg.org/#concept-bodyinit-extract), we parse a USVString ("\uFEFFtest=\uFEFF") and correctly encode it into UTF-8 ("\xef\xbb\xbftest=\xef\xbb\xbf").

However, if we later analyze that request or response by calling formData(), json() or text(), we can see that the initial "\uFEFF" is lost and we're left with "test=\uFEFF". https://url.spec.whatwg.org/#urlencoded-parsing says we should just decode the UTF-8 string without BOM, we shouldn't be stripping the BOM from the beginning.

The BOM is being stripped because TextResourceDecoder::Decode() unconditionally decodes content without the BOM if it finds one.