PHP Internals News: Episode 96: User Defined Operator Overloads

PHP Internals News: Episode 96: User Defined Operator Overloads

In this episode of "PHP Internals News" I chat with Jordan LeDoux (GitHub) about the "User Defined Operator Overloads" RFC.

The RSS feed for this podcast is, you can download this episode's MP3 file, and it's available on Spotify and iTunes. There is a dedicated website:


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 96. Today I'm talking with Jordan, about a user defined operator overloads RFC that he's proposing. Jordan, would you please introduce yourself?

Jordan LeDoux 0:33

My name is Jordan LeDoux. I've been working in PHP for quite a while now. This is the second time I have ventured to propose an RFC.

Derick Rethans 0:44

What was the first one?

Jordan LeDoux 0:45

The first one was the "never for parameter types", which was much more exploratory. And we talked about it a little bit. And it generated a lot of good discussion that contributed to kind of the idea formation, which was what I hope to get out of it.

Derick Rethans 1:01

Okay, but that didn't end up making it into a PHP release. As far as I understand, right?

Jordan LeDoux 1:07

No, I withdrew it actually, it was clear that the better way to approach the problem it was trying to solve was with a much more comprehensive solution. That particular solution was something that only required a seven line change to the engine. So I wanted to see if it was something people were okay with, or thought was a decent idea for that particular problem, much more comprehensive, like template classes, or something like that is probably the better route to go.

Derick Rethans 1:35

Well, I think the RFC that we're talking about today, is going to require quite a bit more than seven lines of code?

Jordan LeDoux 1:41

Quite a bit more. Yeah.

Derick Rethans 1:42

So what is this RFC that we're talking about today?

Jordan LeDoux 1:45

Well, user defined operator overloads is a way for PHP developers to define the ways in which objects interact with specific operators. So for instance, the plus operator, the plus sign. It's a way for those objects to kind of define their own logic as far as how that's handled, which right now, as of PHP 8.0, those were all switched to type errors. So it's not possible currently to write any code that doesn't result in a fatal error, where objects are used with operators.

Derick Rethans 2:25

Usually, I ask about every RFC, what problem are you trying to solve this? So what problem are you trying to solve this RFC?

Jordan LeDoux 2:31

The biggest problem that this solves is that objects contain, so objects in most programs represent a value or multiple values that have a program context. That's the most powerful thing about objects is they're contextual, and they understand the state, they understand what state the object is in, and sometimes even what state the whole program is in. And that's necessary for a lot of things. Like for instance, if you're tracking a distance, you know, you might measure that meters, and that would have a number you might have 30 meters of distance, but it also has a unit of meters. You could just represent that as an int. And then the program just knows internally, hey this is always in meters. But if you need to convert that to a different unit, then that becomes: Okay, well, now I need a special case some things, or I need a function just for converting, and I need to remember which unit my number is in. In a lot of cases, you handle that with objects because objects understand state, and they understand state transitions, which is what a lot of methods are about; transitioning the state of the object from one state to another. Operators are also about state transitions. And they're about very specific kinds of state transitions. It's natural in a lot of ways to think that you, you should be able to define how those two things interact. But currently, it's just not possible within PHP.

Derick Rethans 4:00

Well, does them this magic operator overloading?

Jordan LeDoux 4:04

It allows PHP developers to define an implementation logic, which is much like you define a function body that describes how does this object interact with this operator. That's essentially it. There's a lot of other details as to how it does that and what are the restrictions, but that's really the core of the idea.

Derick Rethans 4:26

And in what kind of situations would you use that?

Jordan LeDoux 4:28

A lot of them are situations where you're doing very complicated mathematics, or scientific computing or machine learning or things of that nature, where you are going to routinely encounter numbers that have state to them or that have multiple dimensions to them. So for instance, vector mathematics is one where the way that vectors interact with a lot of the operators that we're familiar with, like the multiplication sign is very different than how the number five interacts with the multiplication sign. Complex numbers is another one, you know, to multiply two complex numbers together, you have to treat it like a polynomial where you're multiplying it with the FOIL method: first, outside, inside, last. You know, there's a lot of those sorts of circumstances. But it also could potentially be very useful for some things that are not really mathematical but more quality of life for PHP developers. For instance, scalar objects is something that a lot of developers in PHP have, you know, wanted for a while. It's a thing that's a little more difficult to pin down, how exactly would you go about doing this within the engine, and it's a thing that the engine would kind of have to be very opinionated about by its nature. PHP developers can't provide their own scalar objects. And the main reason for this is that scalars interact with operators and objects can't. So simply allowing PHP developers to define a way for objects to interact with operators would allow user land to develop their own scalar object replacements. It wouldn't make every scalar that object; scalar objects within the engine still has, it's a separate feature. And it's still a thing that would be desirable, probably to a lot of people. But it gets quite a bit of the way there.

Derick Rethans 6:20

It is always interesting that people come up with the example of complex numbers, because I'm not sure how useful that is in a PHP user land context. And then beyond the scalars, I then sometimes struggle to see where this could be used. With the only exception is probably doing calculations with money related issues. The moment you bring up operator overloading, you'll also get people to say that this is going to get abused. Examples of that, in my opinion at least, is where in C++ you have like the << operator to put things into the stream and stuff like that. What answer would you have to kind of comments?

Jordan LeDoux 6:58

Abuse of operator overloads to do things that can create unmaintainable code, because that's really the concern for developers is, does a language feature promote code that's difficult to maintain, that's difficult to understand, that's difficult to follow, and develop, and you know, work with. The RFC, the way that I've gone about this implementation, has had that in mind, because I also have experienced that. This is not a thing where I coming down from the academic high tower with, you know, whatever my my concept of this is, and no, no real world experience with these things. I share a lot of those concerns. Actually, I think this is a very useful feature that has a lot of applications I've encountered. I have had to work with matrix maths, I have had to work with complex numbers, I've had to work with arbitrary precision numbers, and all of those situations would have been served so much better by having operator overloads. I was fighting with the language the entire time, I was trying to do those. But I understand you know, in a lot of web applications, those are not common problems to encounter. My experience of that isn't typical. The thing about the way that it's done is it tries to head off a lot of the ways that it could be misused. An example of that is that the RFC requires typing of the parameters. You can't define an operator method and leave the types blank. If you do, then you get a fatal error during compile. It tells you you must explicitly define a type. And the reason for this is that blank types are assumed to be mixed. So it's the same as putting mixed for the type within the engine. And a mixed type says I can take anything, it doesn't matter what you give me, I can take anything. But that simply isn't true for operators. It's never true. Because even if you think hey, I can accept floats, ints, I can accept any objects, I can figure something out with them. You know, even if you think that's true, what happens when somebody passes you a stream resource? I mean, that's part of mixed. Any implementation that says mixed is probably lying. This RFC requires you to document what are the types that you know how to interact with for this operator. And that's the thing that that developers are kind of going to be forced to think about when they implement this. You know, and that's one example. But there's several other things within the RFC that kind of try and take that concern very seriously. And say, what are the strategies we could design something that is going to be used correctly, most of the time, just by design.

Derick Rethans 9:42

Would just not then create an inconsistency in the language where for some methods, you simply have to type the arguments.

Jordan LeDoux 9:50

So yes, it's it is different than how other functions are defined. And methods are defined on classes, but that's one of the reasons that I believe very strongly that using a keyword other than function is a good idea. That's one of the other things that this RFC proposes is, instead of saying function plus or whatever, you say, operator plus. One of the things that that does is that signals to the developer, this is a different thing. That's not a trivial aspect of the RFC. It's not something that can just kind of be thrown away. It's like, oh, that sugar. In a very real way communicates to the developers, this is not like other functions, this is a different thing. It is a function internally within the engine. But that's because that's faster to do it that way. And it's a better way to implement it internally, within core. Developers should not be treating it in PHP as a function, it shouldn't be used that way. It's an engine hook.

Derick Rethans 10:51

When you're writing the code. If you do operator plus, for example, then at that point, it's clear what the plus does, but not necessarily, when you read the code, and you see the plots, you don't necessarily know what it means, right? Which I think is one of the bigger criticisms of having operator overloading support. But then you can also make the argument saying that well, operators they have a specific meaning in normal language, right. The plus means adding two things. So the argument would be that only use the plus operator for adding things together, not for example, adding a comment to a blog post, which you technically could do, right?

Jordan LeDoux 11:25

You could.

Derick Rethans 11:26

I definitely say that is something you should definitely not do, which you could, for example.

Jordan LeDoux 11:30

That's another reason to kind of not treat them as functions in the syntax. You know, I think that having that operator keyword there really communicates that strongly to PHP developers. You know, when you look at a line of code, that's variable A plus variable B, and you're sitting there thinking: Hmm, I wonder if there's an operator overload involved here, because that might be a thing you do have to think about if this were included in core. While that's an additional thing that might have to be investigated, you know, by developers, and that that's not a trivial thing, I completely acknowledge that. It's also not a thing that would happen by accident, it would have to be intentional, because all objects error, if they're used with an operator currently, and after this is introduced, all objects will continue to error unless they define their own overload within the class that's being called, or one of its parents obviously, because inheritance is respected. It's not a thing that would happen by accident, there's no code that's going to accidentally inject an object into an operator, and all of a sudden, PHP makes wild assumptions and your code is spitting out a number that doesn't make sense, or something like that, because it's simply going to error. This is going to error very early. So you're going to get that feedback from the engine right away, when you do something like that. Maybe you didn't intend or that maybe was ambiguous.

Derick Rethans 12:55

I've just realized that in languages like C++, you can define multiple versions of the same operator, because you can have method overloading. This is not something you can do in PHP with normal methods either. So do I understand correctly that you can't do that in this case, either it, you need to accept multiple types in the overloaded operator, and then make a decision yourself.

Jordan LeDoux 13:17

It was suggested to me by a couple of people who gave me very early feedback that, hey, C++ accomplishes this with method overloading, you should do method overloading. And I took one look at that and said: One, I'm already doing a lot of work for this, that sounds like double the work. And two, I'm not convinced that's the best way to do it. Three, that's a huge separate change, that should probably be considered separately. And four, I don't think it's necessary. You can accomplish it with Union types, which we have. And that's another thing that maybe this is a guardrail for PHP developers using it incorrectly. If you're unioning, eight different types, and maybe you're not using it correctly. I mean, that'll look ugly. And I'm people might complain: hey, I don't want to have to Union all these things. I want to be able to overload the method directly with multiple versions. Having that feedback, right in your code that: Hey, this looks ugly. Maybe I'm doing it wrong. I see it as a positive thing, in a lot of ways.

Derick Rethans 14:19

I agree. First of all, it's a separate subject that should be discussed separately. Now, so far, we've only mentioned the operator keyword, but we haven't spoken about the rest of the syntax yet. So how would you define an overloaded operator?

Jordan LeDoux 14:33

As we were discussing, there's the keyword operator. So you would define it very similar to how you would define a function. You can give it a visibility, but it can only accept the visibility public, you can omit that if you want. But it can be abstract or final. So you can have an abstract class that forces an implementation, or you can have a class that disallows overriding of the method. You use the keyword on operator, and then where the function name would go for any other function, you use the symbol that you want to overload, so you don't name it the English word plus, you use the actual symbol '+'. And then the rest of it is the way you would define any other function or method because it has a lot of the same concerns that functions do. But it visually looks very different, which I think is another good guardrail. Another good bit of feedback to developers.

Derick Rethans 15:28

What are the arguments that the overload is operating methods need to accept?

Jordan LeDoux 15:33

Most of them accept and actually require two arguments. The first is the corresponding operand. The things that are to the right and the left of your operator, they're called operands. And one of them will have this overload and the other one will be some kind of value. You need to accept the other value. And then the second parameter is the operand position, whether or not the operator overload being called; whether it's on the left side of the operator or the right side of the operator, because some some operations depend on whether or not it's on the left or right side.

Derick Rethans 16:13

Would you say that most of the time, the operators will be used on two objects of the same class, in which case that doesn't really matter?

Jordan LeDoux 16:22

A lot of the time, I think good implementations of this feature would involve objects that share a base class, share a parent class, or are the same class. I think it would be a very rare circumstance where a good usage of this feature would involve accepting a class that doesn't meet either of those criteria. Maybe it could happen, but I think in most situations, that would be another one of those things that kind of gives you you know, the code smell that a something may be wrong.

Derick Rethans 16:55

Then of course, with the exception that, for example, vectors, you can multiply with a number. And I define number very loosely here. And then in that case, the order is important. So the RFC has a table of having a whole list of operators, but it doesn't include all of them. What kind of categories are included, which ones aren't?

Jordan LeDoux 17:12

There's two main categories of operators that are proposed in this RFC, the mathematical operators, you know, your plus, minus, divide, multiply, the pow operator, and the modulo operator. And then the second class of operators are all the bitwise operators. So bitwise and, bitwise or, bitwise not, shift left, shift right, that kind of thing.

Derick Rethans 17:37

And let's see in the table that It all says equals in the spaceship operators in there. But what I don't see in there, it's larger than, or smaller than operators.

Jordan LeDoux 17:46

I made the decision very early when I was developing this RFC that I didn't want to support the comparison operators independently. And what I mean by that is, I didn't want to have an object that defined separate logic for the greater than sign than they did for the less than sign. That was mainly to avoid situations where reversing things would change the Boolean logic. Instead, there's a single operator, the comparison operator, or the spaceship operator, that allows you to overload all of them, but only in a way that's self consistent. By implementing that operator overload, you can cover all of the inequality operators, but it will always be consistent with its own output. It's never going to give you things that are logical contradictions with its own data.

Derick Rethans 18:43

Would the overloaded spaceship operator implementation also be used for other comparisons, like greater than, less than and greater than equals?

Jordan LeDoux 18:52

That's correct. Going into the implementation just a little bit. Internally, all of those operators, the greater than sign, the less than sign, greater than, and equals to, all of those are internally done as a comparison. That type of comparison where you're outputting, negative one, zero or positive one, they indicate, is it larger? Is it smaller? Is it equal? This actually keeps the PHP user land implementations more consistent with how things are done internally within the engine and makes it much easier to support all of those things, not just consistently, you know, without logical contradictions, but as far as how it gets done within the engine, it makes it much easier to handle those.

Derick Rethans 19:39

Yeah, I see there's another few implied operators in there. For example, if you're like the -= operator, then that gets implied as $a = $a - $b and stuff like; that all seems to be fairly sensible there. And similar it like ++$a, you get $a = $a + 1, which is basically what that means. You mentioned the word implementation detail. And I have a question myself here is: The symbol tables contrary to support a plus or minus? So do they get transformed into a specific name, for example?

Jordan LeDoux 20:12

Internally, the function name for a method on a class is stored as a Zend string, which can handle the symbols, it just doesn't. And that's mainly because the lexer can't; the parser is restricted from doing that, because it's kind of ambiguous in all contexts. For instance, outside of a class, following a function, using arbitrary symbols might cause some issues. But that's another thing that the operator keyword makes simpler. The operator keyword in the parser makes allowing the symbols much smaller implementation hurdle, I think that would be something that would be very difficult to do with the function keyword. But internally, it actually does get stored as the symbol. And then it gets put as a kind of an internal pointer with the other Magic Methods. Because internally, it's treated kind of like a magic method.

Derick Rethans 21:07

Are they flagged with a specific flag or a bit, showing that they are overloaded methods?

Jordan LeDoux 21:13

Yes, there's a new flag that's added as part of this. That's only for methods, ZEND_ACC_OPERATOR.

Derick Rethans 21:21

Which I think becomes important if you start looking things like reflection. Because if you list all the methods on a class on the reflection class, then you sort of need to know, what are the already overloaded operator methods or normal methods?

Jordan LeDoux 21:37

Yes, that's, that is something that became very important when I went into do the reflection implementation for this, which has also been completed at this point. As part of reflection, actually, I very much didn't want to return the operators with other methods. Because again, I don't think that developers should be encouraged to think of these as methods, in most circumstances. That having the flag there made that a very simple change. It was like three or four lines of code per implementation per method that was affected on the reflection classes, check the flag, and then we're done. We're out.

Derick Rethans 22:13

In addition to that, of course, you gets operator specific reflection methods, right? Because you do want to check whether you have them.

Jordan LeDoux 22:20

For normal methods, you have getMethod, getMethods, and hasMethod. And so there's three additional methods that are added to reflection class, getOperator, getOperators, and hasOperator, and they behave exactly the same way as the corresponding method ones, but they only deal with the operators.

Derick Rethans 22:43

The RFC is talking about it an operator methods will be represented by reflection methods, which makes sense, but as you indicate there aren't really methods. And you shouldn't really think of them as methods. So would it not make sense to have a reflection operator method perhaps?

Jordan LeDoux 22:59

I did consider that. So when I was looking at the implementation for ReflectionMethod, I was looking at the methods that you have on that. And I was saying to myself, is this something that shouldn't be there for operators that not only, you know, maybe it doesn't provide useful information, like for instance, isPrivate will always be false for operators because you can't make operators private, but it doesn't break for operators, it still works. And all of the methods on ReflectionMethod were of that nature. Some of them were not super useful for operators, but none of them were things that were broken, or that were totally didn't make sense. And so because of that, I thought, well, maybe it's better to just have ReflectionMethod and just use that again, instead of creating a separate one that doesn't really have any additional functionality. It's just a copy, essentially, so that they don't have to be maintained separately.

Derick Rethans 23:57

I see in the RFC, that you're also adding the isOperator methods to reflection methods, so that you can distinguish between normal methods and operator overloaded methods, right, which is then I suppose the alternative to having a different instance class that represents either the method or the operator?

Jordan LeDoux 24:15

So that was the only thing that I really saw as being necessary, necessarily different, is being able to tell is my instance of ReflectionMethod a normal method or an operator method. That could be solved by having a child class instead, that would be another way to do it, I can definitely see advantages of doing it that way. And I thought about doing it that way. It's already a very big RFC. I kind of wanted to reduce the amount of things that people had to think about or that people had to say, well, this is something different. This is already very different from a lot of things in PHP. And it was one of those things where I was like, that seems like a place where it's not necessary for me to create something new for people to consider.

Derick Rethans 24:56

As you say, this is quite a long and complicated RFC. What's been the feedback been so far?

Jordan LeDoux 25:02

A lot of the feedback so far has revolved around the new keyword, the operator keyword. You know, questions about why is this necessary, as opposed to using the function keyword, which we talked about already a little bit. And kind of going through, what are the implications of that, not just within PHP, but also downstream for tooling to things like Psalm, Rector, tools that PHP developers use IDEs, PhpStorm, you know, what are they going to have to do to handle this? And is that more difficult or less difficult with a keyword? Depending on what the answer to that is? Is that trade off worth it?

Derick Rethans 25:41

Has there been any of the expected feedback saying: Oh, this is just going to be abused by users all over the place?

Jordan LeDoux 25:47

There's been one or two so far, you know, I think operator overloading as a concept as a feature in programming. And this isn't restricted to PHP as a language. This is something that comes up in other languages, too. I think, as a concept, this feature is something that's always kind of been that way to a lot of languages. There's very few languages where people don't have strong opinions about it. Even in those languages, people don't really encounter that often. But it's the kind of thing that people feel strongly about. So I would always imagine that there are going to be people who, quite rightly, from their own experience, believe that this is just a bad idea. And I can understand why they would think that. I disagree, but I can understand why they would think that. I think about the only language I'm aware of that doesn't have that kind of thing going on is maybe R, but R is a language that's kind of designed around nothing but mathematics. So the idea of being able to control operators is kind of central to what the language does. So it's maybe the only example I can think of, but the rest of them, you know, it is somewhat controversial. And I think it kind of always will be, even if it gets accepted.

Derick Rethans 26:54

Talking about that. When do you think you'd be opening voting for this?

Jordan LeDoux 26:58

I'm thinking more along the lines of early January. I think holding the vote two weeks after I announced it on internals a second time, it would be right almost on top of Christmas, I think that would also kind of be a bit unkind, and also may not serve the RFC well. So I think waiting till January is probably the right idea.

Derick Rethans 27:18

I think that's the nicer way of doing it as well. Yes. Do you have anything to add that we forgot to speak about?

Jordan LeDoux 27:25

I wanted to mention going back to the operator keyword, and kind of the discussion around that. And the feedback that's been generated so far on that, a really good way to think about it is that the operator keyword is very similar to the enum keyword. Enums are classes, they simply are, but they're classes with very specific restrictions on them. The operator is a function, but it's a function with very specific restrictions on them. And it's for a lot of the same reasons. Enums are intended to be used for a very specific purpose. Operator overloads are also intended to be used for a very specific purpose. And that's one of the reasons that I think it's not not as bad of a thing. And I think that people really should be thinking about it more in terms of why we have the enum keyword instead of terms like, why don't we just use another magic method or something like that? You absolutely could do it that way, the same way that you could do enums it's just classes, but there's value there and doing it with its own keyword, I think.

Derick Rethans 28:29

Well, thank you, Jordan for taking the time this morning or your night, to talk about the operator overloads proposal.

Jordan LeDoux 28:35

Yeah, thank you for having me.

Derick Rethans 28:41

Because I've been on hiatus for a while I wanted to jump in with a few newsworthy items. First of all, I would like to thank Nikita for the many years he worked on PHP, while being an employee of JetBrains. He has decided that he wants to work on something else besides PHP and choose to leave JetBrains to work on LLVM. This means that I will be speaking to him on this podcast a lot less, if at all.

With Nikita's departure the PHP protect now has nobody working full time on it, as it is desirable for the continuation Nikita's old employer, JetBrains, has banded together with members of the PHP community, including core contributors, companies and sponsors to set up a foundation to fund contributors to work on PHP. Once this is up and running, I will make sure to dedicate an episode to this exciting new development. I have included a link to the foundation on Open Collective in the show notes.

Just before Nikita left the project two more RFCs were passed. The first one was to move the PHP bug tracker from to repository now accepts your bug reports, whereas the system has been largely retired. We still accept security bugs on the old issue tracker because we can discuss these in private there before making them public.

The second RFC implemented the deprecation of dynamic properties with PHP 8.2. Instead of allowing codes to define a rights to undeclared properties, they will now need to be defined in your class definition, otherwise, you will get a deprecation warning. I have included the link to this RFC in the show notes as well. I'm not sure whether I will produce a specific episode on the subject.

With all the news out of the way, I'd like to thank you for listening to this installment of PHP internals news, a podcast dedicated to demystifying development of the PHP language. I maintain a Patreon for an account for sponsors of this podcast as well as the Xdebug debugging tool. You should sign up for Patreon at If you have comments or suggestions, feel free to email them to Thank you for listening, and I'll see you next time.

PHP Internals News: Episode 95: PHP 8.1 Celebrations

PHP Internals News: Episode 95: PHP 8.1 Celebrations

In this episode of "PHP Internals News" we're looking back at all the RFCs that we discussed on this podcast for PHP 8.1. In their own words, the RFC authors explain what these features are, with your host interjecting his own comments on the state of affairs.

The RSS feed for this podcast is, you can download this episode's MP3 file, and it's available on Spotify and iTunes. There is a dedicated website:


Derick Rethans 0:14

Hi, I'm Derick, and this is PHP internals news, a weekly podcast dedicated to demystifying the development of the PHP language.

Derick Rethans 0:23

This is episode 95. I've been absent on the podcast for the last few months due to other commitments. It takes approximately four hours to make each episode. And I can now unfortunately not really justify spending the time to work on it. I have yet to decide whether I will continue with it next year to bring you all the exciting development news for PHP 8.2.

Derick Rethans 0:44

However, back to today, PHP eight one is going to be released today, November 25. In this episode, I'll look back at the previous episodes this year to highlight a new features that are being introduced in PHP 8.1. I am not revisiting the proposals that did not end up making it into PHP 8.1 feature two features I will let my original interview speak. I think you will hear Nikita Popov a lot as he's been so prolific, proposing and implementing many of the features of this new release. However, in the first episode of the year, I spoke with Larry about enumerations, which he was proposing together with Ilija Tovilo. I asked him what enumerations are.

Larry Garfield 1:26

Enumerations, or enums, are a feature of a lot of programming languages. What they look like varies a lot depending on the language, but the basic concept is creating a type that has a fixed finite set of possible values. The classic example is booleans. Boolean is a type that has two and only two possible values true and false. Enumerations are way to let you define your own types like that, to say this type has two values Sort Ascending or Sort Descending. This type has four values for the four different card suits, and a standard card deck. Or a user can be in one of four states pending, approved, cancelled or active. And so those are the four possible values that this variable type can have. What that looks like varies widely depending on the language. In a language like C or C++, it's just a thin layer on top of integer constants, which means they get compiled away to introduce at compile time, and they don't actually do all that much they're a little bit to help for reading. On the other end of the spectrum, you have languages like rust or Swift, where enumerations are a robust, advanced data type and data construct of their own. That also supports algebraic data types. We'll get into that a bit more later. And is a core part of how a lot of the system actually works in practice, and a lot of other languages are somewhere in the middle. Our goal with this RFC is to give PHP more towards the advanced end of enumerations. Because there are perfectly good use cases for it, so let's not cheap out on it.

Derick Rethans 3:14

In the next episode, I spoke with Aaron Piotrowski about another big new feature: fibres.

Aaron Piotrowski 3:20

A few other languages already have Fibers like Ruby. And they're sort of similar to threads in that they contain a separate call stack and a separate memory stack. But they differ from threads in that they exist only within a single process and that they have to be switched to cooperatively by that process rather than pre-emptively by the OS like threads. And so the main motivation behind wanting to add this feature is to make asynchronous programming in PHP much easier and eliminate the distinction that usually exists between async code that has these promises and synchronous code that we're all used to.

Derick Rethans 4:03

I also asked Aaron about small PHP I actually have a slightly related question that pops into my head as like. There's also something called Swoole PHP, which does something similar but from what I understand actually allows things to run in threats. How would you compare these two frameworks or approaches is probably the better word?

Aaron Piotrowski 4:25

Swoole is they try and be the Swiss Army Knife in a lot of ways where they provide tools to do just about everything. And they provide a lot of opinionated API's for things that in this case, I'm trying to provide just the lowest level just the only the very necessary tools that would be required in core to implement Fibers.

Derick Rethans 4:48

Although I discussed several deprecations from Nikita and the last year, I only want to focus on the new features. In episode 76. I spoke with him about array unpacking, after talking about changes to Null in internal functions.

Nikita Popov 5:01

The old background is set we have unpacking calls. If you have the arguments for the call in an array, then you write the free dots and the array is unpacked intellectual arguments. Now what this RFC is about is to do same change for array unpacking, so allow you to also use string keys.

Derick Rethans 5:24

In another episode, I spoke with David Gebler on a more specific addition of a new function fsync. David explains the reason why he wants to add this to PHP.

David Gebler 5:34

It's an interesting question, I suppose in one sense, I've always felt that the absence of fsync and some interface to fsync is provided by most other high level languages has always been something of an oversight in PHP. But the other reason was that it was an exercise for me in familiarizing myself with PHP core getting to learn the source code. And it's a very small contribution, but it's one that I feel is potentially useful. And it was easy for me to do as a learning exercise.

Derick Rethans 5:58

And that is how things are added to PHP sometimes, to learn something new and add something useful at the same time. After discussing the move of the PHP documentation to GIT an episode 78, in Episode 79, I spoke with Nikita about his new in initializers RFC. He says:

Nikita Popov 6:15

So my addition is a very small one, actually, my own will, I'm only allowing a single new thing and that's using new. So you can use new whatever as a parameter default, property default, and so on.

Derick Rethans 6:29

The addition of this change also makes it possible to use nested attributes. Nikita explains:

Nikita Popov 6:34

I have to be honest, I didn't think about attributes at all, when writing this proposal. What I had in mind is mainly parameter defaults and property defaults. But yeah, attribute arguments also use the same mechanism and are under the same limitations. So now you can use new as an attribute argument. And this can be used to effectively nest attributes.

Derick Rethans 6:59

Static Analysis tools are used more and more with PHP, and I spoke to the authors of the two main tools, Matt Brown, of Psalm, and Ondrej Mirtes of PHPStan. They propose to get her to add a new return type called noreturn. I asked him what it does and what it is used for.

Ondrej Mirtes 7:14

Right now the PHP community most likely waits for someone to implement generics and intersection types, which are also widely adopted in PHP docs. But there's also noreturn, a little bit more subtle concept that would also benefit from being in the language. It marks functions and methods that always throw an exception. Or always exit or enter an infinite loop. Calling such function or method guarantees that nothing will be executed after it. This is useful for static analysis, because we can use it for type inference.

Derick Rethans 7:49

Beyond syntax, each new version of PHP also adds new functions and classes. We already touched on the new fsync function, but Mel Dafort proposed to out the IntlDatePatternGenerator class to help with formatting dates according to specific locales in a more specific way. She explains:

Mel Dafert 8:07

Currently, PHP exposes the ability for locale dependent date formatting with the IntlDateFormat class, it says basically only three options for the format long, medium and short. These options are not flexible in enough in some cases, however, for example, the most common German format is de dot numerical month dot long version of the year. However, neither the medium nor the short version provide and they use either the long version of the month or a short version of the year, neither of which were acceptable in my situation.

Derick Rethans 8:40

And she continues with her proposal:

Mel Dafert 8:42

ICU exposes a class called DateTimePatternGenerator, which you can pass a locale and so called skeleton and it generates the correct formatting pattern for you. The skeleton just includes which parts are supposed to include it to be included in the pattern, for example, the numerical date, numerical months and the long year, and this will generate exactly the pattern I wanted earlier. This is also a lot more flexible. For example, the skeleton can also just consist of the month and the year, which was also not possible so far. I'm proposing to add IntlDatePatternGenerator class to PHP, which can be constructed for locales and exposes the get best pattern method that generates a pattern from a skeleton for that locale.

Derick Rethans 9:26

Locales and internationalization have always been an interest for me, and I'm glad that this made it into PHP 8.1. I spoke at length with Nikita about his property accessors RFC, in which he was suggesting to add a rich set of features with regard to accessibility of properties, including read only, get/set function calls, and asymmetric visibility. He did not end up proposing this RFC, which he already hinted that during our chat:

Nikita Popov 9:53

I am still considering if I want to explore the simpler alternatives. First, there was already a proposal, another rejected proposal for Read Only properties probably was called Write Once Properties at the time. But yeah, I kind of do think that it might make sense to try something like that again before going to the full accessors proposal, or instead.

Derick Rethans 10:18

He did then later proposed a simpler RFC read only properties, which did get included into PHP eight as a new syntax feature. He explains again:

Nikita Popov 10:27

This RFC is proposing read only properties, which means that a property can only be initialized once and then not changed afterwards. Again, the idea here is that since PHP 7.4, we have Type Properties. Remaining problem with them is that people are not confident making public type properties because they still ensure that the type is correct, but they might not be upholding other invariants. For example, if you have some, like additional checks in your constructor, that a string property is actually a non empty string property, then you might not want to make it public because then it could be modified to an empty value. For example, one nowadays fairly common case is where properties are actually only initialized in the constructor and not changed afterwards any more. So I think this kind of mutable object pattern is becoming more and more popular in PHP.

Derick Rethans 11:21

Nikita, of course, meant this kind of immutable object pattern, which we didn't pick up on during the episode. Another big change was the PHP type system, where George Peter proposed out pure intersection types. He explains what it is:

George Peter Banyard 11:35

I think the easiest way to explain intersection types is to use something which we already have, which are union types. So union types tells you I want X or Y, whereas intersection types tell you that I want x and y to be true at the same time. The easiest example I can come up with is a traversable that you want to be countable as well.

Derick Rethans 11:54

To explain our pure George Peter says:

George Peter Banyard 11:58

So the word pure here is not very semantically, it's more that you cannot mix union types and intersection types together.

Derick Rethans 12:06

Just after the feature freeze for PHP 8.1 happened in July, another RFC was proposed by Nicolas Grekas to allow the new pure intersection types to be nullable as well. But as that RFC was too late, and would change the pure intersection type to just intersection types, it was ultimately rejected.

Derick Rethans 12:23

The last feature that I discussed in a normal run of the podcasts was Nikita's first class callable syntax support. He explains why the current callable syntax that uses strings and arrays with strings has problems:

Nikita Popov 12:35

So the current callable syntax has a couple of issues. I think the core issue is that it's not really analysable. So if you see this kind of like array with two string signs inside it, it could just be an array with two strings, you don't know if that's supposed to actually be a static method reference. If you look at the context of where it is used, you might be able to figure out that actually, this is a callable. And like in your IDE, if you rename this method, then this array should also be this array elements will also be renamed. But that's like a lot of complex reasoning that the static analyser has to perform. That's one side of the issue. The second one is that colour bulls are not scope independent. For example, if you have a private method, then like at the point where you create your, your callable, like as an array, it might be callable there, but then you pass it to some other function, and that's in a different scope. And suddenly that method is not callable there. So this is a general issue with both the like this callable syntax based on arrays, and also the callable type, is callable at exactly this point, not callable at a later point. This is what the new syntax essentially addresses. So it provides a syntax that like clearly indicates that yes, this really is a callable, and it performs the callable culpability check at the point where it's created, and also binds the scope at that time. So if you pass it to a different function in a different scope, it still remains callable.

Derick Rethans 14:08

This new feature is a subset of another RFC called partial function applications, which was proposed by Paul Crovella, Levi Morrison, Joe Watkins, and Larry Garfield, but ultimately got declined. So there we have it, a whirlwind tour of the major new features in PHP 8.1. I hope you will enjoy them. As I said in the introduction, I'm not sure if I will continue with the podcast to talk about PHP 8.2 features in 2022 due to time constraints. Let me know if you have any suggestions.

Derick Rethans 14:41

Thank you for listening to this installment of PHP internals news, a podcast dedicated to demystifying the development of the PHP language. I maintain a Patreon account for supporters of this podcast as well as the Xdebug debugging tool. You can sign up for Patreon at If you have comments or suggestions, feel free to email them to Thank you for listening and I'll see you next time.

PHP Internals News: Episode 94: Unwrap Reference After Foreach

PHP Internals News: Episode 94: Unwrap Reference After Foreach

In this episode of "PHP Internals News" I chat with Nikita Popov (Twitter, GitHub, Website) about the "First Class Callable Syntax" RFC.

The RSS feed for this podcast is, you can download this episode's MP3 file, and it's available on Spotify and iTunes. There is a dedicated website:


Derick Rethans 0:14

Hi, I'm Derick. Welcome to PHP internals news, the podcast dedicated to explaining the latest developments in the PHP language. This is Episode 94. Today I'm talking with Nikita Popov about the unwrap reference after foreach RFC that he's proposing. Nikita, would you please introduce yourself?

Nikita Popov 0:33

Hi, Derick. I'm Nikita and I work at JetBrains on PHP core development.

Derick Rethans 0:38

So no changes compared to the last time.

Nikita Popov 0:41

Yes, at the time before that.

Derick Rethans 0:43

So what is the problem that is RFC is going to solve?

Nikita Popov 0:46

Well, it's really a very minor thing. I think it's a relatively well known problem for the more experienced PHP programmers. It's like a classic example, you have a foreach loop by reference. So foreach array as value by reference, and then you do a second loop after that, foreach array as value at the same it's by value. So without the reference sign. The result of that is that your last two array elements are going to be the same, which is kind of unexpected. If you're not familiar with how references in PHP work and scoping in PHP works. So I think it's worth explaining what's going on there.

Derick Rethans 1:27

Can you quickly explain the scoping or rather the lack of it, I suppose?

Nikita Popov 1:31

Yeah, it's really the lack of PHP really only has function scoping. So if you have a foreach array as value, then the value variable is going to stay alive, even after the foreach loop. And usually, that won't make much of a difference. So you will just have like reference to the last element of the array, might even be useful for some cases, you know, before we added the array, I think, array_key_last function. If the last element now is a reference, so if you have a reference to the last element, then you're write into that variable is also going to modify the last element of the array. So if you now have a second foreach loop, using the same variable, that's actually not just modifying that variable, but it's also always modifying the last element of the array.

Derick Rethans 2:15

Okay, just to clarify, it isn't necessarily the last element in the foreach loop. It's the last one that's been assigned to?

Nikita Popov 2:22

Yeah, that's, that's true.

Derick Rethans 2:24

Is this not something that people actually use for some useful reasons?

Nikita Popov 2:28

As mentioned before, technically, you could use it to get a reference to the last element and then modify the last element outside the foreach loop. I don't think this is a particularly common use case. But I'm sure people have used in here there. This is a use case we would break with the proposed RFC.

Derick Rethans 2:47

I think it is one I have used in the past, it's probably not how I would do it now. But I'm pretty sure I have some point in the past. What are you proposing to change with this RFC?

Nikita Popov 2:57

The change is pretty simple. And that's to unwrap or to break the reference after the loop. You will still have like after the loop, the variable will still contain the value of the last element, or of the last like visited element, but it will no longer be a reference to it. If you write into the variable, it will not modify the original array. And if you have a second loop that writes into the variable that also doesn't modify the original error any more.

Derick Rethans 3:25

At which point and how is this reference broken?

Nikita Popov 3:29

It's at the end of the foreach loop, or as you say, if you break out too early, then of course, it would also get broken. So it's referenced inside the foreach loop and stops being referenced outside the loop.

Derick Rethans 3:41

And that would happen also, if I would use a goto for example?

Nikita Popov 3:45

Oh, that that's a trick question, actually, yes, it should happen. But now that you have mentioned it, I think my current implementation does not handle that particular case, I will have to double check it. But that should happen, yes.

Derick Rethans 4:00

It's good to know that you've thought about it then.

Nikita Popov 4:02

Well, I didn't think about it. Because I mean, I guess I can mention it here, the way this works is that well, at the end of the foreach loop, we have like an instruction that frees the loop variable. And I can just add an additional one that breaks reference. But if you use things like goto or multi level breaks, or something like that, then we insert these clean-up instructions before the jump. We have to make sure to actually insert the reference breaking instruction there as well. So it's like not automatically handled.

Derick Rethans 4:38

Is this going to be a separate instruction or as we tend to call them opcodes?

Nikita Popov 4:43

I'm using a separate one, but one could run it as a flag into the instruction that frees the loop variable, but I think it's cleaner to have a separate instruction for it. Like technically one could optimize it away in some cases, like I wouldn't bother but it's like semantically a different thing.

Derick Rethans 5:01

I think it'd be nicer result, because it makes it easier to visualize what's happening, right?

Nikita Popov 5:06

Yeah, it is.

Derick Rethans 5:07

Did you actually check whether some code uses this construct?

Nikita Popov 5:10

I have to admit, I tried checking it using a very basic approach, just look at foreach loops by reference. And then if the variable is used after that. But that kind of primitive approach has way too many false positives, for example, you have a foreach loop inside, and if, and then the variable is reused inside an else. So it like wouldn't flow from the if into the else. So you would have to do some kind of more sophisticated control flow analysis. It's something that can be done, but I didn't bother doing it for a one off backwards compatibility check. So I don't have any hard data on how much code is actually using something like this.

Derick Rethans 5:51

So this is where I'm a little bit on the fence about this change, because it is changing behaviour, that's going to be pretty hard to figure out what is actually going to affect your codebase.

Nikita Popov 6:01

It should be possible to very reliably detect that. It's just something you have to actually implement. But you're right now there is no easy way to check that.

Derick Rethans 6:13

It's something that static analysers could probably have a look at.

Nikita Popov 6:16

Yeah, expect that maybe Psalm or PHPStan, something like that will be easier to implement, because they already have control flow information.

Derick Rethans 6:23

You don't really know how impactful this, which is, in my opinion, a bit of the scary bit. How important do you think you'll find it to have this RFC going through and implemented?

Nikita Popov 6:33

I don't think it's super important. It's mostly like, small quality of life fix for newer developers . People who have already encountered this issue once won't forget about it again. In fact, it's somewhat common recommendation that you should always unset the loop variable after a foreach by reference loop. So I've seen that as like a policy some people use, that could be avoided. So yeah, I don't think it's a critical feature, just a small improvement.

Derick Rethans 7:08

Would it be an alternative idea to instead deprecate the foreach by reference?

Nikita Popov 7:14

Okay, that's the radical approach. Everything is possible. I think that foreach by reference is relatively, I mean, I think it's one of the most common uses of references we have, and one of the most reasonable ones. I mean, the alternative is search into by value loop, and then you modify it by looking up the element by key again, which is a bit more ugly, I would say. I think we shouldn't deprecate foreach by reference, though it would be kind of nice to have a different way to achieve the same. One other unfortunate thing about foreach by reference is that it leaves behind references in the array. The case I'm looking at here is this reference to the last element, where you have like reference structure that's pointed to both from inside the array, and from this loop variable. The other thing that foreach by reference does is that for all the other array elements, you will actually leave behind the reference wrapper that's just used in this one single place for this single array element. Essentially, you are wasting memory, because we will leave behind this that reference wrapper. So after you do the foreach by reference loop over the array, the array will actually grow larger. So if you're storing like integers, and it may grow significantly larger, like from a technical perspective, foreach by references, also not great. But like from a usability perspective, it's nicer then modifying values by key lookup.

Derick Rethans 8:53

I guess it's going to depend on how big the array is, right? I mean, if it's a few elements, it probably doesn't matter.

Nikita Popov 8:58

But if you have like a 100,000 element array, then you paying for 100,000 reference wrappers that you don't need afterwards any more.

Derick Rethans 9:07

In that case, it's rather better to just modify it through the key that you obtained by doing foreach key as value.

Nikita Popov 9:14

Right. But it's also worth noting that foreach reference actually has different semantics then foreach value, because foreach by value works on the copy of the array. Like it's not an actual copy just like semantically. If you modify the array inside the foreach by value loop, then we will copy the array. Doing the modification with a separate key lookup and foreach by value loop will actually copy the array at that point, while foreach by reference takes account modifications of the array. So even if you like add or remove elements in the array in the foreach by reference loop, it will try on the like best effort basis to still iterate on in a reasonable way on the modified array. It's like not a straightforward replacement.

Derick Rethans 10:00

It all depends on what people intended to do with it. Right? Do you think there are any further situations that are a bit strange? That could benefit from having some subtle changes to the language semantics?

Nikita Popov 10:13

Nothing can who comes to mind immediately.

Derick Rethans 10:16

Yeah, I can't think of any either. But I thought maybe maybe have something in the pipeline. Would you have anything else to add to this RFC?

Nikita Popov 10:23

Well, one more thing that's discussed in the RFC is the case of complex variables. A little known fact, in the foreach loop, you don't have to assign to a simple variable, you can also assign to something like an object property, or an object property on the result of a function call that that means that in the loop, this function is getting called on every iteration, and then you assign it to a property on the result. So you can do that kind of weird stuff, we allow it.

Derick Rethans 10:52

And does it the work without any weird side effects?

Nikita Popov 10:56

Depends on what you consider weird, but basically does what you expect as if you had written an explicit assignment to the complex variable.

Derick Rethans 11:04

I reckon that's how it's instructed out in the oparray then as well.

Nikita Popov 11:07

Yeah, exactly. As far as this RFC is concerned, the problem there is that to unwrap the reference of the loop, we actually have to evaluate the variable again. And if it's a complex variable that might have side effects, for example, the function call. And that's why the RFC says that if the variable is complex, we are not going to do that, like that's probably going to be more unexpected than leaving a reference wrapper around. So we have this extra weird edge case. In the internals discussion, some people already suggested that maybe we should just deprecate support for these kind of complex assignments. One could also mention that an alternative that has been suggested is to actually make the loop variable, scoped to the foreach loop. So we could unset it entirely after the loop, rather than just breaking the reference, which is, of course, a larger change, larger backwards compatibility break. It also doesn't really align with PHP semantics of only having function scope and not block scope.

Derick Rethans 12:06

I probably agree without, it's too much of a change to do that. Because then you sort of expect that all the language constructs should have a scope. I mean, it needs to be either one or the other.

Nikita Popov 12:15

Yeah, I mean, other languages like JavaScript have solved that by introducing a separate way to declare scoped variables. So that will be "let", just changing the behaviour in one place is probably not a good idea.

Derick Rethans 12:30

I probably agree with you though. It was a bit of a shorter RFC this time. That's okay with me.

Nikita Popov 12:35

Yes, I used that as an excuse to discuss some foreach behaviour details.

Derick Rethans 12:40

Fair enough. Thank you for taking the time this morning to come and talk to me about the references after foreach RFC.

Nikita Popov 12:47

Thanks for having me, Derick, once again.

Derick Rethans 12:53

Thank you for listening to this installment of PHP internals news, a podcast dedicated to demystifying the development of the PHP language. I maintain a Patreon account for supporters of this podcast, as well as the Xdebug debugging tool. You can sign up for Patreon at If you have comments or suggestions, feel free to email them to Thank you for listening. I'll see you next time.

PHP Internals News: Episode 93: Never For Parameter Types

PHP Internals News: Episode 93: Never For Parameter Types

In this episode of "PHP Internals News" I chat with Jordan LeDoux (GitHub) about the "Never For Parameter Types" RFC.

The RSS feed for this podcast is, you can download this episode's MP3 file, and it's available on Spotify and iTunes. There is a dedicated website:


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 93. It's been quiet over the last month, so it didn't really have a chance to talk about upcoming RFCs mostly because there were none. However, PHP eight one's feature freeze has happened now, a new RFCs are being targeted for the next version of PHP eight two. Today I'm talking with Jordan LeDoux, about the Never For Parameter Types RFC, the first one targeting this upcoming PHP version. Jordan, would you please introduce yourself?

Jordan LeDoux 0:50

Certainly. And thanks for having me. My name is Jordan. I've worked as a developer for about 15 years now. Most of my career has been spent working in PHP. Although professionally, I've had experience working in C#, Python, TypeScript, mostly in the form of JavaScript, but a little bit of Node and, you know, a variety of other languages that I haven't spent enough time in to really be proficient in any real way. But recently, I decided to do something that I have thought about doing for many years, but never actually jumped into which is exploring the PHP engine itself and how I could possibly contribute to it.

Derick Rethans 1:32

And here we are, but your first our thing.

Jordan LeDoux 1:35

Yeah, it's exciting.

Derick Rethans 1:36

What is this RFC about, what does it propose?

Jordan LeDoux 1:39

Well, this RFC proposes allowing the never type, which was added in 8.1 as a return value, to parameters for functions and methods on objects. The main idea behind that is that when never was proposed as a return type, it was meant to signal that the function would never return. Not that it returns void, which of course, void signifies which is returning no value or returning, returning without any specified information. And never return signifies that the function will never return, which is a concept that exists in many other languages. And for that purpose in other languages, what's usually used is something called a bottom type. And that's what never ended up being. And I'm proposing that we extend the use of that bottom type to other areas where the type may be helpful.

Derick Rethans 2:38

So a bottom type, that might be a new term for many people, it will certainly for me when I looked at the never RFC for as return types. Can you sort of explain what a bottom type is, especially thinking about object oriented theory with something that we'd like to call the Liskov Substitution Principle? And also, how does it apply to argument types?

Jordan LeDoux 2:59

Let's start with the Liskov Substitution. The general idea behind Liskov Substitution is that if A is a subtype of B, then anywhere that A exists, you should be able to substitute B. It has to do with when you have a class hierarchy in in an object oriented language, that that class hierarchy guarantees certain things about substitutionality, like whether or not something can be substituted for something else. That affects language design in ways that a lot of programmers are kind of intuitively familiar with, but maybe not familiar with the theory and the ideas behind it more concretely. But LSP is the principle in SOLID, that's the L and SOLID. And it represents a portion of the whole idea of object oriented programming in PHP. Part of being able to substitute one object for another, based on their class hierarchy, and what they implement, and what they provide is there part of their ability to be substituted is whether or not they can fulfil the same kind of contractual requirements of typing. And with Liskov, that means that preconditions can never be strengthened, and post conditions can never be weakened. So a precondition would be a parameter type requirement. If you require that a parameters accepts an object, for instance, in PHP, you can't strengthen that requirement beyond just any object to a particular object. But you can weaken it from an object to an object or an integer with with unions. That's an example of the precondition side of it. The post condition side of it is that you can't, you can't weaken it. So if you have have, you know, if you have a return type of int, you can't have an inherited implementation return int or float, because that broadens the possible return types. They go in opposite directions. And one of them is covariance and one of them is contravariance.

Derick Rethans 5:18

I can never remember which one is which.

Jordan LeDoux 5:21

Yeah, basically contravariance go up the tree and covariance go down the tree. If you're thinking about widening, or sorry, narrowing. If you're thinking about narrowing, then covariance go down the implementation tree and contravariance go up the implementation tree.

Derick Rethans 5:40

Okay, so how does the bottom type fit in here?

Jordan LeDoux 5:43

The bottom type in any type system represents like the base type that all types originate from. And the best way in my mind to think about it is kind of just integer math. It's a thing that every programmer is going to be familiar with. And it fits, it fits all right. So you can think of the bottom type as zero integers, a lot of people would think of null as zero if they're thinking about a type system, but null is more like negative one. It's like the entire negative side of the integer system. We could say that the string type is one, and the integer type is two. And the float type is three, and maybe int or float, the union of them is five, which would be the two numbers added together. And if you describe type systems this way, then you can say, hey, if I take any type and add the value that represents the other type, then I get my result type. So if I take zero, the bottom type, and I add one, the string type, what I end up with is one, still the string type. So the bottom type is whatever type system or whatever, whatever type, when you add any other type to it, you get the type you added to it and nothing else, you just get your original thing. That's why it's called the union identity for the type system. The top type in PHP is mixed. And that's the opposite side of it. It's just like zero is the additive identity. One is the multiplicative identity. If you multiply anything by one, you're going to get what you originally had. And if you add anything to zero, you'll get what you originally had. So mixed ends up being, or the top type in general, ends up being the intersection identity, and the bottom type, or never, in PHP's case, ends up being the union identity. And all this is like deep type theory, but most programmers don't have to interact with it. It's more something that affects language design usually.

Derick Rethans 7:48

Could you think of mixed as being infinity?

Jordan LeDoux 7:50

That's actually with my with my crude integer analogy, yeah, it would be like all types, all possible types are added together.

Derick Rethans 7:59

That makes sense then. Okay, so we have explained what the bottom type is, but, and never being the bottom type. So why is it useful to use the bottom type, or never, as this RFC proposes, as a method argument type for parameters?

Jordan LeDoux 8:14

The largest benefit has to do with what we were talking about when it comes to covariance versus contravariance. You know, can you strengthen the requirements? Or can you weaken the requirements? When you inherit a method in a system that preserves Liskov Substitution, the parameters can be widened, they can accept more things. If I had an interface that said, it has one parameter, and that parameter is typed as int, then in any implementation, I could say, okay, but actually, the parameter type is int or float, I'm going to accept both. And I could do that in the implementation. But I would have to accept int, because that's part of the contract. That's part of the interface. So I have to accept whatever type is in the interface. I can just additionally add things on top of that. If I make my original definition, my root definition, the bottom type, then I can add any type to it. And I will just get that type. From never, if I had an interface with, you know, a method foo, and it has one argument, and that argument is typed never, I can re-type that argument as int, or I could re-type that argument as string, and all of them would be valid inheritances.

Derick Rethans 9:38

So it's a way of getting around having at least one concrete type like int in an interface.

Jordan LeDoux 9:44

Right. And in fact, never is a concrete type. It's a concrete type, that means this code can never be called. For return type, it means this type will never return. But if you're, if what you're trying to do is call it then you can never call it.

Derick Rethans 10:01

Which is then why never actually makes sense as a type name, because one of my further questions is going to be how does never as a name make sense? But you've now explained it in such a way that it actually does make sense. So there we go.

Jordan LeDoux 10:15

One of the questions that did come up in the internals discussion on this so far, has been a round that choosing never, and because a lot of the examples for use cases are around inheritance, it makes some kind of intuitive sense, if you're describing the inheritance behaviour, to name it something like any or anything, or you know, something about its how permissive it is in its inheritance definition. But the type should really describe the code that it's actually written in. You know, getting outside the idea of an interface or an abstract or something like that. If you have a function, and that function types it as never, or whatever you name, the bottom type. There's no data that can satisfy that. Because any data will have some type other than never. It'll have string, or int, or something. Even null has the null type. You can't provide any data to a function that requires never that will satisfy its type requirement. So that code can never be called. It's really when you start considering how does this affect inheritance that you start getting into this concept of Oh, maybe a different word makes sense. But then that doesn't really reflect the opposite side on the return side, when you're talking about covariance instead of contravariance. And that's why the bottom type for most languages is something like never, or nothing, or nil, something along those lines.

Derick Rethans 11:48

So if you type an argument to a method as never, the engine will, of course, enforce that you can't call it with any data because it wouldn't satisfy. But would it also automatically make a method abstract so that inheritance inheriting classes have to widen it or not?

Jordan LeDoux 12:04

So that was part of my original idea behind the RFC, was forcing something that implements it or something that inherits it, to widen it. However, there were a lot of good arguments about why that may not be a good idea. One is that in PHP, an empty type isn't an empty type, it's mixed, not widening, would actually just be saying the type is mixed, which is valid, you can go from the bottom type to the top type in a contravariant way, that's a totally valid way to do it. The problem is that PHP has a weakly enforced typing system. Like that's only a problem in this context, it's actually a very powerful feature of the language in a lot of other contexts. So we don't necessarily want to get rid of that. In addition to that, the actual mixed type as a literal that can be used in the language was only added in 8.0. It would kind of represent a much larger backwards compatibility break to require it to be explicitly widened. That was part of my original concept for a lot of the reasons that you were just talking about. But for PHP, specifically, it would probably present more problems than it would provide kind of solutions and utility. I've kind of been convinced off that point a little bit by the arguments of others.

Derick Rethans 13:22

Would it be possible to instantiate a class that has a method with its argument typed as never?

Jordan LeDoux 13:28

Yes, as long as he never called that argument directly.Using a never type in a constructor would definitely be a definitely be a No, no, that would result in a type error. As soon as you try to instantiate the class.

Derick Rethans 13:41

It would basically make the constructor private.

Jordan LeDoux 13:43

Yeah, you would get a slightly less useful error.

Derick Rethans 13:47


Jordan LeDoux 13:48

Then you do if you make the constructor private, and then try and call it.

Derick Rethans 13:53

It makes no sense to do it. The RFC slightly touches on generics. And it goes in a way talking about why this is sort of slightly like generics. Could you explain the interaction between these two concepts?

Jordan LeDoux 14:07

Generics is a feature obviously, that a lot of PHP developers want. And it's also a very complicated feature to do. Way outside of what I was willing to consider for, for my first, my first attempt at something useful. One of the most common ways that generics are used, is within an inheritance structure to allow something similar to type widening, particularly for parameters. Being able to say, I want the type to be able to be widened, but I don't know exactly how it will be widened. That's something that generics offer. Generics offer many other features as well and many other capabilities. But that particular one is something that this can do. It comes with a cost though, because this isn't generics. It is not really the right way to do that. It can be done without generating compile errors now, if you use never, that's the main difference. You still would encounter errors in static analysis and IDE hints, for instance. The IDE, he wouldn't be able to tell if you typed against a interface that had never as a parameter type, it wouldn't be able to tell what sorts of types the implementers have. Because that's not really the point of accepting an interface as a type for a parameter, or for a function call, or something like that. The point is that any implementer of this will be an acceptable type. But that means that from a static analysis perspective, it won't know what the type requirements are, because it won't know what concrete implementations are being provided in the code. So obviously, this is a limitation. And this limitation does not exist in a in an actual generics implementation of some kind. I do view it as an improvement personally. And the main reason is that before, if you tried to do something like this, then the errors that you generated and the problems that you caused, were in your code, in its actual execution. Now, the errors and the problems that you need to solve are going to be in your static analysis or in your IDE. And that's, that's something that's a lot safer for the code in general. It can present some maintainability challenges, it can be annoying for developers to deal with, all of that's absolutely true. And it doesn't provide all the things that you would want from being able to do that. It moves the where the safety problems are from being in code execution, where it can cause real problems into code writing, which is where you have the opportunity to kind of think about it, reason about it and catch it.

Derick Rethans 16:54

From what I understand is that if you have a class doing implementing some kind of generics, then you'd often expect that where this generic type is used in either argument parameter, or return value, that'd be the same for all the methods, whereas with never, you can of course not enforce it, that it isn't being the same type being used, in all the other places where you would otherwise expect or enforce with generics to be the same type. So I reckon that's one of the differences. I think that was a better explanation that what I read from the RFC.

Jordan LeDoux 17:25

That was a explanation and in argument that I wasn't forced to actually articulate until I presented it to other people, because I went through several days of research before even writing the RFC. And sometimes when you do that, particularly for you know, for programming related things, you just absorb the information and you kind of forget about what did I, which things were new, like which things that I just learned, and which things that I already know. You just integrate it all into your new programming knowledge. And so that was definitely something that when I wrote the first draft of the RFC, I didn't, didn't articulate particularly well, because I it just made sense in my head. And I had forgotten: Hey, this is something I didn't know a week ago, I should probably explain it to others.

Derick Rethans 18:10

That's why we have the RFC process, right? So that other people also voice the opinion, and perhaps all the slightly confused language that make perfect sense to the author, but not necessarily to other people that read it, right?

Jordan LeDoux 18:23


Derick Rethans 18:24

The RFC kindly mentioned that there's no backwards incompatible changes. So that's always good news, that makes it a little bit easy to accept. What was sort of the biggest pushback against the RFC?

Jordan LeDoux 18:36

Initially, actually, the most pushback that I got, when I presented it, was a round choosing never as the name, which I thought would, in my mind, I thought would be something that was barely discussed, actually. But I mean, that again, just like you said, that's why this process exists. So that everybody can actually understand, you know, the things that go into it. I did do the research into that prior, but I couldn't find a single language that had more than one bottom type. The concept of more than one bottom type itself, if you if you go back to the integers concept, that'd be like having what positive zero and negative zero or something like that. It's a concept that just intuitively when you, when you understand how the type system itself works, you feel like okay, so there's probably something wrong from a design perspective, if you have more than one bottom type. It's very easy to not be able to see that intuitively. If you don't go in and take a really deep look at how the type system works or, or how types in general work and what they mean and stuff like that. That was the biggest pushback that I got initially. That mostly just involved explaining the things that I just said about like, why is that the case? Why does that make sense? What are some examples of other languages? Just going into that kind of information. The biggest blocker at the moment, really is about, does it make sense to provide never as a parameter type? If you can't use that statically? If you can't use it in a static analysis situation, then does it make sense to ever use it? And if it doesn't make sense to ever use it, then even if it provides contravariance, is it worth adding? And that's the main discussion that's being had right now, that's not entirely resolved. I think that there is still value in doing that. I think that that argument would carry a lot more weight with me personally, if PHP had a better way of handling the situations where that might happen. That's a much larger undertaking, which I would be interested in, but is also not really not really something that would be as kind from a backward compatibility standpoint.

Derick Rethans 20:54

Which we are quite keen on.

Jordan LeDoux 20:57

Right, exactly. I don't personally see a way that this type of functionality could be provided, that could satisfy that concern, and not also invalidate enormous amounts of code that currently exist. You kind of have to choose one or the other from my understanding. I will be very pleased if somebody is able to provide me a way to, even if it's a lot of work, provide me a way that that can be accomplished without breaking a lot of code. That was one of my goals is I don't want this, I want this to be an addition to what currently exists, not not something that breaks the current typing that we have, or the way that PHP developers currently interact with the type system or anything like that. That's the current discussion. And I think the thing that is most unresolved.

Derick Rethans 21:46

I think you'll have a hard time trying to introduce breaking changes into the language. And I also think rightfully so.

Jordan LeDoux 21:54

Yeah, it's it's a good safeguard. And it's a good principle to have in general, I think. I think the way that I described it is that this type of breaking change, the type that would be necessary, wouldn't really be like the type of breaking change you expect in a major version, it would more be like a parallel language syntax. It'd be more like rewriting PHP into a different language with very similar semantics, but very different idea of what the language means underneath. Because fundamentally, in order to do that, typing could no longer be optional anywhere. It would have to be every variable, every piece of data, every function, would have to have an explicit type that the developer is able to control and modify and mutate as they want. I don't even know if type juggling would be possible with the kind of change that would be necessary. And that's, that's half of what PHP consider PHP, so.

Derick Rethans 22:55

Definitely not requiring types and places. Because I think, as I said, I think you'll have a hard time convincing people to go that way.

Jordan LeDoux 23:02

Which is the reason I didn't consider going that way, really, so.

Derick Rethans 23:06

Would you have anything else to add that I'm, that we missed discussing this RFC?

Jordan LeDoux 23:10

This RFC is really interesting to me personally, just from an intellectual perspective. I think that for a lot of users, there's not a lot of use cases where you would use it in your own programs. If you did, it would end up being in like the very core base systems, and only in a few places. In those places, it might really shine, it might be something that's absolutely incredible. Most places in most programs you would never be using, well, you would never be using this type. It's much more critical for some of the internal features things like array access, that interface, and the typing that it requires for parameters. The typing for that's pretty broken. This could be a way that we could fix that possibly, and a couple of the other internal engine features that are implemented through interfaces and things like that would also probably be helped quite a bit by this.

Derick Rethans 24:10

Thank you very much for taking the time today to talk about the Never For Parameter Types RFC.

Jordan LeDoux 24:16

Thank you for having me.

Derick Rethans 24:20

Thank you for listening to this installment of PHP internals news, a podcast dedicated to demystifying the development of a PHP language. I maintain a Patreon account for supporters of this podcast as well as the Xdebug debugging tool, you can sign up for Patreon at If you have comments or suggestions, feel free to email them to Thank you for listening and I'll see you next time.

PHP Internals News: Episode 92: First Class Callable Syntax

PHP Internals News: Episode 92: First Class Callable Syntax

In this episode of "PHP Internals News" I chat with Nikita Popov (Twitter, GitHub, Website) about the "First Class Callable Syntax" RFC.

The RSS feed for this podcast is, you can download this episode's MP3 file, and it's available on Spotify and iTunes. There is a dedicated website:


Derick Rethans 0:14

Hi, I'm Derick. Welcome to PHP internals news, the podcast dedicated to explaining the latest developments in the PHP language. This is Episode 92. Today I'm talking with Nikita Popov about a first class callable syntax RFC that he's proposing together with Joe Watkins. Nikita, would you please introduce yourself?

Nikita Popov 0:36

Hi, Derick. I'm Nikita and I am still working at JetBrains. And still working on PHP core development.

Derick Rethans 0:43

Just like about half an hour ago when we recorded an earlier episode.

Nikita Popov 0:47


Derick Rethans 0:48

This RFC has no relation to read only properties. What is the first class callable syntax RFC about?

Nikita Popov 0:55

The context here is that PHP has the callable syntax based on literals, which is that if you just use a plain string, it's interpreted as a function name, and an array where the first element is an object, and the second one is a method name, that's methods. Or the first element is the class name, and the second one is method name, that's a static method.

Derick Rethans 1:17

I would consider this concept a bit of a hack, especially the the one with the arrays, and I reckon you feel similar and hence this RFC?

Nikita Popov 1:27

Yes, I do. So the current callable syntax has a couple of issues. I think the core issue is that it's not really analysable. So if you see this kind of like array with two strings inside it, it could just be an array with two strings, you don't know if that's supposed to actually be a static method reference. If you look at the context of where it is used, you might be able to figure out that actually, this is a callable. And like in your IDE, if you rename this method, then this array should also be this array element will also be renamed. But there's like a lot of complex reasoning that the static analyser has to perform. That's one side of the issue. The second one is that callables are not scope independent. For example, if you have a private method, then like at the point where you create your callable, like as an array, it might be callable there, but then you pass it to some other function. And that's in a different scope. And suddenly that method is not callable there. So this is a general issue with both like this callable syntax based on arrays, and also the callable type. It's a callable at exactly this point, not callable at a later point. This is what the new syntax essentially addresses. So it provides a syntax that like clearly indicates that yes, this really is a callable, and it performs the callable callability check at the point where it's created, and also binds the scope at that time. So if you pass it to a different function in a different scope, it still remains callable.

Derick Rethans 3:01

And it's guaranteed to always be callable.

Nikita Popov 3:03

Yeah, exactly.

Derick Rethans 3:04

What does the syntax like?

Nikita Popov 3:06

The syntax is the funny bit. As a bit of context. This proposal was created as an alternative or as a subset of the partial function application RFC.

Derick Rethans 3:17

That is just as hard to pronounce as first class callable syntax RFC.

Nikita Popov 3:21

Yes, that's why we say PFA. The PFA RFC has a more general feature. It also allows you to create a reference to a callable as a side effect. But more generally, it allows you to also bind some of the arguments to a fixed value. And has like finer control over for example, you can create a callable that has three required parameters, by passing three question mark arguments. While the new syntax only allows you to use the signature of the original function. But the syntax between both of those is compatible. So the new RFC is a subset of PFA. And that's why it uses the syntax where you do a normal function call, but then pass three dots or an ellipsis as arguments.

Derick Rethans 4:08

Instead of passing the function's or method's normal arguments, you use the three dots.

Nikita Popov 4:14

I think like the way to think about the syntax is that this is similar to like a variadic argument, or to the argument unpacking syntax, just that the arguments haven't yet been provided, they will be provided during the actual call. But I think the syntax was definitely the most contentious bit in the discussion of the RFC. I think this is mainly related to the fact that if you the see this code snippet, it looks a bit like, like the example code where the arguments haven't been filled in. While now this is like actual syntax.

Derick Rethans 4:44

I'm sure there's quite a few tutorials out there explaining how PHP works by using dot dot dot. That is not something you can avoid.

Nikita Popov 4:54

Well, we can avoid it, but it's fairly tricky question. I mean, the reason for this dot dot dot syntax, on one hand, this the compatibility with partial functions. I mean, the PFA, RFC has recently been declined. But in the future, we could extend the current syntax to full partial functions. And we would not end up with two different ways. So that's one benefit of the syntax. But the other part is that PHP has different symbol tables for different kinds of symbols. People often ask, why can't you just write like strlen as a plain name, not inside a string, and have that be treated as a reference to this function? And the answer to that is that we can't do that because you can't have a constant that's called strlen. Normally, that would be reference to constant and the same actually applies to all other callable types as well. So if you have something like methods, like object or method name, that would right now be interpreted as a property access. And for static methods, it will be interpreted as a as a class constant access. So we have this ambiguity here. Even if we add an additional symbol to this, for example, like for classes, we have the syntax, class name, and then scope operator class, that gives you the class name. We could do something like strlen, scope operator function, or fn, or whatever, and have that return the callable. That would work, but it also has some ambiguities. For example, if you have something like object, arrow methods, and then scope operator fn, you have this ambiguity. Is this referencing the method of that name? Or is it referencing a callable stored inside the property of that name? This is like fundamentally ambiguous. The way we would resolve it is we will just say that this index is only usable with real simple, so it will always refer to a method, and you couldn't use the syntax to convert the callable stored in a property into a proper callable. I'm actually not sure how I should distinguish these two concepts, because we have the existing callable, strings and arrays, and the first class callables, which are really closure objects.

Derick Rethans 7:11

Which actually sort of brings me to the next question which just popped in my head, which is: Does this first class scalable syntax, what is returned as return a closure or an existing callable type as we have now, with a callable type being a single string, or this array syntax that we now use.

Nikita Popov 7:28

The syntax returns a closure. Actually, the syntax works essentially the same way as the closureFromCallable method. And we do need to return a closure otherwise, we don't get this behaviour where the scope is bound at the time where the callable is created, rather than called. I think maybe going forward, I would generally recommend that people use a closure type, instead of a callable type in type declarations. I mean, you already cannot use callable for property types. Exactly due to this problem that callability is context dependent. While we only forbid it in property types, the same general problem also exists for argument and return types. And especially with the new syntax being introduced here, I think it's best to use closure instead of callable in the future.

Derick Rethans 8:18

Does that sort of mean that first class scalable syntax is syntactic sugar? Or does it do more than the closureFromCallable method?

Nikita Popov 8:27

No, I think it's effectively just syntactic sugar for closureFromCallable.

Derick Rethans 8:34

I'm actually not sure whether Xdebug is be able to do anything with these closure from callable things to begin with. So that is something I'm going to have to investigate.

Nikita Popov 8:44

Be able as in like, display that it actually refers to a specific method rather than just some kind of closure?

Derick Rethans 8:51

Yeah, because at the moment, it shows you the file name and the line numbers, it doesn't have a name right if you create normal closures, but in this case, it's important to know that it actually refers to specific methods, which is the same thing as the closureFromCallable syntax would also do, but I've never done anything with that.

Nikita Popov 9:10

But I think there is a way to get like the underlying prototype for the closure, and you should be able to determine it from there.

Derick Rethans 9:18

The first class callable syntax, are there situations where you can't use it?

Nikita Popov 9:22

One place where you don't want to use the new syntax as if you don't want to actually create a closure object, and validate callability at the point of creation. For example, creating this first class callable also implies that you have to autoload the class for a static method. If you have some kind of like large definition of of handler, of static handler methods for routes or something like that, then using the first class callable syntax would imply that you have to immediately create closure objects for all of these and immediately load all those classes. That's a use case where you might want to stick with the old syntax.

Derick Rethans 10:01

But wouldn't opcache resolve that issue really?

Nikita Popov 10:04

No, opcache is really exactly the reason why you wouldn't want to do that. For example, for my fastroute library, I cache all the data as a static array. And that's something that OpCache can cache very efficiently because it's in shared memory and accessing it is essentially zero cost. If you include something like first class callables in it, then those have to always be created at runtime, because we don't have concept like, like a persistent object. That means that this can no longer, I mean, the whole script can be in shared memory, but it still has to be executed always at runtime to construct the whole data structure. And that's going to be less efficient. To give a more clear answer to your question is that the first class callable syntax has a cost when creating the callable, and if you are in a situation where avoiding that cost is really critical for performance, that's why you wouldn't want to use it.

Derick Rethans 11:00

And instead you'd have to use the old scalable syntax that we already have.

Nikita Popov 11:03

Exactly. So for that reason, I think that the old syntax is not going to be removed in the near future at least, though maybe we can deprecate certain aspects of it. For example, the syntax also allows you to do highly context dependent things like referencing self, which is even worse than the situation with a private method, because self could refer to something different every time you call it. Those are some things we might want to deprecate early, but the main syntax itself was probably going to stay for a while.

Derick Rethans 11:34

Because callability is checked when you create the closures does that mean it also checks for strictness then? If your PHP file has been declared with strict types?

Nikita Popov 11:45

Strictness is handled the same as with closureFromCallable.The strictness is still determined at the time where the call is made, not where the callable was created, which actually, I am not a fan of how PHP handles strict types together with dynamic calls. But that's like a pre existing problem. And this isn't touching on this.

Derick Rethans 12:06

The language has many issues that probably could have been done better if it was designed from scratch. But that ship has sailed 26 years ago.

Nikita Popov 12:15

The strict types are not quite that old.

Derick Rethans 12:17

No, that is true. The language itself is of course.

Derick Rethans 12:23

Okay, thank you very much then, for taking the time this morning to talk to me about first class scalable syntax.

Nikita Popov 12:29

Thanks for having me, Derick.

Derick Rethans 12:30

Thank you for listening to this installment of PHP internals news, a podcast dedicated to demystifying the development of the PHP language. I maintain a Patreon account for supporters of this podcast as well as the Xdebug debugging tool. You can sign up for Patreon at If you have comments or suggestions, feel free to email them to Thank you for listening and I'll see you next time.

PHP Internals News: Episode 91: is_literal

PHP Internals News: Episode 91: is_literal

In this episode of "PHP Internals News" I chat with Craig Francis (Twitter, GitHub, Website), and Joe Watkins (Twitter, GitHub, Website) about the "is_literal" RFC.

The RSS feed for this podcast is, you can download this episode's MP3 file, and it's available on Spotify and iTunes. There is a dedicated website:


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 91. Today I'm talking with Craig Francis and Joe Watkins, talking about the is_literal RFC that they have been proposing. Craig, would you please introduce yourself?

Craig Francis 0:34

Hi, I'm Craig Francis. I've been a PHP developer for about 20 years, doing code auditing, pentesting, training. And I'm also the co-lead for the Bristol chapter of OWASP, which is the open web application security project.

Derick Rethans 0:48

Very well. And Joe, will you introduce yourself as well, please?

Joe Watkins 0:51

Hi, everyone. I'm Joe, the same Joe from last time.

Derick Rethans 0:56

Well, it's good to have you back, Joe, and welcome to the podcast Craig. Let's dive straight in. What is the problem that this proposal's trying to resolve?

Craig Francis 1:05

So we try to address the problem where injection vulnerabilities are being introduced by developers. When they use libraries incorrectly, we will have people using the libraries, but they still introduce injection vulnerabilities because they use it incorrectly.

Derick Rethans 1:17

What is this RFC proposing?

Craig Francis 1:19

We're providing a function for libraries to easily check that certain strings have been written by the developer. It's an idea developed by Christoph Kern in 2016. There is a link in the video, and the Google using this to prevent injection vulnerabilities in their Java and Go libraries. It works because libraries know how to handle these data safely, typically using parameterised queries, or escaping where appropriate, but they still require certain values to be written by the developer. So for example, when using a query a database, the developer might need to write a complex WHERE clause or maybe they're using functions like datediff, round, if null, although obviously, this function could be used by developers themselves if they want to, but the primary purpose is for the library to check these values.

Derick Rethans 2:05

That is a method of doing it. What is this RFC adding to PHP itself?

Craig Francis 2:09

It just simply provides a function which just returns true or false if the variable is a literal, and that's basically a string that was written by the developer. It's a bit like if you did is_int or is_string, it's just a different way of just sort of saying, has this variable been written by the developer?

Derick Rethans 2:28

Is that basically it?

Craig Francis 2:30

That's it? Yeah.

Joe Watkins 2:32

It would also return true for variables that are the result of concatenation of other variables that would pass the is literal check. Now, this differs from Google, because they introduced that at the language level, but not only at the language level, at the idiom level. So that when you open a file that's got queries in PHP, commonly, if they're long, basic concatenation is used to build the query and format it in the file so that it's readable. So that it wouldn't really be very useful if those queries that you see everywhere in stuff like PHPMyAdmin, and WordPress, and Drupal and just normal code weren't considered literal, just because they're spread over several lines with the concatenation operator. It's strictly not just stuff that's written by the programmer, but also stuff that was written by the programmer or concatenated, with other stuff that was written by the programmer.

Derick Rethans 3:33

Now in the past, we have seen something about adding taint supports to PHP, right? How is this different, or perhaps similar, to taint checking?

Craig Francis 3:44

At the moment today, there is a taint extension, which is something you need to go out your way to install, and actually learn about and how to use. But the main difference is that taint checking goes on the basis of say, this variable is safe or unsafe. And the problem is that it considers anything that had been through an escaping function like html_entities as safe. But of course, the problem is that escaping is difficult. And it's very easy to make mistakes with that. A classic example is if you take a value from a user, an SSH SSH, their homepage URL, if you use HTML encoding, and then put it into the href attribute of a link, that can also result in HTML injection vulnerability, because the escaping is not aware of the context which is used. Because if the evil user put in a JavaScript URL, that is in inline JavaScript, that has created a problem because taint checking would assume that because you use HTML encoding it is safe, and all I'm saying is that is it creates a false sense of security. And by stripping out all that support for escaping, it means that you can focus on libraries doing that work because they know the context, they understand the domain, and we can just keep it a much simpler, and much safer approach.

Derick Rethans 5:02

Would you say that the is_literal feature is mostly aimed at library authors and not individual developers?

Craig Francis 5:09

Yeah, exactly. Because the library authors know what they're doing. They're using well tested code, many eyes over it. The problem libraries have at the moment is that they trust the developer to write things themselves. And unfortunately, developers introduce a lot of injection vulnerabilities with those strings before they even get into the library.

Derick Rethans 5:30

How would a library deal with with strings that aren't literal then?

Craig Francis 5:35

So it really depends on each individual example. And the RFC does include quite a lot of examples of how each one will be dealt with. The classic one is, let's say you're sorting by a column in a database, because if we're dealing with SQL, the field name might come from the user. But that is also quite a risky thing to do if you start including whatever field name the user wrote. So in the RFC, I've created a very simple example where the developer would create an array of fields that you can sort by, and then whatever the user provides, you search through that array, and you pull out the one that you that matches and is fine. And therefore you are pulling out a literal and including into the SQL. To be fair, these ones are quite unique. And each one needs to be dealt with in its own way. But I've yet to find an example where you can't do it with a literal. Having said that, I think Larry Garfield actually gave an example where a content management system changed its database structure. And the way that would work is the library would have to deal with it, they would receive the value for a field, and then that field would be escaped and treated as a field, it understands it as a field, and it will process it as such, then it can include into the SQL, knowing full well that everything else in that SQL is a literal, and then it can just build up SQL in its own way internally.

Derick Rethans 6:58

Okay, talking a little bit about the implementation here. Since PHP seven, we have this concept of interned strings, or maybe even before that actually, I don't quite remember. Which is pretty much a flag on each string and PHP that says, this's been created by the engine, or by coconut. Why would strings have to have an extra flag here to remember that it is created by the programmer?

Joe Watkins 7:21

Well, interned does not mean literal. It's an optimization in the engine, should we use strings. We're free to do whatever we want with that. At the moment, it by happenstance, most interned strings are those written by the programmer. If you think about the sort of strings that are written by the programmer, like a class name, when those things are declared internally, by an extension, or by core code, those things are interned as if they were written by the programmer. They don't mean literal, we're free to use interned strings for whatever we want. For example, a while ago, someone suggested that we should intern keys while JSON decoding or unserializing. It didn't happen, but it could happen. And then we'd have the problem of, well, how do we separate out all this other input. There is another optimization attached to interned strings, which is one character strings, where if you type only one character, or you call a Class A or B, or whatever, the permanent interned string will be used. That results in when the chr function is called, that results in the return of that function always being marked as interned. So it would show as literal, which is not a very nice side effect. And that's just a side effect that we can see today. We don't want to reuse the string really, it does need to be distinct. Also, if you're going to concatenate, whether you do it with the VM or a specific function, obviously, you need to be able to distinguish between an interned string and a literal string, which interned means it has a specific life cycle and specific value. And we can't break that.

Derick Rethans 9:00

So there are really two different concepts, is what you're saying, and hence, they need to have a special flag for that?

Joe Watkins 9:06

Yeah, they're very, two very separate concepts. And we don't we don't want to restrict the future of what interned strings may be used for. We don't want to muddy the concept of a literal.

Derick Rethans 9:16

Of course, any sort of mechanism that languages built into solve or prevent injections in any sort of form, there's always ways around it. Theoretically, how would you go around the is_literal checks to still get a user inputted value into something that passes the is_literal check?

Craig Francis 9:36

Generally speaking, you would never need it because the library should know how to deal with every scenario anyway. And it's not that difficult. We're only talking about things like in the database world, you'll be taking value from field names and therefore it should receive field names or table names. And, you know, we are providing a guardrail as a safety net. And what should happen is that the default way in which programmers work should guide them, to do it the right way. We're not saying that you can't do weird things to intentionally work around this. A really ugly version, which you should never do, but use eval and var_export together, it's horrible. But if you are so desperate, you need to get around this. That's what we're doing it. But in reality, we can't find any examples where you'd actually need to do this.

Joe Watkins 10:22

I would say that, hey, there's this idea that most people writing PHP are using libraries, and they're using frameworks. I don't actually find that to be true. I've been working in PHP for a long time. And most of the big projects I've worked on for a long time did not start out using frameworks. And they did not start out using libraries. They look a bit like that today, but their core, they are custom. There may be a framework buried in there. But there is so much code that the framework is a component and is not the main deal. Most code, we actually do write ourselves, because that's what we're paid to do. I think we don't decide how people are going to use it, and we don't decide where they're going to use it. The fact is, like Craig said, it's a guardrail that you can work around easily. And if you find a use case for doing that, then we shouldn't prejudge, and say, well, that's the wrong thing to do. It might not be the wrong thing to do. For example, an earlier version of the idea included support for integers. We considered integers safe, regardless of their source. If you wanted to do that, in your application, you could do that very easily and still retain the integrity of the guardrail is not compromised. I wouldn't focus on this is for libraries, and this is for frameworks, because these things become so small in the scheme of things that they're meaningless. I mean, most of the code we work on is code that we wrote, it is not frameworks.

Derick Rethans 11:48

That also nicely answers my next question, which is what's happened to integers, which have now nicely covered. The RFC talks about that as hard to educate people to do the right thing. And that is_literal is more focused, so to say, on libraries, and perhaps query building frameworks as the RFC alludes to. But I would say that most of these query building tools or libraries already deal with escaping from input value. So why would it make sense for them to start using is_literal if you're handling most of these cases already anyway?

Craig Francis 12:24

If you look at the intro of the RFC, there's a link to show examples of how libraries currently receive the strings. And you're right about the Query Builder approach is a risky thing, I would still argue it's an important part. That's why libraries still provide them. Doctrine has a nice example of DQL. The doctrine query language is an abstraction that they've created, which is also vulnerable to injection vulnerabilities. And it gives the developer a lot more control over a very basic API. I still think people should try and use the higher level API's because they do provide a nice safe default, but that depends on which library use, they're not always safe by default. So for example, when you're sort of saying: I want to find all records where field parameter one, is equal to value two, a lot of the libraries assumed that the first parameter there is safe and written by the developer. They can't just necessarily simply escape it as though it's a field because that value might be something like date, bracket, field, bracket, and it's sort of relying on the developer to write that correctly, and not make any mistakes. And that hasn't proven to be the case, you know, they do include user values in there.

Derick Rethans 13:43

Just going back a little bit about some of the feedback, because feedback to the RFC has happened for quite some time now. And there were lots of different approaches first tried as well, and suggested to add additional functions and stuff like that. So what's been the major pushback to this latest iteration of the RFC?

Joe Watkins 14:01

So I think the most pushback has come from an earlier suggestion that we could allow integers to be concatenated and considered literal. We experimented with that, and it is possible, but in order to make it possible, you have to disable an optimization in the engine, that would not be an acceptable implementation detail for Dmitri. It turns out we didn't actually, we don't need to track their source technically, but it made people extremely uncomfortable when we said that, and even when we got an independent security expert to comment on the RFC, and he tried to explain that it was no problem, but it was just not accepted by the general public. I'm not sure why.

Derick Rethans 14:45

All right. Do you have anything to add Craig?

Craig Francis 14:48

The explanation given by people is they liked the simpler definition of what that was as if it's a string written by the developer. Once you start introducing integers from any source, while it is safe, it made people feel, yeah, what is this. And that's where we also had the slight issue because we had to find a new name for it. And I did the silly thing of sort of asking for suggestions, and then bringing up a vote. And then we had, I think it's 18 to three people saying that it should be called is_trusted, and you have that sinking moment of going, Oh, this is going to cause problems, but hey, democracy. It creates that illusion that it's something more. So that's why we sort of went actually, while I like Scott's idea of having the idea of maybe calling it is_noble. It is a vague concept, which people have to understand. And it's a bit strange. Whereas going back to the simpler, original example, they've all seem to grasp grasp of that one. And we could just keep with the original name of is_literal, which I've not heard any real complaints about.

Derick Rethans 15:53

I think some people were equivalenting is_trusted with something that we've had before in PHP called Safe mode, which was anything but of course.

Craig Francis 16:02

Yes, no, definitely.

Derick Rethans 16:03

We're sort of coming to the end of what to chat about here. Does the introduction of is literal introduce any BC breaks?

Craig Francis 16:11

Only if the user land version of is_literal, which I'm fairly sure is going to be unlikely. So on dividing their own function called that.

Derick Rethans 16:18

Did you check for it?

Craig Francis 16:20


Derick Rethans 16:21

So if you haven't found it, then it's unlikely to to exist.

Craig Francis 16:24

There are still private repositories, we can't shop through all their show, check through all their code. But yeah.

Derick Rethans 16:29

Did I miss anything?

Craig Francis 16:31

We covered future scope, which is the potential for a first class type, which I think would be useful for IDs and static analysers. But this is very much a secondary discussion, because that could build on things like intersection types, but we still need to focus on what the flag does. And there's also possibility of using this with the native functions themselves, but we do have to be careful with that one, because, you know, we got things like PHPMyAdmin. We have to be able to make the output from libraries as trusted because they're unlikely to still be providing a literal string at the end of it. So that's a discussion for the future. And the only other thing is that, you know, the vote ends on the 19th of July.

Derick Rethans 17:08

Which is the upcoming Monday. How is the vote going? Are you confident that it will pass?

Craig Francis 17:13

Not at the moment, we're sort of trying to talk to the people who voted against it. And we've not actually had any complaints as such. The only person who sort of mentioned anything was saying that we should rely on documentation and the documentation is already there. And it's not working. I think a lot of people just voted no, because they just sort of going well, that's the safe default. I don't think it's necessary. Or, you know, I'd like the status quo. And we still are trying to sell the idea and say: Look, it's really simple. It's not really having a performance impact. And it can really help libraries solve a problem, which is actually happening.

Derick Rethans 17:46

Is this something that came out of the people that write PHP libraries or something that you came up with?

Craig Francis 17:52

So I've come gone to the library authors and suggested you know, this is how Google do it. Would you like something similar? And we've certainly had red bean and Propel ORM saw show positive support for that. And I've also talked to Matthew Brown, who works on the Psalm static checking analysis. He's very positive about it, so much so that Psalm now also includes this as well. Obviously, static analysis is not going to be used by everyone. So we would like to bring this back to PHP so that libraries can use it without relying on all developers using static analysis.

Derick Rethans 18:25

Thank you very much. Glad that you were both here to explain what this is_literal RFC is about.

Craig Francis 18:31

Thank you very much, Derick.

Joe Watkins 18:33

Thanks for having us.

Derick Rethans 18:37

Thank you for listening to this installment of PHP internals news, a podcast dedicated to demystifying the development of the PHP language. I maintain a Patreon account for supporters of this podcast as well as the Xdebug debugging tool. You can sign up for Patreon at If you have comments or suggestions, feel free to email them to Thank you for listening and I'll see you next time.

PHP Internals News: Episode 90: Read Only Properties

PHP Internals News: Episode 90: Read Only Properties

In this episode of "PHP Internals News" I chat with Nikita Popov (Twitter, GitHub, Website) about the "Read Only Properties" RFC.

The RSS feed for this podcast is, you can download this episode's MP3 file, and it's available on Spotify and iTunes. There is a dedicated website:


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 90. Today I'm talking with Nikita Popov about the read only properties version two RFC that he's proposing. Nikita, would you please introduce yourself?

Nikita Popov 0:33

Hi, Derick. I'm Nikita and I do PHP core development work by JetBrains.

Derick Rethans 0:39

What does this RFC proposing?

Nikita Popov 0:41

This RFC is proposing read only properties, which means that the property can only be initialized once and then not changed afterwards. Again, the idea here is that since PHP 7.4, we have typed properties. A remaining problem with them is that people are not confident making public type properties because they still ensure that the type is correct, but they might not be upholding other invariants. For example, if you have some, like additional checks in your constructor, that string property is actually a non empty string property, then you might not want to make it public because then it could be modified to an empty value for example. One nowadays fairly common case is where properties are actually only initialized in the constructor and not changed afterwards any more. So I think this kind of mutable object pattern is becoming more and more popular in PHP.

Derick Rethans 1:35

You mean the immutable object?

Nikita Popov 1:37

Sorry, immutable. And read only properties address that case. So you can simply put a public read only typed property in your class, and then it can be initialized once in the constructor and you can be... You don't have to be afraid that someone outside the class is going to modify it afterwards. That's the basic premise of this RFC.

Derick Rethans 1:57

But it also means that objects of the class itself can modify that value any more, either.

Nikita Popov 2:01

Exactly. So that's, I think, a primary distinction we have to make. Genuinely, there are two ways to make this read only concept work. One is like actually read only or maybe more precisely init once, which is what this RFC proposes. We can only set that once and then even in the same class, you can't modify it again. And the alternative is the asymmetric visibility approach where you say that, okay, only in the public scope, the property can only be read, but in the private scope, you can modify it. I think the distinction there is very important, because read only property tells you that it's genuinely read only, like, if you access a property multiple times in sequence, you will always get back the same value. While the asymmetric visibility only says that the public interface is read only, but internally, it could be mutated. And that might like be, you know, intentional, just that you want to like have your state management private, but that the property is not supposed to be immutable.

Derick Rethans 3:05

How's this RFC different from read only properties, version one?

Nikita Popov 3:09

Read only properties version one was called write once properties. I think the naming is kind of one of the more important differences. The new RFC is also effectively write once, but I think it's really important to view it from an API perspective as read only because that's what the user gets to see. While write once gives you this impression, that is know that you can externally from outside the class like passing the value once I know like dependency injection, that is what they would think of when they hear write ones. And from the technical site, there is a related difference. And that difference is that new RFC only allows you to initialize read only properties inside the class scope. That means if you do something really weird, like leaving a property uninitialized in the constructor, it's not possible for someone outside the class to initialize it instead, just like an extra safety check. Of course, you can, as usual bypass that, like if you're writing a serializer or hydrator, you can use reflection to initialize it outside the class. But normally you won't be able to.

Derick Rethans 4:13

Does that mean that these read only properties can also be initialized from a normal method instead of just from the constructor?

Nikita Popov 4:19

That's true. Yes, that's possible.

Derick Rethans 4:21

So the RFC talks about that a read only property cannot be assigned from outside a class. Does that mean it can be set by a different object of the same class?

Nikita Popov 4:29

Yeah, that's how scoping works in PHP. Scoping is always class based, not object based, it's a common misconception that if you have like private scope, you can access different objects of the same class.

Derick Rethans 4:42

That was a surprise to me the first time I ran into that, but once you know, it's obvious that it's should work all over the place right. Now, the RFC states that you can only use read only with typed properties. Why is that?

Nikita Popov 4:55

So this is related to the initialisation concept, that typed introduced. Typed properties start out, if you don't give them a default value, they start out in an uninitialized state. And we reuse that state for read only properties. You can only assign to the property while it's uninitialized. And once it's initialized, you cannot assigned to it any more or even unset it back to an uninitialised state. For non typed properties, you also can get into this uninitialized state by explicitly unsetting it. But the problem is that this is not the default state. Untyped properties always have no default value, even if you don't specify one, which effectively means that these properties are always initialized. So they will be kind of useless if you used read only with them. Which is why we make this distinction to avoid any confusion. And if you want to use an untyped read only property, you do that by using the mixed type, which is the same but has the initialisation semantics of typed properties.

Derick Rethans 5:56

What would that mean if you have say, a resource or class typed property with a read only keyword? Can you not read or write to a resource any more? Or modify properties on an object, that is the value of a read only typed property?

Nikita Popov 6:12

No, you can still modify those, because we have to distinguish the concepts of like exterior and interior mutability here. So objects and resources are... Well, I mean, we often say they are passed by reference, which is not strictly true, because those are not PHP references. But the important part is that they only pass around some kind of handle and you can still modify the inside of that handle. What you can't do is you can't reassign to a different resource or reassign to a different object, it's always the same object, the same resource, but the insides of the objects, those can change. Of course, if your object also only contains read only properties, then you won't be able to change this.

Derick Rethans 6:55

Okay, and that answered another question that I have: how it possible to make the whole object read only or on all of its properties? Where the answer is by setting all the properties to read only.

Nikita Popov 7:05

We could like add a read only class modifier that makes all the properties implicitly read only but maybe future scope.

Derick Rethans 7:13

Is there a reason why read only properties can't have a default value?

Nikita Popov 7:18

So this is, again, same issue with initialization. If they have a default value, then they're already initialized. So you can't overwrite it. Like we could allow it, but you just would never be able to change them from the default value, which is something we could allow it just wouldn't be very useful.

Derick Rethans 7:36

in PHP, eight zero PHP introduced promoted or constructor promoted properties, I think it's the full name of it. How does this read only property tie in with that? Because can you set the read only flag on a constructor promoted property?

Nikita Popov 7:49

Yeah, you can. So that works as expected. Important bit there is that with the promoted properties, you can set the default value. And the reason is that for promoted properties, the default value is not the default value of the property. It's the default value of the parameter. And that works just fine.

Derick Rethans 8:07

But again, it would be a constant.

Nikita Popov 8:10

No, in that case, it's not a constant, because it's just a default value for the parameter and the code we generate, we just assign this parameter to the property IN the constructor.

Derick Rethans 8:19

Which means that if you instantiate the class with different arguments, they would of course, override the default value of the constructor argument value, right?

Nikita Popov 8:28

That's right. If you pass it explicitly, then we use the explicit value. If you don't pass an argument, then we use the default of the argument, but it still gets assigned to the property through the like automatically generated code for the promotion.

Derick Rethans 8:42

Promotion isn't actually... there isn't actually really a language feature, but more of a copy and paste mechanism.

Nikita Popov 8:50

Yes, this is pure bit of syntax sugar.

Derick Rethans 8:53

Which sometimes can be handy. But all kinds of interesting things that we added to properties or type system in general, usually inheritance comes into play. Are there any issues here with read only and inheritance? What does it do to variants or traits or things like that, all the usual things that we need to take into consideration?

Nikita Popov 9:13

Most of our rules for properties, most of our inheritance rules for properties are invariant, which means that the property in the child class has to basically look the same as the property in the parent class. And this is also true for read only properties. So we say that if the parent property is read only the child property has to be read only, and the same the other direction and for the type as well. So we say that the type of the read only property has to match with the parent property. Those rules are very conservative and like they could theoretically maybe be relaxed in some cases. For read only properties on read only is like kind of return type and return types in PHP are covariant. So one could argue that the type should also be covariant here The problem is that like here, we get into this little detail that read only is really not read only, but init once. So there is this one assignment. And assignment is more like an argument. So is contravariant. So if we made them covariant, then we would basically lie about this one initializing assignment. And we don't like to lie about these things, at least without some further consideration.

Derick Rethans 10:24

Read only doesn't change the variance rules at all, which is different from the property accessors RFC we spoke about a couple of months ago. Talking about that this is actually a competing RFC, or do they tie together?

Nikita Popov 10:37

Well, I should first say that probably the variance stuff defined in the accessors RFC maybe isn't right, for the same reason and should also like just keep things invariant. But it's more complicated there because it also has abstract properties, or abstract accessors. And then the abstract case, that's where the different variance rules are safe. To get back to your question, they are kind of competing, but could also be seen as just complimentary. So read only is basically a small subset of the accessors feature. I mean, accessors also allow you to implement read only properties as part of a much more general framework. And read only properties cover like only this small but very useful corner in a way that's much simpler. And I think that's not just simpler from a technical perspective, but also simpler to understand for the programmer, because there are a lot of less edge cases involved.

Derick Rethans 11:32

Because we're getting pretty close to feature freeze now. Are you still intending to take the property accessors RFC forward for PHP eight one?

Nikita Popov 11:41

No, definitely not.

Derick Rethans 11:43

But you have taken the read only properties one forwards because we're already voting on it. At the moment, it looks like the vote is going to pass quite easily. So there's that. We don't have to speculate about that, like we had to do with previous RFCs. Do you think I've missed anything talking about read only properties? Or have we covered everything?

Nikita Popov 12:01

We've missed one important bit. And that's the cloning. So the read only properties RFC is not entirely uncontroversial. And basically all the controversy is related to cloning. The problem are wither methods as they're used by PSR seven, for example, which are commonly implemented by cloning the original object, and then modifying one property in it. This doesn't work with read only properties, because you know, if you clone the object, then it already has all the properties initialized. So if you try to change something, then you'll get an error that you're modifying read only property. So these patterns are pretty fundamentally incompatible. There are possible solutions to the problem primarily a dedicated syntax that goes into the code name of cloneWith where you clone an object, but override certain properties directly as part of the syntax, which could bypass the read only property checks. The contention here is basically there are some people consider this clone support and these wither methods are very important. Well, I mean, read only properties are about immutable objects. And like PSR seven is a fairly popular read only object pattern, they think that without support for cloning or for withers, the feature loses all value. That's why the alternative suggestions are either to first introduce some more cloning functionality. So like this mentioned, cloneWith or instead implement asymmetric visibility, which does not have this problem because inside the class, you can always modify the property. So it works perfectly fine with cloning. My personal view on this is well I personally use cloning in PHP approximately never. So this is not a big loss for me, and I'm happy for this from my perspective, edge case, to be addressed at a later time. I can understand that other people put more value on this then I do.

Derick Rethans 13:55

There doesn't seem to be enough push against that for this RFC to fail. So there is that. Thanks Nikita for explaining the read only properties RFC which will very likely see in PHP eight one. Thanks for taking the time.

Nikita Popov 14:08

Thanks for having me Derick, once again.

Derick Rethans 14:13

Thank you for listening to this installment of PHP internals news, a podcast dedicated to demystifying the development of the PHP language. I maintain a Patreon account for supporters of this podcast as well as the Xdebug debugging tool, you can sign up for Patreon at If you have comments or suggestions, feel free to email them to Thank you for listening and I'll see you next time.

PHP Internals News: Episode 89: Partial Function Applications

PHP Internals News: Episode 89: Partial Function Applications

In this episode of "PHP Internals News" I chat with Larry Garfield (Twitter) and Joe Watkins (Twitter, GitHub, Blog about the "Partial Function Applications" RFC.

The RSS feed for this podcast is, you can download this episode's MP3 file, and it's available on Spotify and iTunes. There is a dedicated website:


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 89. Today I'm talking with Larry Garfield and Joe Watkins about a partial function application RFC that they're proposing with Paul Crevela and Levi Morrison. Larry, would you please introduce yourself?

Larry Garfield 0:36

Hello World. I'm Larry Garfield or Crell on most social medias. I'm a staff engineer for Typo3 the CMS. And I've been getting more involved in internals these days, mostly as a general nudge and project manager.

Derick Rethans 0:52

And hello, Joe, would you please introduce yourself as well?

Joe Watkins 0:55

Hi, I'm Joe, or Krakjoe, I do various PHP stuff. That's all there is to say about that really.

Derick Rethans 1:02

I think you do quite a bit more than just a little bit. In any case, I think for this RFC, you, you wrote the implementation of it, whereas Larry, as he said, did some of the project management, I'm sure there's more to it than I've just paraphrased in a single sentence. But can one of you explain in one sentence, or if you must, maybe two or three, what partial function applications, or I hope for short, partials are?

Larry Garfield 1:27

Partial function application, in the broadest sense, is taking a function that has some number of parameters, and making a new function that pre fills some of those parameters. So if you have a function that takes four parameters, or four arguments, you can produce a new function that takes two arguments. And those other two you've already provided a value for in advance.

Derick Rethans 1:54

Okay, I feel we'll get into the details in a moment. But what are its main benefits of doing this? What would you use this for?

Larry Garfield 2:01

Oh, there's a couple of places that you can use partial application. It is what got me interested. It's very common in functional programming. But it's also really helpful when you want to, you have a function that like, let's say, string replace takes three arguments, two of which are instructions for what to replace, and one of which is the thing in which you want to replace. If you want to reuse that a bunch of times, you could build an object and pass in constructor values and save those and then call a function. Or you can just partially apply string replace with the things to search for, and the things to replace with and get back a function that takes one argument and will do that replacement on it. And you can then reuse that over and over again. There are a lot of cases like that, usually use in combination with functions that wants a callback. And that callback takes one argument. So array map or array filter are cases where very often you want to give it a function that takes one argument, you have a function that takes three arguments, you want to fill in those first ones first, and then pass the result that only takes one argument to array map or a filter, or whatever. So that's the one of the common use cases for it.

Derick Rethans 3:15

That's the benefits and some of its background comes from functional programming, as you've just mentioned. What is the syntax that you're proposing and some of the semantics?

Larry Garfield 3:26

The syntax that we've developed, are two placeholders that you can use in a function call. So if you're calling a function as you normally would, but for one of the arguments, you pass a question mark, or at the tail end, you have an ellipsis (dot dot dot), then that tells the engine: This is not a function call. This is a partial application. And what it will do is return not the result of the function but return a closure object that has the the arguments that correspond to those question marks. And then when called with those arguments, we'll pass those along with the original function. Probably easier to explain, if I use a concrete example, using the string replace example we talked about before, you would call it with str_replace, the example from the RFC, hello, hi, question mark. What that gives you is a callable, a closure that has one argument, which will take its type and name from str_replace. So the third argument to str_replace essentially gets copied into that closure. And what closure does internally when you call it with that one argument is it just calls string replace with hello, hi, and whatever argument you gave it and returns that value. It is conceptually very, very similar to just writing a short lambda or an arrow function that takes one arguments and calls string replace hello, hi, and that argument. In most cases, it ends up functioning almost exactly like that. There's a few subtle differences in a few places. But most of the time, you can think of it working essentially like that. The question mark means one required argument only. The dot dot dot means zero or more arguments, if you want to, say provide the first argument to a function, and then dot dot dot would mean: And then all of the other arguments, however many there are, even if it's that zero, those are what's left, which languages other languages that have partial application as a first class feature, usually end up doing it that way where you can only pre fill from the left. PHP, because the placeholder lets us do it in any order. So we can skip over arguments if we want to, which is quite nice. But it means that you can take a function and reduce it to, I want to prefill just these two arguments and leave these three arguments for the new function, or I want to prefill these arguments from the left, and then everything else, whatever it is, is left. It also lets you do cute things like if you provide all of the arguments to a function, and then just tack on a dot dot dot the end of it, then you get back a closure that takes essentially zero arguments. But when called, will call that other function. So it's lets lets you really easily build a delayed function as you need to.

Derick Rethans 6:15

When do the arguments to the function get evaluated then?

Larry Garfield 6:18

Arguments are evaluated in advance. So this is the subtle difference between partial application and the short lambda syntax. In a short lambda, what happens is, essentially, that entire expression on the right hand side gets wrapped up into a closure. And so any arguments that are compound like they have a function call that is inside one of the placeholders, or one of the arguments, that'll get evaluated later. With partial application, the function that is in a parameter position gets evaluated first and reduced to a value. And that value gets partially applied to the function. 90% of the time, that's not going to be an issue. There are a few cases where doing it one way or the other may be subtly different, but you'll spot those fairly easily.

Derick Rethans 7:02

So the RFC talks about things that you can do, but also a few things that you cannot do or don't want to do yet. What are these things that partials won't support, or run support yet, at least?

Larry Garfield 7:13

The main thing that it doesn't support is named placeholders. You can pre fill a value or an argument with a named named argument. But not a named placeholder. Those have to be positional. Named placeholders are complicated to implement, and run into a question of, if you provide those in a different order, does that also change the order of the arguments in the partially applied function that you get back in that closure? And there's a good argument to be made that either way is logical. And so we're like, no, does not deal with it, too complicated. We'll just positional only. And you cannot specify an optional arguments either. It's just again, too complicated. Things get too weird. If you have those advanced cases, use our short lambda, that works just fine. If you want to just make a new function that defers to a new function, and change its API in the process, short lambda works fine. And it's still quite short.

Derick Rethans 8:13

I know the RFC talks a little bit about references, but I don't like talking about references. So let's skip that part. In my opinion, they should be removed from the language. But I know we can't.

Larry Garfield 8:22

There's occasionally used for them. But very occasionally.

Derick Rethans 8:25

There's a bunch of technical things that I also want to chat about. And hopefully, Joe, if you want to fill in, I'd be more than welcome to hear your opinions on these things. But the first one is that PHP has this thing called func_get_args. How does that work with these partials? How does that tie in together?

Joe Watkins 8:42

It should mostly behave as if you've invoked the function directly. We don't want there to be a huge discrepancy between. The callee know whether they've been called through partial application or complete application. It should be the same.

Derick Rethans 8:58

That is good to know. I mean, I always like it how things work as people expect them to work, right?

Joe Watkins 9:03


Derick Rethans 9:04

We already have used the dot dot dot operator for variadics. But you're reusing the dot dot dot, or ellipses, as you more eloquently call it earlier. Here again, as well, is that not going to cause issues? Or does that tie in well together?

Joe Watkins 9:18

Well, there's quite a lot of debate about what's the right symbol to use. I think it's dot dot dot, and I think Larry agrees with me. But there's some people who want to stick an extra question mark on the end, which to me looks like it reads zero to one. And to Larry, it looks like an extra character that's just not needed. Other people say it makes sense for them. But if you can type three characters and not four, I mean, you need a really good argument. The arguments that have been put forward so far don't really make very much sense for me. Maybe we should ask that question and it doesn't really matter. In the end, what the syntax is, is if it's a difference between it getting in and not getting in, then we'll just put the extra question mark on there. I don't really have a really good argument to change it like to be like that.

Derick Rethans 10:05

To be honest, to me, it looks like you then have two placeholders.

Joe Watkins 10:09


Derick Rethans 10:10

I don't feel the need for it.

Joe Watkins 10:11

That's also another argument because we've introduced this one symbol, and then this other symbol, and then you put them together. And that's two things. I mean, you can't have one and one equals one.

Derick Rethans 10:20

Fair enough. The RFC does touch on another quite interesting thing, I think, which is constructors, which it also be able to partially apply. But of course, you've mentioned that, that arguments get applied immediately when you do the substitution, when you do the partial application. But of course, the constructor is a bit weird because a constructor runs immediately after an object has been constructed. So how does that work together with partials?

Joe Watkins 10:47

So at first, we made it so like if you invoke a constructor with reflection, and you just invoke it over and over again, it'll invoke it on the same object, you won't get back a new object. It's not the constructor that returns the object, it's the new operator. So first, we had a bit dumb. And we did just like what reflection does. And if you applied to a constructor, you'd get back a closure that just repeatedly invokes the constructor, which is, as Larry called it, quite naive. So we went back and revisited that. And so now it acts like a factory. Every time you invoke the closure return from an application, you get a brand new object, which is more in line with what people expect. And it's also quite cool. It's one of my favourite bits, actually as it turns out.

Derick Rethans 11:31

In my opinion, it also makes more sense than then having an apply to the same object over and over again. Whether I'd like it or not, I don't know yet.

Joe Watkins 11:39

Oh, the other option is traditional constructors to avoid the surprising behaviour. But that would be just a strange.

Larry Garfield 11:45

There are a lot of use cases where you want to take a bunch of values, convert them to objects using an array map, supporting constructors for that makes total sense to me.

Derick Rethans 11:54

And I would probably say, though, that I would prefer not allowing it over it applying over the same object over again. You've touched a little bit on some common cases where you want to use this, do you perhaps have some other ideas where this might be really useful?

Larry Garfield 12:10

So there's three use cases that we think are probably going to be the lion's share. One is to just use the dot dot dot operator. So you have some function or method call, call it with dot dot dot, and that's it. You prefill nothing, which gives you back a closure that is identical in signature to the function or the method that you're applying it to. Everything we've said about functions applies the methods here as well. Which means we now effectively have a new way to refer to a function or a method and make a callable out of it, that doesn't involve just sticking it into a string. You just say, hey, function called dot dot dot, or an arrow bar, parentheses, dot dot dot, parentheses. And now you can turn any function or method into a callable and pass that around. And it's still, it's not wrapped up into the silly array format, it's still accessible to static analysers and refactoring tools. Hopefully, with this, you will never need to refer to a function name using a string ever again, never refer to a method call as an array of object and method. So that that just is not needed any more in the vast majority of cases.

Derick Rethans 13:20

That alone is probably worth having them, maybe.

Larry Garfield 13:23

And Nikita had an RFC that was doing just that, and nothing else. It's kind of a junior version of this. I don't think that's necessary, the full full scope here works, and gives us that. The second use case that I think is going to be common are unary functions. That's functions that take a single argument. More to the point, as I mentioned before, a lot of functions take a callback. And that callback needs a single argument, array map, array filter, some validation routines, a lot of other things like that. So it's now stupidly easy to take any arbitrary function or method and turn it into a single parameter function, which you can then pass as a callback to array map, array filter, all these other tools, and it just becomes really easy to pre fill things that way. The third is the other one I mentioned earlier, if you pre fill all the arguments, and then just put a dot dot dot at the very end, which means zero or more, you now have a function that takes no arguments, but calls the original function you specified with all the arguments you specified. This often the case for default values, where I want to have a default value available, but don't want to take the time to compute it in advance because it might be expensive. Whatever function it is that will determine that default value, I just partially apply that and give it all the arguments and I get back a callable. That creating a callable is dirt cheap, but when I actually need that value, I can then call it at that time, but it won't actually get called unless I need it. That's another use case that we expect to be common. There are no doubt others that we haven't thought of, or that will be less common, but still useful. I think this will probably replace a large chunk of the use cases for short lambdas. Not because short lambdas are bad, they're wonderful. But so many of them convert a function to a simpler function. And this gives us an even more compact, more readable syntax for that, with even less extra symbols and flotsam around it.

Derick Rethans 15:24

I saw, hopefully as a joke, saying that, instead of using the question mark, we should use dollar sign dollar sign, and then we should call the token name T_BLING.

Larry Garfield 15:36

This RFC actually has a storied history. Several years ago, Sara Golemon had proposed porting the pipe operator from Hack to PHP. The pipe operator is an operator available in a lot of different languages that lets you string together a series of functions. So you pass a function, pass an argument into one function, its results you pass to the next function, its results, you pass the next function and so on, which is a good case for unary functions. In Hack's syntax, they don't use a function on the right hand side, they use an arbitrary expression, and then dollar dollar as a placeholder for where to put the value from the left hand side from the previous step. It's the only language that does that.

Derick Rethans 16:20

The other language that does it is bison.

Larry Garfield 16:23

Or Bison also does that style of?

Derick Rethans 16:25

It does something weird like that, yeah. Have a look at the grammar file.

Larry Garfield 16:29

I've looked in there. It's scary. So at the time, she didn't actually put an implementation in for it. But there was some discussion about it. I joked that if she wanted to do that, she should call it T_BLING. And she thought it was hilarious, but never went anywhere. A year ago, I started working on a pipe operator RFC that did just the pipe part, but used a callable on the right hand side, instead of an expression, more like F#, and Haskell, and other languages that have a pipe operator. And their main response to that was, we'd like this, this is cool, except that just using short lambdas on the right all the time to make unaries is too ugly. We want partial application first. So I spent a while trying to bribe someone with more experience and knowledge than me to work on partial application. I tried bribing Ilya Tovolo, to do so by working with him on enumerations. And we got enumerations in, but he doesn't have the time to work on partial application. Levi and Paul had already written an RFC for partial application that had no implementation. It's just a skunkworks, essentially. Then a few weeks ago, Joe pops up and starts working on an implementation for partials. And I, to this day, don't know what interested him in it. But I'm very happy about this fact. So as we updated the RFC, I knew that people want a bike shed about syntax. So I threw that in as a joke. I don't think we're actually going to do that. It's just a little inside reference that is now no longer inside.

Derick Rethans 17:56

Joe what made you work on partials, then?

Joe Watkins 17:58

It's interesting to write. I've had my fun whether it gets in or not.

Derick Rethans 18:02

Sometimes that's the case, right? So just working on this is all the fun.

Larry Garfield 18:06

Sometimes it's fun to just run down rabbit holes for the heck of it. And sometimes really cool things can come out of that sometimes.

Derick Rethans 18:12

At some point, I might have to implement support for partials like I have for closures in Xdebug as well. Because at some point, people might want to debug these things. So I'm a little bit interested in how do these the closures that it generates? Where does it store the already applied arguments?

Joe Watkins 18:29

So partials have the same binary struct up to this point or of the closure, and then after that there's some extra fields.

Derick Rethans 18:36

Would they still have the names?

Joe Watkins 18:39

No, because named arguments aren't actually named, that information is lost. By the time we've got them, we don't have any name information. We've only got their correct position, according to the call that was made.

Derick Rethans 18:50

And every argument that hasn't been filled and doesn't have a special placeholder in there, or does it keep track of which ones have been filled in?

Joe Watkins 18:56

We've got two special placeholders internally, you won't see as undef or null or anything.

Derick Rethans 19:02

Okay, that's good to know. What has the reaction been so far?

Larry Garfield 19:05

Slightly positive. There were a lot of discussions early on about do we support argument reordering? And should it use a single placeholder or two separate placeholders? Originally, we had one and realized after a while, that doesn't actually work. There're use cases where that will be confusing. Overall, the feedback has been quite positive, and I fully expect that to pass. Really the only question people are still debating about at this point is ellipsis versus ellipses question mark.

Joe Watkins 19:34

Yeah, I think the first version of the RFC was quite well received. Someone said we could document it as to make a partial sprinkle or question mark over it and hope for the best.

Derick Rethans 19:44

Oh, that's good to hear. With feature freeze coming not pretty soon now. When do you think you're putting this up for a vote?

Larry Garfield 19:51

Probably in the next couple of days. The only question I think is whether we include a second question for which variadic placeholder to use, which syntax/ Or if we just say it's dot dot dot, go away. Other than that it should go to a vote probably before this episode airs.

Derick Rethans 20:06

Thank you very much, both of you for taking the time to me today to talk about partials.

Larry Garfield 20:11

Thank you again Derick, hopefully see you once more on this season.

Joe Watkins 20:15

Thanks Derick, see you soon.

Derick Rethans 20:21

Thank you for listening to this installment of PHP internals news, a podcast dedicated to demystifying the development of the PHP language. I maintain a Patreon account for supporters of this podcast as well as the Xdebug debugging tool. You can sign up for Patreon at If you have comments or suggestions, feel free to email them to Thank you for listening, and I'll see you next time.

PHP Internals News: Episode 88: Pure Intersection Types

PHP Internals News: Episode 88: Pure Intersection Types

In this episode of "PHP Internals News" I talk with George Peter Banyard (Website, Twitter, GitHub, GitLab) about the "Pure Intersection Types" RFC that he has proposed.

The RSS feed for this podcast is, you can download this episode's MP3 file, and it's available on Spotify and iTunes. There is a dedicated website:


Derick Rethans 0:14

Welcome to PHP internals news, a podcast dedicated to explaining the latest developments in the PHP language. This is Episode 88. Today I'm talking with George Peter Banyard about pure intersection types. George, could you please introduce yourself?

George Peter Banyard 0:30

Hello, my name is George Peter Banyard. I work on PHP code development in my free time. And on the PHP Docs.

Derick Rethans 0:36

This RFC is about intersection types. What are intersection types?

George Peter Banyard 0:40

I think the easiest way to explain intersection types is to use something which we already have, which are union types. So union types tells you I want X or Y, whereas intersection types tell you that I want X and Y to be true at the same time. The easiest example I can come up with is a traversable that you want to be countable as well. So traversable and countable. Currently, you can do intersection types in very hacky ways. So you can either create a new interface which extends both traversable and countable, but then all the classes that you want to be using this fashion, you need to make them implement the interface, which might not be possible if you using a library or other things like that. The other very hacky way of doing it is using reference and typed properties. You assign two typed properties by reference, one being traversable, one being countable, and then your actual property, you type alias reference it, with both of these properties. And then my PHP will check: does the property respect type A those reference? If yes, move to the next one. It doesn't respect type B, which basically gives you intersection types.

Derick Rethans 1:44

Yeah, I saw that in the RFC. And I was wondering like, well, people actually do that?

George Peter Banyard 1:49

The only reason I know that is because of Nikita's slide.

Derick Rethans 1:51

The thing is, if it is possible, people will do it, right. And that's how that works.

George Peter Banyard 1:56

Yeah, most of the times.

Derick Rethans 1:57

The RFC isn't actually called intersection types. It's called pure intersection types. What does the word pure do here?

George Peter Banyard 2:05

So the word pure here is not very semantic. But it's more that you cannot mix union types and intersection types together. The reasons for it are mostly technical. One reason is how do you mix and match intersection types and union types? One way is to have like union types take precedence over intersection types, but some people don't like that and want to explicit it grouping all the time. So you need to do parentheses, A intersection B, close parentheses, pipe for the union, and then the other type. But I think the main reason is mostly the variance, like the variance checks for inheritance are already kind of complicated and kind of mind boggling.

Derick Rethans 2:44

I'm sure we'll get into the variance rules in a moment. What is it actually what you're proposing to add here. What is the syntax, for example?

George Peter Banyard 2:52

So the syntax is any class type with an ampersand, and any other class type gives you an intersection type, which is the usual way of doing and.

Derick Rethans 3:01

When you say class types, do you also mean interfaces?

George Peter Banyard 3:04

Yes, PHP has a concept of class types, which are mostly any class in any interface. There's also a weird exception where parent and self are considered class types, but those are not allowed.

Derick Rethans 3:20

Okay, so it's just the classes that you've defined and the class that are part of the language but not a special keywords, self and parent and static, I suppose?

George Peter Banyard 3:28

Yes, the reason for that is standard types are not allowed to be part of an intersection, because nothing can be an integer and a string at the same time. Now, there are some of the built in types, which can be kind of true. You could have a callable, which is a string, because callables can be arrays, or can be a closure. But that's like very weird and not very great. The other one is iterable. If when you expand that out, you get redundant types, which we can talk about later. And the final thing is parent, self, and static, just makes for some very weird design questions, in my opinion, like, if you ask for something to be an intersection with itself, you basically can only enforce conditions on subclasses. You have a class and you say: Oh, I want it to return self, but also be countable for some reason, but I'm not countable. So if you extend me, then you need to be countable, but I'm not. So it's very weird. parent has kind of the very same weird semantics where you can ask a parent, but it's like, if the base class doesn't support it, and you ask for a parent to be an intersection, then you basically need the child to implement the interface and then a child to return the first child. If you do that main question. Why? Because I don't see any good reasons to do it. And it just makes everything harder.

Derick Rethans 4:40

You've only added for the sake of completeness instead of it being useful. Let's move on birds. You've mentioned which types are supported, which is class names and interface names. You already hinted a little bit at redundant types. What are redundant types?

George Peter Banyard 4:56

Currently, PHP already does that with union types. If you repeat the type twice in a union, you'll get a compile error. This only affects compiled time known aliases. If you use a use statement, then PHP knows that you basically using the same type. However you use a runtime alias, then it can't detect that.

Derick Rethans 5:13

A runtime alias, what's that?

George Peter Banyard 5:15

So if you use the function class_alias.

Derick Rethans 5:16

It's new to me!

George Peter Banyard 5:18

it technically exists. It also doesn't guarantee basically that the type is minimal, because it can only see those was in its own file. For example, if you say I want A and B, but B is a child class of A, then the intersection basically resolves to only B. But you can only know that at runtime if classes are defined in different files. So the type isn't minimal. But if you do redundant types, basically, it's a easy way to check if you might be typing a bug.

Derick Rethans 5:46

You try to do your best to warn people about that. But you never know for certain.

George Peter Banyard 5:51

You never know for certain because PHP doesn't compile everything into like one big program like in check. Static analyser can help for that.

Derick Rethans 5:59

Let's talk a little bit about technical aspects, because I recommend that implementing intersection types are quite different from implementing union types. What kind of hacks that you have to make in a parser and compiler for this?

George Peter Banyard 6:11

Our parser has being very weird. The parsing syntax should be the same as union types. So I just copy pasted what Nikita did. I tried it. It worked for return types without an issue. It didn't work with argument types, because bison, which is the tool which generates our parser, was giving a shift reduce conflict, which basically tells: Oh, I got two possible states I can go in, and I don't know which branch I need to go, because the PHP parser only does one look ahead. Because it was conflicting, the ampersand, either for the intersection type or for to mark a reference. Normally, if the paster is more developed, or does more look ahead, it is not a conflict. And it shouldn't be. Ilia managed to came up with this ingenious idea, which is just redefine the ampersand token twice and have very complicated names, and just use them in different contexts. And bison just: now I have no issue. It is the same token, it is the same character. Now that you have two different tokens it manages to disambiguate, like it's shift produce. So that's a very weird.

Derick Rethans 7:17

I'll have a look at what that actually does, because I'm curious now myself. Beyond the parser, I think the biggest and most complicated part of this is implementing the variance rules for these intersection types. Can you give a short summary of what a variance rules are, and potentially how you've actually implemented them?

George Peter Banyard 7:38

Since PHP seven point four, return types and up covariant, and parameter types are contravariant. Covariant means you can like restrict, we can be more specific. And contravariance means you can be broader or like more generic. Union types already gives some interesting covariance implications. Usually, you would think, well, a union is always broader than a single type, you say: Oh, I want either a traversable or accountable, it seems that you're expanding the type sphere. However, a single type can have as a subtype, a union type. For example, you say,:Oh, my base type is a Class A, and I have two child classes, which are B and C. I can type covariantly that I want either B or C, because B or C is more specific than just A. That's what union types over there allows you to do. And the way how it's implemented. And how to check for that is you traverse the list of child types, and check that the child type is an instance of at least one of the parents types. An intersection by virtue of you adding constraints on the type itself will always be more specific than just a single type. If you say: Oh, I want a class A, then more specifically, so I want something of class A and I want it to be countable. So you're already restrict this, which gives some very interesting implications, meaning that a child type can have more types attached to itself than a parent type. That's mostly due how PHP implements its type system, to make the distinctions, basically, I've added the flag, which is either this is a union, meaning that you need to check it is part of one, or it's an intersection. The thing with intersection types is that you need to reverse the order in how you check the types. So you basically need to check that the parent is at least an instance of one of the child types, but not that none of the child types is a super type of the parent type. Let's say you have class C, which extends Class B and Class B extends Class A. If I say let's say my base type is B to any function, and I give something which is a intersection T, any interface, this would not be a valid subtyping relation to underneath B. Because if you looked it was a Venn diagram in some sense, you've got A which is this massive sphere, you've got B which is inside it, and C which is inside it. A intersection something intersects the whole of A with something else, which might also intersect with B in a subset, but it is wider than just B, which means like the whole variance is very complicated in how you check it because you can't really reuse the same loop.

Derick Rethans 10:13

I can't imagine how much more complicated this gets when you have both intersection and union types in the same return type or parameter argument type.

George Peter Banyard 10:22

One of the primary reasons why it's currently not in the RFC, because it is already mind boggling. And although I think it shouldn't be that hard to like, add support for it down the line, because I've already split it mostly up so it should be easy to check: Oh, is this an intersection? Is this a union? And then you need to branch.

Derick Rethans 10:42

Luckily because standard types aren't included here, you also don't really have to think about coercive mode and strict mode for these types. Because that's simply not a thing.

George Peter Banyard 10:50

That's very convenient.

Derick Rethans 10:52

Is the future scope to this RFC?

George Peter Banyard 10:54

The obvious future scope is what I call composite types, is you have unions and intersections available in the same type. The main issue is mostly variance, because it's already complicated, adding more scope to it, it's going to make the variance go even harder. I think with most programming languages, the variance code is always complicated to read. While I was researching some of it, I managed to hit a couple of failures, which where with I think was Julia and the research paper I was it was just like focusing on a specific subset. And like, basically proving that it is correct. It's not a very big field. Professors at Imperial, which I've talked to, have been kind of helpful with giving some pointers. They mostly work with basically proper languages or compiled languages, which have this whole other set of implications. Apparently, they have like a bunch of issues about how you normalize the types like in an economical form, to make it easier to check. Which is probably one of the problems that will need to be addressed, when you get like such a intersection and union type. First, you normalize it to some canonical form, and then you work with it. But then the second issue is like how do you want the composite types to actually be? Is it oh, you have got parentheses when you want to mix and match? Or can you use like union precedence? I've heard both opinions. Basically, some people are very dead against using Union as a precedent.

Derick Rethans 12:14

My question is going to be, is this actually something people would use a lot?

George Peter Banyard 12:21

I don't think it would be used a ton. The moment you want to use it, it is very useful. One example is with the PSRs, the HTTP interfaces. Or if you want the link interface. Combining these multiple things gets it convenient. One of the reasons why I personally wanted as well, it's for streams. So currently, streams don't have any interface, don't have any classes. PHP basically internally checks when you call like certain string methods. For example, if you try to seek and you provide a user stream, it basically checks if you implement a seek method, which should be an interface. But you can't currently do that. Ideally, you would want to stream maybe like a base class, instead of having like a seekable stream, and rewindabe stream, or things like that. You basically just have interfaces. And then like if somebody wants a specific type of stream, just like a stream, which is seekable, which is rewindable. And other things. We already have that in SPL because there's an iterator. And we have a seekable iterator interface, which basically just ask: Oh, this is there's a seek method. I think it depends how you program. So if you separate the many things into interfaces, then you'll probably use intersections types a lot. If you use a maybe a more traditional PHP code base, which uses union types a lot. Union types are like going to be easier. And you want to reduce that.

Derick Rethans 13:32

Would you think that lots of people already use union types because it's pretty new as well. Isn't it?

George Peter Banyard 13:38

Union types are being implemented in various different libraries. PSRs are updating the interfaces to use union types. One use case, I also have a special method, which was taken the date, it takes a union of like a DateTime interface, a string or an integer. Although intersections types are really new, you hear people when union types were being introduced, you heard people saying, I would promote bad cleaning habits, you shouldn't have one specific type. And if you're using a union, you have a design issue. And I had many people complaining to me why and intersection types of see? Why they haven't intersection types being introduced first, because intersection types are more useful. But then you see other people telling us like, I don't see the point in intersection types. Why would you use an intersection type, just use your concrete class, because that's what you're going to type anyway.

Derick Rethans 14:21

I can give you a reason why union types have implemented first, over intersection types, I think, which is that it's easier to implement.

George Peter Banyard 14:28

It's easier to implement. And it's more useful for PHP as a whole, because PHP functions accepts a union or return a union. Functions return false for error states instead of null. It makes sense why union types were introduced first, because they are mostly more useful within the scope of what PHP does.

Derick Rethans 14:46

Do you think you have anything else to add about intersection types? At the moment, it's already up for voting, when is that supposed to end?

George Peter Banyard 14:54

So the vote is meant to end on the 17th of June.

Derick Rethans 14:57

At the moment I see there's 15 votes for and two against so it's looking good. What's been your most pushback on this? If there was any at all?

George Peter Banyard 15:05

Mostly: I don't see the point in it. However, I do think proper reasons why you don't want it, compared to like some other features where it's more like have thoughts on what you think design wise. But it is undeniable that you you add complexity to the variance. And to the variance check. It is already kind of complicated. I have like a hard time reading it initially. There's the whole parser hackery thing, which is kind of not great. It's probably just because we use like a restricted parser because it's faster and more efficient.

Derick Rethans 15:36

I think I spoke with Nikita about parsers some time ago and what the difference between them were. If I remember which episode it was all the to the show notes.

George Peter Banyard 15:44

And I think the last reason against it is that it only accepts pure intersections. You could argue that, well, if you're adding intersections, you should add the whole feature set. It might impact the implementation of type aliases, because if you type alias T to be a union of A and B, and then you use type T in an intersection, you basically get a mixture of unions and intersections, that you need to be able to work with. The crux of this whole feature is the variance implementation. And being able to rationalize the variance implementation and been to extend it, I think it's the hardest bit.

Derick Rethans 16:18

I guess the next thing still missing would be type aliases, right? Like names for types, which you can't define just yet, which I think you also mentioned in the RFC is future scope.

George Peter Banyard 16:29


Derick Rethans 16:30

Thank you, George, for taking the time today to talk to me about pure intersection types.

George Peter Banyard 16:36

Thanks for having me on the show.

Derick Rethans 16:41

Thank you for listening to this installment of PHP internals news, the podcast dedicated to demystifying the development of the PHP language. I maintain a Patreon account for supporters of this podcast as well as the Xdebug debugging tool. You can sign up for Patreon at If you have comments or suggestions, feel free to email them to Thank you for listening and I'll see you next time.

PHP Internals News: Episode 87: Deprecating Ticks

PHP Internals News: Episode 87: Deprecating Ticks

In this episode of "PHP Internals News" I chat with Nikita Popov (Twitter, GitHub, Website) about the "Deprecating Ticks" RFC.

The RSS feed for this podcast is, you can download this episode's MP3 file, and it's available on Spotify and iTunes. There is a dedicated website:


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 87. Today I'm talking with Nikita Popov about a much smaller RFC this time: Deprecating Ticks. Nikita, would you please introduce yourself.

Nikita Popov 0:34

Hi Derick, I'm Nikita, and I'm working on PHP core development on behalf of JetBrains.

Derick Rethans 0:40

Let's jump straight into what this RFC is about, and that's the word ticks. What are ticks?

Nikita Popov 0:46

Ticks are a declare directive,. You write declare ticks equals one at the top of your file, and then PHP we'll call a tick function after every statement execution. Or if you write ticks equals two, then as we'll call it the function after every two statement executions.

Derick Rethans 1:05

Do you have to specify which function that calls?

Nikita Popov 1:08

Of course, so there is also a register tick function and unregister tick function and that's how you specify the function that should be called rather the functions.

Derick Rethans 1:17

How does this work, historically, because the RFC talks about the change being made in PHP seven?

Nikita Popov 1:22

Technically ticks work by introducing an opcode after every statement that calls the tick function depending on current count. The difference that was introduced in PHP seven is to what the tick declaration applies. The way PHP language semantics are supposed to work, is that declare directives are always local. The same way that strict types, only applies to a single file, ticks should also only apply to a single file. Prior to PHP seven, it didn't work out way. So if you had declare ticks, somewhere in your file, it would just enable ticks from that point forward. If you included the different file or even if the autoloader was triggered and included a different file that one would also make use of ticks. That was fixed in PHP seven, so now it is actually file local, but that also means that the ticks functionality at that point behaviour became, like, not very useful. Because usually if you want to use tics you actually want them to apply it to your whole codebase. There are ways around that. I'm afraid to say that people have approached me after this RFC and told me that they actually do that. The way around that is to register a stream wrapper. It's possible in PHP to unregister the file stream wrapper and register your own one, and then it's possible to intercept all the file includes and rewrite the file contents to include the declare ticks at the top of the file. I do use that general mechanism for real things in other places, but apparently people actually use that to like instrument, a whole application with ticks, and essentially restore the behaviour we had in PHP 5.

Derick Rethans 3:03

What was the intended use case for ticks to begin with?

Nikita Popov 3:07

Well I'm not sure what was the intended use case, but at least it was the main use case, and that's signal handling. In the PCNTL extension allows you to register a signal handler, and when the signal arrives, we can't just directly call that signal handler, because signals are only allowed to call functions without that our async signal safe. Which excludes things like memory allocation, and a lot of other things that PHP uses. What we do instead is we only set the flag that okay signal has arrived and then we have to actually run the signal handler at some later point in time. In PHP five, that worked using ticks. You declare ticks, and the PCNTL extension registered the tick handler, and then after this flag was set, it would execute your callback on the next tick. In PHP seven, an attentive mechanism was introduced, that is based on virtual machine interrupts. Those were originally introduced for time-out handling, because there we have a similar problem, that when timeout arrives, we might be in some kind of inconsistent state, like the middle of the allocator right now, and if we just bail out at that point, we are likely to see crashes down the road. So that was a significant problem in PHP five. PHP seven changed that. We now set an interrupt flag on timeout, and then the virtual machine checks this flag at certain points. The interrupt flag is not checked after every instruction, but only, like, just often enough to make sure that it's checked, at some point. So that you can't like go in an infinite loop, that ends up never checking. These points are basically function calls, and jumps that go higher up in the function, PCNTL signals can now use the same mechanism. If you call PCNTL async signals true, then those will also set the interrupt flag, and execute the signal handler on the next opportunity. The next time the interrupt flag is checked. The nice thing about that is that it's essentially free. I mean we already, we already have to do these checks for the interrupt like anyway, adding the handling for PCNTL signals doesn't add any cost on top. Unlike ticks, which have to be like executed on every instruction or at least regularly, and that does add significant cost.

Derick Rethans 5:28

Execution time itself because it's an opcode that needs to be executed.

Nikita Popov 5:32


Derick Rethans 5:33

So what are you proposing to do but the ticks in PHP eight one then?

Nikita Popov 5:36

I want to deprecate that. So both the declared directive itself, and the register tick function, unregister tick function.

Derick Rethans 5:44

How could users emulate the same behaviour as ticks allows them to do so now?

Nikita Popov 5:49

That's a good question. As I mentioned, if the use case is, use case of ticks was signal handling, then by using async symbols. If it was something else, then you have a problem. My assumption when writing this RFC was basically that signal handling was really the main remaining use case of ticks, because other use cases require this kind of you know stream wrapper instrumentation, and I didn't expect that people will be crazy enough to use something like that in production.

Derick Rethans 6:21

Hopefully they catch these rewritten files?

Nikita Popov 6:23

Probably yeah. I think it's possible to make this integrate with opcache. If you use it for other purposes, then, I don't think there is a really good replacement. So I think what they use it for is some kind of well instrumentation, so profiling, memory profiling, for example, and the alternative there of course is to use a tool that is appropriate for that job, for example, Xdebug contains a profiler, but of course it is not a production profiler, but I think there are also production profilers.

Derick Rethans 6:54

As far as I know all the production or APM solutions. They do this on their own without having to use sticks. They don't need any user land modifications.

Nikita Popov 7:03

Yeah, definitely. All the APM solutions support this, they use internal handlers.

Derick Rethans 7:08

Because it's actually removing functionalities that some people use, what's the reaction been to removing this functionality?

Nikita Popov 7:14

Well on the mailing list at least positive, but as I mentioned at least some people have like pointed out on the pull request that they are using the functionality.

Derick Rethans 7:23

Enough in such a way to sway for not deprecating them? What is the benefits of getting rid of ticks, if you don't use them?

Nikita Popov 7:31

That's, I think the thing, that there is not really a big benefit to getting rid of them. Like they don't add a lot of technical complexity to the engine. They're pretty simple in that sense. I haven't seen those responses. I'm kind of rolling a bit unsure if we should really remove them, because you could argue that well they don't really hurt anyone. I do have to say that I think all the things that people use sticks for, all the cases I have heard about, and all of those cases ticks are not the right way to solve the problem. They are not the right way to solve the signal handler problem, they are not the right way to solve the profiling problem. And the other one I heard is also they're not the right way to solve the heartbeat problem, to make sure a service stays connected. While people do use them I think they use them for questionable purposes.

Derick Rethans 8:24

Developers, if they're using something to rewrite the PHP file to introduce ticks, they can also technically rewrite a file to introduce calls to their own functions, after every statement.

Nikita Popov 8:34

Yes, I actually have a very nice PHP fuzzing project that rewrites PHP files to introduce instrumentation functions at certain points. That needs a lot more control than ticks, because it's interested in branching statements in particular. That is definitely also possible, but it's kind of even more crazy than just adding ticks. If you're doing it like this, I think, if we want to keep ticks, then we should change ticks from a declare directive to a ini_set, because this kind of rewriting of files to introduce takes that's like not a great solution. On the other hand, that does mean that if you are, I don't know a library, implementing some code and expecting that, you know, it just runs normally, then someone can with by enabling an ini setting will suddenly run code in the middle of your library file that's like essentially any point. So enabling ticks us a major behaviour change, that's something we really don't like to have in ini settings which is I guess also, why does it declare in the first place, because that limits the scope. And you have to go out of your way if you want to not limit it using this rewriting hack. So I'm not really sure ultimately what to do here.

Derick Rethans 9:44

Are you thinking of bringing this up for vote before PHP eight dot one's feature freeze?

Nikita Popov 9:49

If I decide to go for it, then definitely before. I'm just not completely sure on this topic yet.

Derick Rethans 9:55

it'd be interesting to, to hear what other people think about removing this. I have no opinion about this. Other features I do but in this case, I'm happy with them being there, I'm happy with them not being there, because it's something I'm using myself. In any case, thank you for going through this RFC with me today, and we'll see what happens.

Nikita Popov 10:14

Thanks for having me, Derick.

Derick Rethans 10:18

Thank you for listening to this installment of PHP internals news, a podcast dedicated to demystifying the development of the PHP language. I maintain a Patreon account for supporters of this podcast, as well as the Xdebug and debugging tool. You can sign up for Patreon at If you have comments or suggestions, feel free to email them to Thank you for listening and I'll see you next time.