PHP Internals News: Episode 98: Deprecating utf8_encode and utf8_decode

Thursday, March 3rd 2022, 09:02 GMT

In this episode of "PHP Internals News" I chat with Rowan Tommins (GitHub, Website, Twitter) about the "Deprecate and Remove utf8_encode and utf8_decode" RFC.

The RSS feed for this podcast is https://derickrethans.nl/feed-phpinternalsnews.xml, you can download this episode's MP3 file, and it's available on Spotify and iTunes. There is a dedicated website: https://phpinternals.news

Transcript

Derick Rethans 0:14: Hi, I'm Derick. Welcome to PHP Internals News, a podcast dedicated to explaining the latest developments in the PHP language. This is episode 98. Today I'm talking with Rowan Tommins about the "Deprecate and remove UTF8_encode and UTF8_decode" RFC that he's proposing. Hi, Rowan, would you please introduce yourself?
Rowan Tommins 0:38: Hi, I'm Rowan Tommins. I'm a PHP software architect by day and try and contribute back to the community and have been hanging around in the internals mailing list for about 10 years and contributed to make the language better, where I can.
Derick Rethans 0:57: Excellent. Yeah, that's how I started out as well, many, many more years before that, to be honest. This RFC, what problem is this trying to solve?
Rowan Tommins 1:08: PHP has these two functions, utf8_encode and utf8_decode, which, in themselves, they're not broken. They do what they are designed to do. But they are very frequently misunderstood. Mostly because of their name. And because Character Encodings in general, are not very well understood. People use them wrong, and end up getting in all sorts of pickles that are worse than if the functions weren't there in first place.
Derick Rethans 1:37: What are you proposing with the RFC then?
Rowan Tommins 1:39: Fundamentally, I'm proposing to remove the functions. As of PHP 8.2, there will be a deprecation notice whenever you use them, and then in 9.0, they would be gone forever, and you wouldn't be able to use them by mistake, because they just wouldn't be there.
Derick Rethans 1:56: I reckon there's going to be a way to actually do what people originally intended to do with it at some point, right?
Rowan Tommins 2:02: So yeah, there are alternatives to these functions, which are much clearer in what you're doing, and much more flexible in what you can do with them so that they cover the cases that these functions sound like they're going to do, but don't actually do when you understand what they're really doing.
Derick Rethans 2:20: I think we'll get back to that a little bit later on. You're wanting to deprecate these functions. But what do these functions actually do?
Rowan Tommins 2:27: What they actually do is convert between a character encoding called Latin-1, ISO 8859-1, and UTF-8. So utf8_encode converts from Latin-1 into UTF-8, utf8_decode does the opposite. And that's all they do. Their names make it sound like they're some kind of fix all the UTF 8 things in my text. But they are actually just these one very specific conversion, which is occasionally useful, but not clear from their names.
Derick Rethans 3:01: It's certainly how I have seen it used in the past, where people just throw everything and the kitchen sink at it, and expecting it to be valid UTF 8, and then at the end, decode. I mean, the decoding was not even part much of this, right? It's just throw everything at it, and then magically it will all be UTF 8. But I reckon that's not really quite the case. When and how does that go wrong?
Rowan Tommins 3:26: So what actually ends up happening is, because text doesn't know what encoding it's in. Something that people misunderstand about character encoding is they think it's like, the text is a certain colour, and the computer knows what colour it is. And if you tell the computer to make it a different colour, then it will work. But it's not like that. In the computer, there's just the sequence of binary. And the encoding is how to read that binary as text. And if you tell the computer to read it as Latin 1, it will read it as Latin 1. If you take to convert from Latin 1 to UTF 8, it will assume the input is Latin 1, it will convert to UTF 8 on that basis. If your text actually wasn't Latin 1 in the first place, you're just going to end up with garbage. And some of the worst cases of that is when you already have UTF 8, and then you run utf8_encode on it, because the language doesn't know that you've already got UTF 8, so it tries to read its Latin 1, write it out ass UTF 8 and you get this weird Mojibake. I don't know pronouncing that right.
Derick Rethans 4:27: I think it's pronounced Mojibake.
Rowan Tommins 4:30: Mojibake.
Derick Rethans 4:31: It's a Japanese term, because clearly these things, these issues happened with Japanese text quite a lot because they have a lot more different and difficult characters and encodings as well. With which things often go wrong though?
Rowan Tommins 4:44: Using an unco on text that's already UTF 8 is obviously a big one. Usually obvious, but occasionally people just getting a muddle with that. The other thing that often happens is confusing with similar encoding. Latin 1 is often mistaken for a different coding windows 1252. To the extent that web pages labelled as Latin 1, web browsers will assume that they're actually in Windows 1252. These PHP functions don't make that assumption. If your text is actually in Windows 1252, and it's been mislabelled Latin 1, you might still think you're doing the right thing. So I've got Latin 1 text, but you haven't. And then the characters that are different, are going to get mangled again. And there's a few other related encodings that often look the same. There are a few other encodings that look the same at a glance that again, will go wrong on any character that's different between the different encodings.
Derick Rethans 5:43: How could a function tell which encoding a certain text was in?
Rowan Tommins 5:49: It's tricky. There are libraries out there that try to do it. Some encodings that are sequences of bits that aren't a valid character. So if any of those appear, it's definitely not in that encoding. Unfortunately, a lot of encodings, every pattern of bits has a meaning. It's just not necessarily mean. So you can't look at the string and just tell at a glance. The only way I've seen that does it effectively, is trying to guess based on what language text it might be in. If your text suddenly has a load of symbols in the middle of sentences, you're probably using the wrong encoding. If it's suddenly got a load of capital letters, in the middle of words, you're probably using the wrong encoding. So you can make guesses like that, that ultimately, there are only ever guesses.
Derick Rethans 6:38: It's only always going to be a guess, right? You can't really tell for certain what it it is, which I've seen people assume that she can just tell. We have concluded that utf8_encode and decode don't actually do what they say they don't magically encode everything to UTF 8. What if things go wrong? How are errors handled?
Rowan Tommins 6:58: If you're converting from Latin 1 into UTF 8, there Latin 1 covers all 256 possible eight bit binary strings. Those will correspond directly to a single mapping in Unicode and therefore in UTF 8. So there are no errors as such, when...