Bug 193339 – Either StringView::UpconvertedCharacters::UpconvertedCharacters() or StringImpl::createCFString() is using the wrong encoding

• Status: RESOLVED
• Resolution: INVALID
• Component: Text

[Myles C. Maxfield]

Strings in WebKit have two flavors: UTF-16, and UTF-16 with all the leading 0 bytes removed (if all the code points are <= 0xFF).

StringImpl::createCFString() pretends the second one is kCFStringEncodingISOLatin1, which is not correct.

https://en.wikipedia.org/wiki/ISO/IEC_8859-1#Code_page_layout
https://www.unicode.org/charts/PDF/U0080.pdf

[Myles C. Maxfield]

Looks like https://bugs.webkit.org/show_bug.cgi?id=90720 is the culprit

[Myles C. Maxfield]

There seems to be some confusion over what encoding these strings have.

[Myles C. Maxfield]

Either StringBuilder::allocateBufferUpConvert() and StringView::UpconvertedCharacters::UpconvertedCharacters() are wrong, or StringImpl::createCFString() is wrong.

[Myles C. Maxfield]

StringImpl::copyCharacters() says it's not Latin1

TextBreakIteratorICU says it uses Latin1 (in set8BitText()) but imlements the functionality by calling StringImpl::copyCharacters(), which means TextBreakIteratorICU lies

[Myles C. Maxfield]

bool equal(const LChar* a, const UChar* b, unsigned length) says it's not Latin1

[Myles C. Maxfield]

I thought it wasn't true the Latin1 is just the first 255 characters of Unicode, but I'm checking now.

[Myles C. Maxfield]

Looks like it is true. Sorry for the noise.

var error = U_ZERO_ERROR
var i = Int8(1)
for item in 1 ... 0xFF {
    let source = [Int8(i)]
    let targetCapacity = ucnv_convert_63("UTF-16", "ISO-8859-1", nil, 0, source, Int32(source.count), &error)
    assert(error.rawValue > U_ZERO_ERROR.rawValue)
    error = U_ZERO_ERROR
    var target = [Int8](repeating: 0, count: Int(targetCapacity))
    ucnv_convert_63("UTF-16", "ISO-8859-1", &target, Int32(target.count), source, Int32(source.count), &error)
    target.withUnsafeBytes() {(unsafeRawBufferPointer: UnsafeRawBufferPointer) in
        let unsafeBufferPointer = unsafeRawBufferPointer.bindMemory(to: UInt16.self)
        for j in 0 ..< unsafeBufferPointer.count {
            print("\(String(item, radix: 16)) Code Unit \(j) -> \(String(unsafeBufferPointer[j], radix: 16))")
        }
        assert(unsafeBufferPointer.count == 2)
        assert(unsafeBufferPointer[0] == 0xfeff)
        assert(unsafeBufferPointer[1] == item)
    }
    i = i.addingReportingOverflow(1).partialValue
}


...
Program ended with exit code: 0

[Myles C. Maxfield]

Just verified it backwards, too.

[Darin Adler]

(In reply to Myles C. Maxfield from comment #6)
> I thought it wasn't true the Latin1 is just the first 255 characters of
> Unicode, but I'm checking now.

Here’s one reason you might be confused:

When a website specifies Latin-1 as its character encoding, compatible web browsers treat the content of the website as windows-1252 instead, which is like Latin-1 but the bytes in the range 0x80-0x9F for 32 different characters, rather than for U+0080 through U+009F.

You can see this in the WhatWG encoding specification where the names for windows-1252 include strings like "l1", "latin1", and even "ascii".

That encoding is what TextCodecLatin1.h/cpp implements. TextCodecLatin1.h/cpp could be renamed to avoid confusion with actual Latin-1.