Scroll down to the script below, click on any sentence (including terminal blocks!) to jump to that spot in the video!
SymfonyCon 2018 Presentation by Pedro Magalhães
PHP releases a new minor version every year. Major versions happen when there are enough changes that justify to do so. Who is making those changes and how does that process work? What is the process to get an RFC to vote and the subsequent merge of the code? How can I make my first contribution? Is there anything I can do even if I don't know C? If you are intrigued by PHP internals, this talk is for you.
Okay. So, Hi everyone. Thanks for joining this talk about PHP.
First, a little bit about myself. I'm Portuguese, I'm Pedro Magalhães, by the way. I'm Portuguese, I'm 33 years old. Uh, I'm a PHP developer at the Dutch company called Emesa during the day and I'm a PHP contributor by night. Some nights, at least.
Um, I authored one rejected RFC, which was to get rid of that small "b", uh, before the strings because in PHP you can always put a "b" before any string. Uh, and this is a relic from the past when this was introduced for forward compatibility with PHP 6 and the binary strings and the unicode strings. And that never happened as we all know, there is no PHP 6, but the "b" is still there 10 years now and it still doesn't do absolutely anything. But I tried to get rid of it, and people didn't want to get rid of it. So, it stays. Uh, and well, also one approved RFC - it's the first one that was approved for 8.0.
And uh, yeah, I also tried to contribute a little bit to other open source projects, like I have my very own badge of Symfony code contributor. And, yeah, I like games, and beer, and technology. Yeah, okay.
So in this talk it's called changing PHP and changing PHP can have two meanings, uh, and I'm going to talk about both of them. So one of them is as an objective, um, changing is something in a state of becoming different. And so we will talk a little bit about how PHP is changing lately. Um, and then as a verb. So I really like this definition to make the form, nature, content, future course, etc of something different from what it is or from what it would be if left alone. So this is the two parts of the talk.
So, let's start with the first one - with changing as an objective and, um, let's start with a reality check. So this is where we stand today. Um, 5.6 has 24 more days of security updates and then 5 is done for, forever. 7.0 is already out of security updates. Even though there will still be one less release, uh, that was not planned, but there will still be 7.0.33. Um, but yeah, that's it for 7... for 7.0. Then 7.1 is already security only since five days ago. So, currently the only actively supported version of PHP is 7.2 - is the only one that is getting actually bug fixes. But, the good news is 7.3.0 is released today.
It's not announced yet, uh, on the website. Um, but, but it will be later today. I promise, by night you can have 7.3 in production during the social event we can do it. Um, and yeah, it will be, it will be nice. But yeah, it's really, really great. It's a great timing that 7.3 is released today. So because of that, let's talk a little bit about PHP... about 7.3.
Um, and let's talk about some of the things that it will bring. So, one of the first that I want to mention is JSON_THROW_ON_ERROR as it kind of advertises, it will be possible to pass chase and throw an error to json_encode() and json_decode(). And it will start throwing exceptions instead of you having to go out and check json_last_error() if something wrong happened. And, well, yeah, sometimes it did, sometimes it didn't. And a lot of the libraries for handling JSON are pretty much just wrapping this up, just checking the json_last_error() and then throwing it for you. But now, PHP can do it by itself.
Then for the OCD ones among us. Now you can use a trailing comments in a function calls. Um, so your, your diffs are going to be much nicer now because you don't have to add a comment to the previous line and all of that. I, yeah. For some people it may be, it may be useful or nicer. Then, array_key_first() and array_key_last() also as the name advertises, it gives you the first and the last keys of an array. Um, and um, yeah, without changing the internal pointer of the array. So, the array stays in the same place where it was.
And this way you can check it. It has, didn't fortunate decision in my opinion, of returning NULL, in case you pass an empty array, I would prefer to throw something, but mostly it's, um, it's, um, it's a good addition.
And also, flexible heredoc. Uh, so for heredocs and nowdocs, um, you can now use this, uh, this syntax where the difference is that the closing marker. So, in this case, the end no longer has to be followed by a new line, so you can actually now close the function call that you are doing a after the closing marker. And the other thing is that the closing marker, I'm basically specifies, um, the indentation of the rest of the text. So, in this case, because END is not really at the, at the, um, at the first column, um, and it is at the fourth column, it means that C will only have one space before it when it's printed.
So, basically, the closing marker now dictates, um, where the, um, where the intention starts for all the text. Then, uh, SameSite cookies - it already existed in Symfony since 3.2. So, this is not news for a lot of you, um, and since 4.2, there was 3.2 is already from two years ago and 4.2, it also added support for that for session and "remember me" cookies. And the SameSite cookies for those of you who don't know, basically it's a way to prevent a cross site request forgery. And, uh, basically the way that it works is that if you put it too strict, uh, the cookie won't be sent if the request is cross site. So, if you are on a different website and you'll make, um, yeah, you click a link to your website, the cookies want to be sent with that request.
Um, but because that will be really unfriendly in some cases, uh, you also have the Lax which allows the cookie to be sent if this is considered a safe request and it's safe request in this case is mostly a GET request, or any "read" request that is a top level navigation so that when you are on Gmail, for instance, and you click on the Twitter link, you stay logged in because, uh, if even if they use the SameSite, if it is Lax - than your cookies will also be sent.
But, in PHP, this is the current signature of the setcookie() function. And so we wanted to add one more - SameSite. Can I join you guys? And the guy said: No! So we had to come up with the, with the alternative syntax for this because this was, this went to vote and it was rejected. And so, the solution that ended up coming forward was this one of setting the cookie and you get the name, the value, and then an array of options. And the array of options is of course, all the other options, so
More on 7.3. In this case, so, we have a
case "baz", and then we have a
continue. Who here thinks that the
continue will continues the
while(). Who thinks it will continue the
Okay. Well, that was not a lot of confusion but in case it is confusing because it can be confusing to some people. Now it will throw also a warning letting you know of that, that the continuous the
switch() is just a break and it asks you if you wanted to, if you meant continue to because then you, as with continue and with break, you can add the number to say the number of loops that you want to break out of or continue from.
Another thing also in reflection, in errors, uh, you would see before you would see integer and boolean. But that's not really what you typed down. So now it's, it matches what you use as well. So, now they are always called int and bool, which is nice.
Then we have a new password hashing algorithm if you build it with the, with the password Argon, with the lib... Argon library. Then you can also use now the PASSWORD_ARGON2ID. Um, in FPM, there are also some changes. The entire logging of FPM was refactored and right now you can set the log limit variable to define what will be the maximum length of a line on the log, um, which prevents some before you could lose some information because of, because of the line that was too big. And right now this solves that problem.
And also it allows you now to not decorate the workers outputs because PHP would also always add some information to the, to the output of the workers and now you can actually specify that you don't want that. Um, besides that, also FPM finally got the
getallheaders() supports, which is basically to give you all the requests heathers and well, it didn't work in FPM.
Then, another interesting one is the
hrtime(), which is the highest resolution monotonic clock and the monotonic clock, basically, it's a clock that's always goes in the same direction. That is a guarantee so that it's not affected by a daylight saving and stuff like that. It will always, always, always goes forwards. Um, and it starts at the unspecified point in time, in the past. It's not important, it's just that you can measure it twice and we will get the most precise difference between those two times possible. So this is really useful also to measure stuff, of course.
And of course there are a lot of bug fixes and performance improvements and stuff like that. But who cares? I'm, this is the truth. And sometimes it's also the truth for contributors, like, ah, yeah, I want to do something new and shiny, but yeah.
So, but this is mostly what I have to say about a 7.3. About 7.4 which is coming one year from now. So holds, you have to sit down a little bit to wait for that. But, uh, there it is already showing a lot of very interesting stuff. The first one that I would like to talk to talk about is about the preload, which is already accepted and it's already merged in the master. Um, and it was developed by Dmitry Stogov and to quote what he said because I think it's a good definition.
So on service startup, before any application code is run, we may load the certain set of PHP files into memory and make their contents permanently available to all subsequent requests that will be served by that server or the functions in classes defined on these files will be available to requests out of the box exactly like internal entities. By example, the string length or, or the exception class.
And what does this mean exactly? So, currently, on 7.3, um, so PHP starts and the first request comes. And I imagine that on your index.php where your, your requests are arriving, you do a new A. These will trigger the autoloader because it doesn't know about it yet. So it will trigger autoloader: Hey, what, what is A? And your autoloader imagined that it is composer's autoloader a will then find that file and include it. The file is compiled into OP codes and those OP codes are stored in the OPcache. And after that it will run the constructor of A on the subsequent requests you do new A again. It will again triggered the autoloader, it will, um, the autoloader will include A.php but this time we don't need to compile it anymore because it's already on OPcache.
So now we load the OP codes from OPcache and we run the constructor. On 7.4 with preloads. So we, when we start PHP, we can pass these, uh, these setting, or we can have this ini setting saying
opcache.preload and the file in this case, so, for this example, I'm just preloading A.php directly. Typically, it would preload the file, that would include all the other files that you want to preload.
You can only specify one file and then that file is responsible for preloading the rest of what you want. But in this case, for the sake of simplicity, let's say that we just preload A.php, and at this point A.php is compiled and it's in memory. So for the first request, when you do new A, it runs A, um, A constructor because it already knows what it is. It already, it's already loaded in memory. It doesn't need to hit OPcache at all.
So, yeah, all the requests are now immediately... you don't have these intermediate steps. This, of course, there're the downsides, uh, all the files that are preloaded, if you want to change something in them, uh, you need to restart PHP. So, there is no way to invalidate these files. The only way to change them is by restarting PHP.
Then, typed properties, finally. Um, so yeah, it, it's happening. I'm sorry if you're having a recruiter speak flashbacks. Um, but yeah. So, it's happening. It's going to be on 7.4. Um, and some, some interesting things, so, object types, uh, cannot have a default value unless they are nullable and indicates that they are nullable then you can use null as a default, but you cannot set a default values for, for thing, for properties that are of a specific class and they will not be null. They will be undefined if you don't define them in your constructor. So we have a new states for, for properties which will be undefined, which you shouldn't really encounter if, if you ever encountered undefined is because someone forgot to set the property on your constructor. It shouldn't happen, but, it, uh, it is there.
And, currently this is still being discussed. Uh, but this one was already voted, was already accepted. It's not merged yet because they are still struggling with some implementation details. And the other one, it's currently on discussion. So this one I cannot tell you that it will happen for sure, but I'm pretty confident it will, which is about variance. So in this case, so we imagined that we have a class A, grandparents and then we have a class B which is the parents. And then we have a class C which is the child. And then we define an interface and that interface receives a B, and it returns a B.
We can now, well with variance, we can extend these interface and we can use a contravariance on the, on the, on the, on the parameter types, which means to go up one or multiple levels, so... and, uh, for the return type you can go down. So that's covariance. And, basically, these does not break the, it does not violate the Liskov substitution principle because currently you could already, even just when you have the type hint and on the child, you could just remove the type hint. But this is now more complete, so it will actually, like if your parents can deal with the B, then your child can deal with something that is greater than a B and then can deal with it as the same way that if your parents returns a B, it's safe for the child to return to the C because the C is also a B. So it never breaks, it never violates the Liskov substitution principle.
Okay. So for 8.0, um, there is already a change. So I said that, uh, the RFC, um, I have an RFC that was accepted for 8.0 and it's very simple, but it's basically about this. So if we have an array with the, with this initial structure, and then we add one more element to this array, uh, what will be the index of these new element.
You wish, but it's not. Currently, it's not. So, as the documentation says: If no key is specified the maximum of the existing integer indexes, uh, is taken and the new key will be that maximum value plus one, but at least zero. And I really, really, really disliked this and so I got rid of that part. So, on 8.0, uh, it will be minus 41 while currently it's still zero. And because this may bring some, uh, yeah, BC breaks, um, this was, uh, this was a slate that for, for a new major version.
But for the rest of 8.0, um... OK, I have to switch modes. Well, I want you to contribute to, to PHP and to help shape up 8.0. So now it's really the right time to do that. 7.4 still being developed so you can still introduce new deprecations for 7.4 for things that you really wanted to change on 8.0. So now it's a good time to do that.
And so, let's talk a little bit about how you change PHP, about the verb part of it. Um, but first I'm going to tell you a little bit my first, about my first contribution and tell you a little bit the story about my first contribution. Um, and it was like these. So, one afternoon, in an office somewhere, uh, we were having a security scan, um, and certain requests were hanging on certain conditions. Um, and basically it boiled down to the fact that the text protocol of Memcached cannot really handle new lines in the keys properly.
Um, and yeah, because it would hang or it would even allow injection. So this is actually, this was an, uh, an injection for Memcached was possible to, to abuse these, to get that. And this was easy to fix on our side, right? So just, okay, don't accept new lines in the keys. Uh, but yeah, I started wondering like why, why does PHP let these through? What, what, yeah, why would they do that?
Um, and so, luckily in the PHP world, we are in the world of free software and two of these freedoms specify that you are free to study how the program works and change it so it does your computing as you wish. And I wish Memcached to not be vulnerable to injection and then the freedom to distribute copies of your modified versions to others. By doing this you can give the whole community a chance to benefit from your changes.
Well, so let's, let's try to get that for everyone. And so I started looking at the code of the PHP Memcached extension and I was positively surprised to figure out that, well it's not rocket surgery. It's not that hard. And the change in the end, in terms of code, ended up being just these and maybe it's not very well visible, but I think that this is reasonable for anyone. This is not very complicated. It's C, it's a little bit, syntax is a little bit different from a PHP, but in this case, not even that much and this was all that was needed to, to cover that, to cover that hole.
So then, um, well now what do I do with this? Uh, I emailed the [email protected] because this is a security concern and they told me, yeah, just open a pull request. Don't mention that it is about security, just open a pull request and uh, and we will merge it eventually our. Okay, sounds good. So I did, I opened the pull request, and it got merged, and since PHP Memcached 3.0.0, that's the problem no longer occurs. So, that's nice. So it's no longer vulnerable to at least this type of injection.
Then about your first contribution. First, I wanted to know, did anyone of you already contribute to PHP anything in any way?
Maybe you did, but. Okay. So, let's, uh, first things first. Subscribing to the mailing list is pretty much essential. So this is where all the discussions about PHP happen. Uh, this is where the RFCs are presented. This is, uh, yeah, where a lot of these discussions happen. This is also a nice place to get help if you're trying to fix a bug and you cannot figure a step out, you can always mail internals and there is these very nice website, externals.io that basically just exposes the internals to the outside. Is a nice way to read the mailing lists. It does the whole treating thing and yeah, it's, it's very nice to, to read there. But you can always subscribe to the mailing list on php.net/mailing-lists.php. So this is, for me, it's a good first step.
Then, helping with the documentation. Sometimes there is ambiguous language that should be fixed or updated documentation or you can help with translations to your own language. If it's not english and it's missing something and you can visit edit.php.net, and it gives you this fancy interface with the facebook like button where you can actually edit the documentation there and you can submit your patch to review and then anyone with a php.net account, if they find that this makes sense, they can merge it. And then the website is a rebuilt every 24 hours. So it might take a while, but eventually your change will be there.
There is talks of possibly migrating these to GitHub, which would make sense and make things a little bit easier. But on the other hand here you can also do it anonymously. So we'll see. We'll see what will happen with that.
Then - reporting bugs. So, we do have bugs.php.net. Um, and yeah, as with any bug reporting, you should try to always provide the minimum reproducing code. So, okay, this is all you have to do to have this problem. Any other useful information and of course what you expect versus what you actually got. And in this case it's a, it was a bug reported by Nicolas, caused by me and fixed by me as well. So that was nice.
Then, um, you can also answer bug reports provided that you have an email address or not. Maybe you don't really that and that you have some basic mathematic skills - you can solve that problem you can contribute. Um, and yeah, I mean sometimes you just know how to reproduce it with less codes.
Sometimes you just, you see that the person that posted the question is not giving you enough information, is not giving anyone enough information so you can tell: Hey, can you, please, be more specific or anyone can, can help with this. So replying to bugs is all, so really, really nice. Um, and then who's doing it here, also doing it on StackOverflow, answering people or on a, on the chat of StackOverflow. Um, those are good. Those are good places to help out as well.
And then about fixing the bugs themselves. So, PHP is on GitHub. It's a, it, it's true. it's a mirror of git.php.net. But it has pull requests, you can just do your change, open a pull request, like with any other project and the change can be merged. So, yeah, it's pretty simple to start off.
Um, to help you with the, with your diving into the PHP source. Um, there are some helpful tools. One of them is LXR, which is a way to look at the code a little bit like an advanced idea because you can click on anything and it will take you to that function. And yeah, you can see that one on the room11.org, lxr.room11.org. Um, Room11 is the PHP room on the StackOverflow chat and a lot of PHP maintainers or contributors are also their daily. Uh, so it's also a nice place if you're trying to dig in and you don't, um, yeah, you need some help. It's a good place also to find that help.
Then, you also have 3v4l.org, uh, which is really, really, really great. It's a, yeah, it's a website where you can run a snippet of code and it will run it on 200 plus PHP and HHVM versions so you can see exactly when a behavior was changed in which versions did that happen. Um, and the shout out to Sjon for, for making this and making it available.
Then, you also have wiki.php.net, which has a lot of useful resources if you want to dig in. Uh, so yeah, so is a recommended reading. It has some things that are a little bit outdated, but yeah, you will always be able to find the interesting documentation there. And also on the, on the, on the repository of PHP source itself. It also has the contributing file and it also has some nice information and from these I would say to use your favorite editor to avoid any religious wars. I did this. Um, so yeah, just use your favorite editor as long, in my opinion, as long as you can debug properly with it, uh, feel free to choose whatever you want to use.
Then, writing tests. So, PHP tests look like this. This is a
.phpt file. Um, they have a pretty simple structure. So, basically, here we see it is, it has the test name, the script itself, what it will do and what it expects. It's all good. It will pass if the, what expects is not there, uh, it will fail. You have another, you have a bunch of other sections. You can have EXPECTF instead of expects to define it in a format. You can have INI to change some INI setting specific for this test. Uh, you can have a SKIPIF so to, to run a little snippet. And if, uh, if the output is "skip" - then the test is skipped because the extension is not there or for whatever reason.
And then you can run the entire test suits if you run... make tests on the repository or if you want to only test a specific test or a bunch of tests that are related to each other. You can also use the run tests, um, well
make test uses run tests, but then you can use run tests directly to specify that you will only want to run a certain type of or certain tests.
Then, proposing changes. So this is the RFC process. RFC stands for request for comments. Um, and yeah, your approach should go a little bit like this. So the first thing to do is to check what happened the last time that this was discussed. PHP is a pretty large project. It has been around for a lot of years and most likely what you are thinking about, somebody's already thought about and maybe it was already discussed. So that's always a good starting point is to figure out: Okay, what happened the last time that someone brought this up?
Then, if you're convinced, okay, nobody really, nobody talked about these are the reasons why they rejected it before are not valid anymore. Uh, you should email internals and introducing your proposal and listening to feedback a little bit like this first phase of, uh, of feedback and actively listen to the feedback and try to adjust your proposal to that.
Then, uh, you can write the RFC itself on the wiki and it will produce a fancy document looking like this. Um, and yeah, and here you should really detail everything that has to be said about the proposal, like a feature, uh, what, what if it has any, uh, BC breaks. If it, if it has any future scope, like in the future we could also do this, this and that, the proposal itself. So, yeah. Um, and then, uh, preferably, this is not mandatory, but preferably you would also implement our proposal.
Uh, but this is, as I said, it's not mandatory. It happens sometimes people propose things even if they cannot implement it. Um, and then someone else will pick up the implementation, most of the times. Um, versus the SameSite cookie - a thing that I talked before, it was proposed by someone else, but then they couldn't come up with a, with a implementation and I ended up, I ended up implementing that myself. That with the options, a rare.
Um, then, you should start your discussion period, which should be at least for two weeks. And, uh, well listen again to feedback, change things that have to be changed or fight people until they said that, they say that it's not needed to change anymore.
Um, and then you can start the voting period. The voting period should also last for at least two weeks. Um, people go to the wiki and they can vote on it. Uh, you can, uh, you don't need to have a php.net account to vote. You just need to have a wiki account to vote and some people that are not directly contributors to PHP versus Fabien has, has the, um, the possibility to vote on the, on the RFCs. So, if it is accepted, the PR gets merged into, into PHP.
And that's pretty much it. Um, I would appreciate any feedback that you have about the talk in that JoindIn a link and a, yup, I would take questions. Does anyone have any questions?
Uh, they will start at 0 if you don't say anything. They will start at 0 anyway. Uh, this will only change if for the first keys that you use are negative, right?
Okay. Yeah, that's why we'll wait for a major version. Right? So, instead of like I propose these initially for 7.2 I think yet, and they're like: Oh, of course not. It's not an a minor. Are you crazy? Oh, hey, sorry. Uh, then, well, then I do it in a major. Is that fine? Yeah, on a major is fine, okay. Then I'll do it in a major. So, yeah, so that's basically the reason why we will wait for 8.0 to do this change. Because, yeah, it might break some things, or some people if they write code, like if you write code, if you're using implicit keys to add things to the array and then you use an explicit key to access it - you're probably doing something wrong. Right? I mean if you don't care about the keys, you don't care about the keys. Or either you care or you don't. Make a choice. But, yeah. Anything else? Yup.
I shouldn't say. Come on, I did all the effort to not say, but okay, I use the VS code. Uh, and it works really well for me. I use it on Linux and it integrates really well with the GDB for debugging, so I can just add breakpoints and it breaks and I can see the state of the entire application. For me, it works really, really well. But yeah, I won't say more than that. Anyone else? Yup.
No, not strong. Uh, actually like probably if I will have to write C outside of the PHP context, I would be a little bit lost to be honest. Um, but yeah, it's uh, I don't have a strong background in C. I learned a little bit about it in school, but that was it. And by day I work with PHP, so that's what I'm used to. But uh, yeah, as I said before, it's not rocket surgery. It's not that complicated. It's, yeah, you'll get used to it quickly I would say. Anyone else? Okay. Well, then that's it. Thank you.