PHP Performance Demystified: Internals, Profiling, and the Road to PHP 9 with Volker Dusch

[00:00] Vular, welcome. You're going to be

[00:02] talking about performance in PHP. What

[00:06] exactly will you be telling us all about

[00:08] at PHP UK conference?

[00:11] >> Hey, thank you. Nice to to be here. Um,

[00:14] so the main thing I want to I want to

[00:16] talk about to people is how does the PHP

[00:19] engine actually work and how do you make

[00:21] decisions based on that. So not just uh

[00:24] and we'll also be talking about that not

[00:26] but not how just how to measure the

[00:28] program externally but also to develop a

[00:30] better intuition and understanding of

[00:32] how these things work so that when you

[00:33] look at a piece of code you can also

[00:35] already make some guesses about the PHP

[00:37] side of the performance and then we're

[00:39] also going to talk about the usual

[00:41] performance aspects like how do I

[00:42] measure how long the database is taking

[00:44] and we'll we'll see some other

[00:46] approaches like most people probably

[00:48] have seen how to instrument this in in

[00:50] userland like you make a wrapper around

[00:52] your database calls and you and and and

[00:54] you do that. But I want to show some

[00:56] more um seale approaches, some more PHP

[00:59] extensions you can use in some more

[01:01] low-level tooling to really figure out

[01:02] what's going on without having to

[01:04] necessarily change the code. For

[01:05] example, if you have a problem that only

[01:07] happens in production,

[01:09] >> I see. So if people come along and if

[01:11] they have uh slow running code or code

[01:14] that you know doesn't seem to be

[01:16] performing as well as they would expect

[01:19] then the takeaway from your talk will be

[01:21] one of the takeaways from your talk will

[01:23] be here's how we can measure it and

[01:25] here's will will actually show how to

[01:27] make changes um based on that

[01:29] information.

[01:30] >> Um so what I will be showing is a bit

[01:32] how to work with a profiler. There's a

[01:34] couple great open source tools out there

[01:36] where you can like run a request or run

[01:39] a process and really see every single

[01:41] function call that happens or see like a

[01:43] couple thousand times a second what's

[01:44] what's going on in in in PHP and then

[01:47] how this looks after you make a change

[01:50] but also just how to like reason about

[01:52] performance and how PHP actually

[01:54] executes your code internally so you get

[01:55] a better understanding for what's going

[01:57] on there as well.

[01:58] >> Amazing. And in your work, have you

[02:01] applied these principles on projects

[02:03] that you've been working on? And what's

[02:05] been your biggest performance increase

[02:07] based on this the kind of work the kind

[02:10] of analysis you do?

[02:12] >> So like there's always a couple fun

[02:14] stories to tell, but so um I work for

[02:16] for a company um we'll maybe talk about

[02:19] that a bit later. I'm not going to go

[02:20] into much detail where we help other

[02:22] other people with performance problems

[02:24] both in a consulting fashion but mostly

[02:26] in providing a a software as a service

[02:29] platform to them where they can improve

[02:31] their code and like there is there is

[02:34] like the the hard winds of people like

[02:37] hey I have this service it has like a 7

[02:39] millisecond response time we want to get

[02:40] it down to five like what what can we do

[02:43] what can we measure and those are like

[02:44] very intricate and very specific but

[02:46] then there is also like here's an

[02:48] e-commerce shop and they had installed a

[02:50] plugin that uh every for every single

[02:52] page uh page load that they make like

[02:54] wants to get like a a token for like

[02:56] some tracking service and that just

[02:58] takes one and a half seconds. Uh and

[03:00] it's just like every single request and

[03:02] that token never changes and they could

[03:04] just have it in there but that plug-in

[03:06] just like fetches it for every single

[03:08] page request. And um because it's like a

[03:11] big e-commerce shop with a bunch of

[03:13] plugins that you haven't written,

[03:14] debugging that can actually be quite

[03:16] annoying. So like people can use our

[03:18] tool or other tools to like get that

[03:20] information very quickly and so that was

[03:22] a great great experience with the

[03:24] customer of like their onboarding to the

[03:26] product and the first thing was like oh

[03:28] that's the problem and within minutes

[03:29] you have a solution and it's it's really

[03:32] fun to see like

[03:33] >> how how fast you can make PHP and what

[03:36] you can get out of the language while

[03:37] still having like a nice highle language

[03:39] that allows you to work very quickly uh

[03:42] but then still get really good

[03:44] performance out of it as well.

[03:46] So the PHP haters who say it's too slow,

[03:52] that's not really true, is it? I mean,

[03:54] it I it always depends on like for what

[03:56] use case and what you want to do with

[03:58] it. um like uh if if if you need a

[04:02] real-time application or like if you

[04:04] want to do high frequency trading

[04:06] wherever you would want to do that like

[04:08] I probably wouldn't do that in PHP and

[04:10] um but uh for a website and like if we

[04:13] if we compare PHP to other web languages

[04:17] like JavaScript, Python, Ruby um I'll go

[04:22] a bit in the talk about like some

[04:24] comparisons there as well but I think

[04:25] PHP holds up really really well And um

[04:29] given the given how core works and the

[04:32] and the improvements there and how the

[04:34] engine works, there are some things that

[04:36] PHP is really good at. Like it also

[04:38] checks your types at runtime. So you

[04:40] don't like if you get a JSON response

[04:42] from an endpoint that you didn't expect,

[04:43] it won't magically do the wrong thing

[04:45] because the types didn't match because

[04:47] it actually checks the types. And that's

[04:48] cool. And it still manages to outperform

[04:51] some of the languages that don't do

[04:52] that. So I think there is there's a lot

[04:54] to be said for running PHP.

[04:57] I see. So if people are watching this

[04:59] and maybe they've got a slow web

[05:01] application or e-commerce thing and

[05:03] they're thinking, do we need to change

[05:05] from PHP to another language? The answer

[05:08] would probably be well actually let's

[05:10] look at the performance of our PHP

[05:11] application first. And

[05:13] >> yes,

[05:14] >> I I I think it's not like usually like

[05:17] uh if you rewrite something in a new

[05:19] language, it's not faster because you've

[05:21] used another language, but it's faster

[05:22] because you rewritten the whole thing.

[05:24] Um, and the the process of like

[05:27] re-evaluating all your decisions and

[05:28] reing your approaches and like redoing

[05:30] that usually helps more. And so chances

[05:33] are if you're good at one language and

[05:35] you do it again in that language, you

[05:38] actually end up with a way better result

[05:40] than if you move to a new language where

[05:42] you don't know all the pitfalls and all

[05:43] the potential issues that are coming

[05:45] with that.

[05:46] >> [gasps]

[05:47] >> But yes, but I suppose that a far a far

[05:50] lower effort or engineering effort is to

[05:53] use some of the things you're going to

[05:54] talk about and maybe some of the tools

[05:56] that you're going to talk about in this

[05:58] talk and that will probably illuminate

[06:01] some of the uh some of the issues you

[06:02] might be having uh with with with

[06:04] performance.

[06:05] >> Yes indeed.

[06:08] And have you have you on on well you

[06:10] might not want to answer this but on

[06:11] projects you've worked on have you had a

[06:13] really embarrassing uh find where you

[06:16] suddenly gone oh my goodness this is the

[06:17] reason it's it's it's so slow.

[06:21] >> Um like yes like so due to due to the

[06:25] nature of our product like we we get

[06:27] thousands of requests per second with

[06:29] with customer data um that that we use.

[06:31] So like we have some some endpoints that

[06:33] are rather performance critical and um I

[06:37] mean I wouldn't necessarily like say

[06:39] embarrassing. That's just uh the nature

[06:41] of working in a in a small self-owned

[06:44] company without like billions of startup

[06:47] money where you do stuff and then you

[06:49] see in production if it works and

[06:51] especially if you work on a tool that

[06:53] helps people run their software in

[06:54] production, measure it in production, it

[06:56] also makes sense that we use our own

[06:57] tool to look at our own software. So one

[06:59] of the things um that we did was the the

[07:02] doctrine 3 upgrade and um my boss

[07:05] Benjamin is also one of the core people

[07:07] of the doctrine project and we upgraded

[07:10] that and then during that process we

[07:12] learned that a couple of the custom

[07:13] things that we've done around doctrine

[07:15] to make it faster didn't fully apply to

[07:17] the new version. So like after the

[07:19] upgrade things were quite a bit slower.

[07:21] Uh and it was really fun to see like

[07:23] like we we made the deployment and then

[07:25] our software told us like hey the thing

[07:27] is five times slower now and it's like

[07:29] ah cool um like that is like one of the

[07:31] normal stories where you know you're

[07:33] doing something but also over the years

[07:35] there is a bunch of stuff where we where

[07:37] we looked at something I was like

[07:40] not sure why we're looping over this

[07:42] thing every time to find an element here

[07:44] like this is a very long list uh like

[07:48] this is really like some PHP level thing

[07:50] where like it's a million elements in an

[07:52] array and like we're looping through it

[07:53] to find something and uh we have so many

[07:57] hashmaps in PHP. This might actually be

[07:59] one of the good use cases for that.

[08:01] Let's let's do that. and then like yeah

[08:03] seconds of that stuff goes away or like

[08:05] you're not clearing some some um or m

[08:08] cache or something and that means the

[08:11] thing has to loop through a lot of stuff

[08:12] and it's quite fun to see when when

[08:14] using a tool that shows you like all the

[08:16] function calls that a process makes and

[08:19] then it's like

[08:20] >> string leng 50 million it's like ah that

[08:23] seems a bit off

[08:25] [laughter and gasps]

[08:26] maybe we shouldn't be doing that.

[08:28] >> Wow. So when you look at these things, a

[08:30] lot of the some of the things just jump

[08:32] out almost immediately, do they? Oh,

[08:34] well, that's something to investigate.

[08:36] There's some inner loop we're going

[08:37] around millions of times. And

[08:39] >> yeah, I mean, that's the that's the fun

[08:41] moments when you when you open it up and

[08:42] you immediately find something. But the

[08:44] other the other fun part I think of

[08:46] performance work is also like buckling

[08:48] down with a problem for an hour or two

[08:50] or uh longer and using a tool and like

[08:54] really looking at like the individual

[08:56] steps and where the time is spent and

[08:58] trying to make something like trying to

[08:59] make something 10% faster usually that

[09:02] is already fast is a lot harder than

[09:04] trying to make something a lot faster

[09:06] that when when when there is like these

[09:08] obvious errors. So it's both of these

[09:10] things can be fun. Yeah, I've certainly

[09:13] had some embarrassing uh performance

[09:15] related uh problems on on projects. Uh

[09:18] one of them we were running is ages ago

[09:21] we had a like a Jenkins box for for CI

[09:25] >> and the test suite took 45 minutes to

[09:28] run and back then we were using Garrett

[09:31] so every single commit had to had to go

[09:34] through validation and there were bits

[09:36] where we would leave it on in the

[09:37] evening and come in the next day and it

[09:39] was still running. And we're like, "Oh,

[09:40] we've got to get these tests down." And

[09:42] then somebody looked at the

[09:43] configuration. They said, "Why have you

[09:44] got XD debug running on the Jenkins

[09:48] thing and they like flicked the flag and

[09:50] suddenly the test went from like 45

[09:51] minutes down to 15." I [laughter] mean,

[09:55] it was it was like I spent three days

[09:57] trying to speed it up prior to that. So,

[09:59] that was super embarrassing that it was

[10:00] just something as simple as a as a flag.

[10:03] But uh [laughter]

[10:03] >> it's a good one.

[10:05] >> Yeah. Yeah. These these things happen.

[10:08] Um yeah and and you actually so tell me

[10:12] a bit more about Tideways because you're

[10:15] one of the sponsors for the conference

[10:17] and it's quite related to what you're

[10:20] talking about. So what does Tideways do

[10:22] if people have never heard of them and

[10:24] >> Yes. So

[10:25] >> engage

[10:26] >> yeah we're super happy to sponsor the

[10:28] conference like after you've accepted my

[10:30] talk we we talked internally and like if

[10:32] I'm if I'm going there and if you're so

[10:33] nice to invite me actually we might as

[10:35] well sponsor. Um, so we're a a small

[10:39] company from Germany um with um that

[10:43] focuses exclusively on on PHP and we're

[10:46] an APM like you might know data do or

[10:49] Sentry or other products. Um, and what

[10:53] we're doing and what what our core focus

[10:55] is is that we exclusively focus on PHP

[10:58] on on making you not write all your

[11:00] instrumentation code yourself or

[11:01] installing a bunch of libraries or like

[11:04] doing a lot of the upfront work to get

[11:06] performance insights. But what we want

[11:08] to achieve with with what Tideway is

[11:09] doing is that you can on board with us

[11:12] and you immediately get meaningful

[11:14] insights because the PHP extension that

[11:17] we have auto automatically instruments

[11:19] your code for you and you don't have to

[11:22] do anything except install that

[11:23] extension and then you already get

[11:25] insights um that um and the product is

[11:28] like split in in in three main things.

[11:30] So we have a monitoring product where

[11:32] you can see your performance over time

[11:34] and you see if something gets slower. Um

[11:36] we have error like all the error

[11:38] tracking things um that uh we instrument

[11:42] like all the error handlers of major

[11:44] frameworks and grab the errors from

[11:46] there and um everything that you would

[11:49] expect to to see if your application is

[11:51] healthy. And then we also have like a

[11:52] tracing and profiling thing where you

[11:54] can like with um

[11:57] xh or xdebug or php spx would be some

[12:01] open source tools to mention. Um and

[12:03] with tideways like you can go to your

[12:05] own website click a button and then you

[12:07] see every function call that happens on

[12:08] that page and like how long it took and

[12:10] how everything is structured. But we

[12:12] also have like a low low overhead mode

[12:15] where we grab the most important bits of

[12:17] your application and so you have

[12:19] continuously like oh someone said the

[12:21] website was slow and you don't have to

[12:22] then start looking at it but you can

[12:24] look back at the data that's already

[12:26] collected and see if it's slow. Yeah.

[12:29] And yeah

[12:31] >> I see. So so what we would do if you

[12:34] were using your product is you would

[12:36] just install something on your

[12:37] production server and it would collect

[12:40] all this data in

[12:42] not a very intrusive way.

[12:44] >> And then and then how do you view it

[12:46] all? Is that do you go to the tidalist

[12:48] website to view it all or does all that

[12:50] data?

[12:51] >> Okay. So all the data gets all the

[12:52] relevant data gets transferred and you

[12:54] can you can view it.

[12:55] >> Exactly. So the way this works is you

[12:56] install our PHP extension and after the

[12:58] PHP request is done um that data then

[13:02] gets sent to a local thing running on

[13:04] your server um that's written in in Go

[13:07] where we do a lot of sanitization stuff

[13:10] because we don't want to have your

[13:11] customers email addresses. We don't want

[13:12] to have any private information. We

[13:14] don't want to see the strings in your

[13:15] SQL queries like this is all stuff that

[13:17] you don't want us to see and that we

[13:18] don't want to have. Um so that gets all

[13:21] stripped out on your system and then

[13:22] sent to us where we aggregate it, show

[13:24] you charts, generate alerts um for like

[13:28] we have like a big database of known

[13:30] problems of things that are slow or like

[13:32] of configuration issues and stuff like

[13:33] that. So it al also checks your systems

[13:36] configuration. Um or it says like hey

[13:39] you're running like a magenta juice shop

[13:41] and you you're running into this known

[13:44] problem. Um here is like how to fix

[13:46] that. And so you don't have to you don't

[13:48] even have to spend the time of figuring

[13:50] out what the problem is. You can just

[13:51] use like our observations and our

[13:53] learning to do that. And then on top of

[13:56] that you get like the continuous

[13:57] monitoring to see if something um yeah

[14:01] uh gets gets worse in the in the future.

[14:05] I see. So you can set up what alerts as

[14:08] well. So if things start to slow down

[14:11] then you get alerts say well look this

[14:12] used to be really quick and now it's now

[14:14] it's not so quick. Yes, like we can have

[14:17] you can set you can send us when you

[14:18] make a release and then you get a

[14:20] release comparison for like did it get

[14:22] better or slower? Are there more errors?

[14:25] Like you can get you get all the error

[14:26] monitoring and things you would expect

[14:28] from like a a monitoring solution. But

[14:30] we tried to be like a an affordable like

[14:33] a really affordable one piece package

[14:35] where you maybe you don't get like the

[14:38] full insight that you would get with

[14:39] data dog but that you also don't need a

[14:41] full operations team to look at it and

[14:43] really try to be a solution for people

[14:45] that don't have massive engineering

[14:47] teams. Um, while also allowing people

[14:51] that do performance the whole day to

[14:53] give them the tools with the tracing and

[14:55] the deep insights that they don't have

[14:56] to like set this up on every developer

[14:59] machine and they can use our tools and

[15:01] do that do local testing and do testing

[15:04] in production for performance. Yes. And

[15:06] we do this with with a with a small team

[15:09] um with Benjamin, me and a couple

[15:11] colleagues of mine, one of whom is a PHP

[15:13] core contributor that I'm very happy

[15:15] that Tide is sponsoring some time to

[15:17] work on PHP. Um I do some work on PH for

[15:21] PHP as well and like it's it's very nice

[15:23] that we're also able to give back given

[15:25] like our exclusive focus on PHP. Of

[15:27] course, we also have a vested interest

[15:29] in the language. Um but with with the

[15:31] Sentry open source pledge like we also

[15:33] support that fully and give to the PHP

[15:35] Foundation. I think we're like one of

[15:37] the bigger contributors there and the

[15:39] funding member. Uh so yeah that that

[15:42] make that that makes it really nice to

[15:44] to work as part of the ecosystem there

[15:46] while still like providing a product

[15:48] that uh allows us to to pay people.

[15:51] [snorts]

[15:52] >> Yeah. Nice. And I suppose actually

[15:54] you've you've um referred to the fact

[15:56] you do things for the PHP language, but

[16:00] you've actually been you actually the

[16:01] release manager, aren't you, for

[16:03] >> Yes. So for PHP 8.5, I'm one of the two

[16:06] release managers together with uh Daniel

[16:09] Shhatzer from from the from the States.

[16:12] Um we we put out the 85 release that has

[16:16] uh released with the do.0 in November

[16:19] and now we we get to do all the dot

[16:21] versions. Yeah, that was a lot of fun as

[16:23] well and learning and getting to know

[16:25] more of the core people and getting to

[16:27] know more about how PHP works internally

[16:30] after being on the mailing list for many

[16:32] years and doing small things like my my

[16:35] colleague um Tim also has helped me to

[16:39] write a couple PHP RFC. So there's like

[16:41] four or five small language features

[16:43] that I had a hand in. Like I do write

[16:47] some C but I'm not extremely fluent in

[16:49] PHP core. Like I did m maintain our

[16:52] extension before we found him who does a

[16:55] way smoother job at it. And I'm very

[16:57] happy that that we have him as well uh

[16:59] with with deep PHP engine knowledge like

[17:02] we also managed to make our own

[17:03] instrumentation a bit faster and have

[17:05] some fun there. Yeah.

[17:07] And so I suppose yeah that does work

[17:09] quite well you know getting involved

[17:11] with the core of it and doing what you

[17:12] do for your for your day job. But how on

[17:15] earth did you or what made you think oh

[17:19] I want to be a PHP 8.5 release manager?

[17:23] Um, so when it like when the

[17:26] announcement was made that that they're

[17:28] looking for new people, I think only two

[17:30] or three people put their name up for

[17:32] for a vote. And I thought um

[17:36] like it it would be nice if there would

[17:38] be just a couple more uh people

[17:40] volunteering for these types of things

[17:42] because it's not like it's not that much

[17:44] uh uh work but it's also something that

[17:47] that needs to be done and people should

[17:49] have a choice there um in in how they

[17:52] want who they want to have doing that.

[17:54] So, I put my name up for that and

[17:56] apparently um the the internalist

[17:59] community or the people that are allowed

[18:00] to vote on these things um for having

[18:03] contributed to PHP in the past decided

[18:05] that they would also like me to do that.

[18:07] Um yeah, and it was I I think it was a

[18:10] nice way to to give back um to PHP in a

[18:13] way that doesn't require like extensive

[18:15] C knowledge.

[18:17] >> Um I see like Oh, so you don't you don't

[18:19] need to be wellversed in C to be the

[18:22] release manager then.

[18:24] like it it it helps to be able to read

[18:26] it and um to to be able to judge like

[18:29] pull requests, but even that I would say

[18:31] is optional um like you learn some

[18:34] things along the way and of course like

[18:36] having a strong PHP background also

[18:38] helps but there is not a ton of

[18:39] decisions that you need to make and the

[18:41] great thing is that you have all the um

[18:44] all the great people from from the core

[18:46] team you have all the foundation people

[18:48] to talk to and they're very responsive

[18:50] and very nice and like everyone has a

[18:52] vested interest uh in the fact that the

[18:54] next PHP release is as great as the last

[18:56] one. So, um people are also very willing

[18:59] to help and like it's more of a like

[19:01] going around making sure everyone is

[19:04] happy, all the check boxes are checked

[19:06] and everything is going according to

[19:08] plan and smoothly than it is to like do

[19:10] very complicated engineering work. Uh so

[19:13] I think it's something more people could

[19:16] consider if they if they like want to

[19:17] give back to to to the PHP project in

[19:20] that way.

[19:21] >> Yeah. And and if anyone is uh coming to

[19:24] the PHP UK conference and they want to

[19:26] find out more presumably that you're

[19:27] happy to talk to them in the in the

[19:29] hallway track.

[19:30] >> Absolutely. Like um I'm I'm happy to to

[19:33] walk people through how the RFC process

[19:35] work like how you can contribute

[19:36] features to PHP, how you can propose

[19:38] feature ideas like in

[19:42] I think around the time of the

[19:43] conference at a bit later we will have

[19:45] the voting on um the next uh release

[19:48] manager for for 8.6. I think that

[19:51] process will start soon

[19:54] um as the release is coming out in

[19:56] November and there's like a 3 months I

[19:59] don't know uh I don't know from the top

[20:00] of my head period before that will start

[20:02] with the voting quite soon. So uh yeah

[20:04] if someone's interested in that and

[20:06] wants to hear like the the experience of

[20:08] that which is very positive but I'm

[20:10] happy to go into more detail there in

[20:12] person. Yeah come come find me talk to

[20:13] me.

[20:15] >> Brilliant. And and how much effort does

[20:16] it take? I assume a lot of it is within

[20:19] the final three months before the

[20:21] release or

[20:22] >> yes same throughout the year.

[20:24] >> The the great thing about having two

[20:26] people do that is that um basically

[20:29] every release is done by one person and

[20:31] then you're off for the next one. So um

[20:34] there is a there is a bit more of a time

[20:37] commitment in the um in the time leading

[20:39] up to that where we have around 10

[20:41] releases if everything goes well. So

[20:43] like four alpha releases, couple beta

[20:45] releases, a couple RC's, and then the

[20:48] release candidates are like the the more

[20:50] engaging phase where you make the most

[20:52] of the decisions as a release manager

[20:54] like which which feature still fits in

[20:56] the new next release, which feature

[20:58] needs to wait a bit because it's

[21:00] changing too many things or it's

[21:01] breaking stuff um and figuring figuring

[21:05] these things out. Um, but

[21:08] I say it's a rather manageable time

[21:11] commitment. Like now going forward for

[21:13] the next two years, I think like it's a

[21:16] Tuesday and a Thursday like a couple

[21:17] hours on on a Tuesday and uh an hour on

[21:20] a Thursday every month like once every

[21:23] month. So this is that's quite okay. And

[21:26] before that it's a bit more involved.

[21:28] Um, but it's definitely not uh something

[21:32] that you need to take time off from uh

[21:35] from your day job for to to do that.

[21:37] Especially if you work in a PHP company

[21:39] and this is like something you can talk

[21:41] to them about. It's like, hey, is it

[21:42] cool if I do this for a couple hours uh

[21:45] a month?

[21:47] >> Yeah. Well, thank you very much for all

[21:49] your efforts uh for because it's a great

[21:51] release 8.5.

[21:52] >> Thank you for putting up a conference.

[21:55] [laughter]

[21:55] >> That's right. Um I I because I actually

[21:59] that's that's another thing you talk

[22:00] about releases like people some people

[22:02] might just think oh it's just the big

[22:04] November release but actually there's

[22:05] lots of like patch releases throughout

[22:08] throughout the year. Is is there is it

[22:10] just when something needs doing or did

[22:12] you try and keep it at regular

[22:13] intervals? So this is kept on a regular

[22:16] interval and as a the release manager

[22:18] yourself you don't have to do a lot

[22:19] there because the people that work on

[22:22] core and work on um like assume like

[22:25] let's say there's a bug fix like some

[22:27] some some someone from from the PHP

[22:29] source team um will identify like in

[22:32] which version does that bug fix need to

[22:33] go into and they will make sure it lands

[22:35] in all the right branches. Um, and then

[22:38] you just come by and package all of that

[22:40] up and they probably have already taken

[22:43] care of the news uh as well. So like all

[22:45] you need to do is like look in the news

[22:46] file that's in the repository and then

[22:48] you already have your change log ready.

[22:50] Um so most of that work is done there

[22:53] and the um

[22:56] the more involved bits are more uh

[22:58] decision- making than like running

[23:00] running around and figuring that out

[23:02] just because the people who who work on

[23:04] PHP source are so diligent in in what

[23:06] they do and are so easy to work with.

[23:09] >> Sure. So I suppose in terms of uh

[23:12] deciding which features go in does that

[23:14] mean there might be an RFC which has

[23:15] been approved and all the work has been

[23:17] done but due to the nature of the

[23:20] changes you might say oh we're not going

[23:21] to put this in 8.5 we'll have to wait

[23:23] till 8.6. Is it those kind of decisions

[23:25] you need to make or does every RFC

[23:27] that's complete in time actually make it

[23:29] in?

[23:29] >> Everything that is completed in time

[23:31] makes it in. So it's not like um as a

[23:34] release manager you can say like oh that

[23:35] thing that's already in we're not going

[23:37] to release that uh because that's

[23:39] already in. But there is a there is a

[23:42] couple months time period where um you

[23:44] get to voice your opinions and um

[23:48] potentially reject changes that like oh

[23:50] we already released a couple alphas,

[23:52] people already started testing and now

[23:53] we want to make this like big

[23:54] fundamental change that is just a bit

[23:57] too much or a bit too dangerous or like

[24:00] someone has to make the call on like is

[24:02] this really a bug fix? Is this a

[24:03] feature? Is this a refactoring that is

[24:05] really necessary now? like what's the

[24:07] what's the benefit of doing this now

[24:09] versus doing that later. But yeah, all

[24:11] the work that has that has been done or

[24:13] that's going on during the year will um

[24:16] just be released. Let's go with that.

[24:19] And thankfully like everyone that's

[24:21] contributing to to core also seem like

[24:24] all the all the bugs I've seen all the

[24:27] problems I've seen are usually like just

[24:29] picked up by the people who um worked on

[24:32] that area because they all like want to

[24:34] have the stuff that they care about look

[24:36] good and work well. Um so it's it's it's

[24:38] not a lot of like you don't really have

[24:40] to run after people to get stuff figured

[24:42] out.

[24:44] >> Yeah.

[24:45] >> And I expect you get this question a

[24:46] lot. Who decides that the next release

[24:49] was 8.6 and not version 9? How's that

[24:53] decision made?

[24:55] >> I think there isn't like a single person

[24:57] that decides that. Um

[25:00] my understanding is that this like

[25:04] making the new version a major version

[25:05] just needs an RFC. So it basically needs

[25:08] like someone to say, "Hey, we should

[25:10] make the next version the the 90

[25:13] version." Um I think this is more of a

[25:16] natural thing of people coming together

[25:18] and being like oh like we have all these

[25:20] deprecations we maybe maybe we have like

[25:23] a really killer feature around the

[25:25] corner or like there's a big refactoring

[25:27] that we like someone has made like a

[25:29] great change but that is easier if we

[25:32] could clean up some of PHP's past uh and

[25:36] where the the people that work on on

[25:39] source come together and say like now it

[25:41] feels like a good time. Um, so like I

[25:45] know there has been some discussion if

[25:46] there should be an 86 or if there should

[25:48] be a 9. There's like maybe there is some

[25:51] marketing consideration, maybe there is

[25:53] some consideration of like we don't want

[25:55] to pile up all these deprecations and

[25:57] like do we already have a super clear

[25:59] migration path? How did the last release

[26:01] go?

[26:02] >> Yeah. So it it it's more of a community

[26:04] effort in that sense that just a lot of

[26:06] people discuss this and if it feels like

[26:08] the time is right, someone will make a

[26:10] proposal and then it's voted on.

[26:12] I see. And you largely are trying to get

[26:15] all the breaking changes

[26:17] uh just in like major releases.

[26:20] Is that the Yeah. Yeah. I mean

[26:22] [clears throat]

[26:23] >> minor breaking things in the in the in

[26:25] the you know

[26:26] >> Yeah. I mean some sometimes there's like

[26:28] a bug fix in in an external library that

[26:31] you could consider a breaking change and

[26:33] how it processes stuff, but that is also

[26:35] like technically the right thing it

[26:37] should be doing and a behavior people

[26:39] don't rely on. But like I think PHP has

[26:42] learned a lot from um 53 to 7 and or

[26:48] like when when seven came up like and

[26:51] then 56 I think was it and then we went

[26:53] to seven but like that wasn't an easy

[26:55] migration. Four to five wasn't an easy

[26:57] migration for people. I think um the

[26:59] industry took some lessons for there and

[27:02] then Python 2 to Python 3 reinforced

[27:05] those lessons as well with like there is

[27:08] a bunch of stuff that where people say

[27:10] like oh this looks a bit nicer but it

[27:12] just means you have to touch every every

[27:14] file of code that you have and it's not

[27:16] really a tangible benefit and I think

[27:18] the people that work on PHP have a good

[27:20] sense of how much of the internet is run

[27:23] by PHP and how how important that

[27:26] language is for the health of the web

[27:28] also like being a language that isn't

[27:30] funded by like one big company like

[27:32] there is no single organization that

[27:36] runs PHP or pays for it or does a lot of

[27:38] stuff like if you look at um the open

[27:41] collective page like there is

[27:44] a bunch of companies but but none of

[27:46] them has has has like a major share in

[27:48] the contribution and there is not like

[27:49] there isn't like a company that hire I

[27:52] don't think there is any company that

[27:53] hire employs more than two maybe three

[27:56] PHP core people

[27:58] >> and like those are not full-time either.

[28:01] >> Um I'm sure there is a company uh and

[28:04] I'm I'm I'm sorry for that but like just

[28:05] from the contributions that I'm seeing

[28:07] like this is a lot of the stuff comes

[28:09] from the PHP Foundation. Um, and it's

[28:12] not like there is something like with

[28:14] Golang where you have a billion dollar

[28:17] investment from one of the tech giants

[28:19] or something with TypeScript or

[28:20] JavaScript that big companies have an

[28:23] interest in owning the egg ecosystem to

[28:25] to to sell stuff. Um so yeah this works

[28:28] a bit different for for PHP I think and

[28:31] um that freedom also is quite nice

[28:34] because everyone that works on that

[28:35] really works works on it because they

[28:37] want to make it better and they want to

[28:40] contribute to it and not for oh like I'm

[28:42] just paid to do this like it's really

[28:44] nice attitudes of everyone.

[28:47] Yeah, it's always seems a really nice uh

[28:49] community, you know, when and every time

[28:51] you speak to people like the they're

[28:54] always really friendly and helpful. So,

[28:56] it's it's it's always good, I think.

[28:58] Yeah.

[28:58] >> And then from and then from the adaption

[29:01] that we're seeing like with our

[29:03] customers and hideways, there is of

[29:05] course still people who run seven and I

[29:07] hope they do that on on plans that where

[29:11] you still get external support for PHP.

[29:14] Um I think there is still a couple

[29:17] companies like with with tideways itself

[29:19] we still support PHP 72 or 73

[29:24] um like we we do support PHP 5 as well

[29:27] like we just don't change our um tooling

[29:30] um for that much anymore and um some

[29:33] things still work but I'm not seeing a

[29:35] lot of five out there um I think those

[29:37] projects are uh not being actively

[29:39] worked on um and more in in a in a very

[29:42] maintenance state But um especially for

[29:45] e-commerce which is also a chunk of the

[29:47] people that that work with us um 74 we

[29:50] still see some um and I think that's

[29:53] just the natural the natural way of

[29:55] having a language such a long and

[29:57] storied history and that has provided so

[29:59] much value that there is so much stuff

[30:02] out there that also just works um where

[30:06] I think security and uh is is more of

[30:08] the the reasons we eventually upgrade

[30:10] than anything else.

[30:13] M so that when we do eventually go to

[30:17] version 9 that pathway should be easier

[30:21] for people than say 5 to 7 or 7 to8 that

[30:24] that's that's kind of the hope is it

[30:26] >> I mean it is it I think that has always

[30:28] been the hope um I think we we can also

[30:31] make this a reality this time like um I

[30:34] think the the next um like all that if

[30:39] if you have a code base on 85 without

[30:42] seeing a lot of deprecations. And from

[30:45] what I'm seeing, like we do collect

[30:47] deprecation information from all of the

[30:49] people that run Tideways if they

[30:51] activate that feature. Um, it's looking

[30:54] quite good. Like there isn't a lot of

[30:55] like core stuff that that gets people.

[30:58] Um, I think if you run a mostly

[31:01] deprecationfree thing by either like

[31:04] looking at your logs or having a tool

[31:05] like like tideways or using some other

[31:08] way to to track that, then the upgrade

[31:10] to 9 should be rather painless from what

[31:13] I'm seeing now. But of course there will

[31:16] might also be a reason why we switch to

[31:18] nine to to change some other stuff. Um

[31:21] but I think like I said like everyone in

[31:23] that that works on core is quite aware

[31:25] of the of the impact of that and uh I

[31:28] think one of the important parts to

[31:29] realize about the PHP community is also

[31:32] that it is a lot of smaller communities

[31:35] that all don't talk so much to one

[31:37] another like WordPress is a big thing

[31:40] Drupal is a big thing is a big thing

[31:42] Laravel is a big thing and like all of

[31:44] these like communities exchange ideas

[31:46] but they also all do things a bit in

[31:48] their own way do these things a bit

[31:50] different and I don't think there is

[31:52] like the one right way to to to write

[31:54] PHP which is also the challenge for core

[31:56] like that you can't say like oh like

[31:59] here we here's a feature that should

[32:00] that make makes only sense if you do

[32:02] functional programming I think

[32:04] personally is something that really

[32:05] doesn't fit into PHP uh because like

[32:08] everything that the the core language

[32:09] has should serve all of these

[32:11] communities and like of course there is

[32:14] some APIs in PHP that are like only

[32:16] object-oriented um and that's something

[32:19] that has the language has built around.

[32:21] But like it it managed to stay rather

[32:23] generic to serve all of these different

[32:25] needs. And given that all of these

[32:28] communities also do provide feedback in

[32:30] their own way with for example Symfony

[32:32] and PHPUnit being really great at

[32:34] integrating against the current version

[32:36] like the the master the the master or

[32:38] main branch of the PHP repository. Uh

[32:40] and then you get feedback very early

[32:42] like when we broke something during the

[32:44] the first uh or second alpha release

[32:48] like I I was talking to the folks from

[32:50] composer about like hey like the symfony

[32:53] install broke this broke because of

[32:55] composer this broke because of this

[32:57] thing that like actually I put into PHP

[32:59] and then we got that fixed in a day uh

[33:02] just by having these like very quick

[33:03] feedback cycles. Uh and that's really

[33:06] cool.

[33:08] Yeah, I suppose it's really good that

[33:09] you do have these people kind of running

[33:11] really at the bleeding edge just to feed

[33:14] back those those kind of issues. And I

[33:17] imagine also like the whole all of the

[33:19] tech stack is is getting a lot more

[33:20] mature. Like you've got really good

[33:22] static analysis. You've got tools like

[33:24] recctor to help people upgrade from one

[33:26] version to the next. So I I kind of

[33:28] imagine that um and actually I think

[33:31] people are doing a lot more testing. You

[33:33] know, you've got PHPUnit, you've got

[33:35] pest, I suppose. I suppose that does

[33:37] mean that when people upgrade, you know,

[33:39] if they've at least got a test suite,

[33:41] you know, it can act as some kind of

[33:43] smoke test what what you know the what's

[33:45] going on in PHP. Uh so I suspect if all

[33:49] of those practices improve and people

[33:50] take uh you take up all of these tooling

[33:53] then that does mean that your feedback

[33:55] will be will be a lot quicker I suppose.

[33:58] >> Absolutely. Yes. Like I think there is

[34:00] like the people you talked about now I

[34:03] think mostly are like application

[34:04] developers that work on more custom

[34:06] applications. Then there is of course

[34:08] like a big chunk of the ecosystem is

[34:09] people running software written in PHP

[34:11] is people running Drupal is people

[34:13] running Shopware um or or other things.

[34:17] And for them they like they upgrade when

[34:20] the hostet tells them to upgrade or they

[34:22] update when um

[34:24] >> they they run out of security support

[34:26] for a thing hopefully. and um or they

[34:29] update because the new version has like

[34:31] performance benefits that they've heard

[34:32] about and they want to see that or to

[34:34] get rid of a bug or something like this

[34:35] is usually what motivates or they just

[34:37] like have a well organized IT setup

[34:39] where they just want to be on current

[34:41] versions for like to to be protected uh

[34:45] always and to to manage their

[34:46] infrastructure. And like these are two

[34:49] very different feedbacks that you get

[34:51] from from people who can fix their own

[34:53] bugs or from people that are dependent

[34:55] on vendors to to do stuff. Um but of

[34:57] course uh you get more detailed and more

[35:00] frequent feedback of course from the

[35:01] people who actually like develop these

[35:03] applications than you get from the

[35:05] people who run them.

[35:07] >> Yeah. Yeah. And sometimes you know you

[35:09] see the I don't know sometimes I learn

[35:12] about language features you know when

[35:14] you have this what's new in PHP the

[35:16] latest version and then normally at the

[35:18] end of those talks people have a whole

[35:19] series deprecations and stuff they've

[35:21] been removed and often I learn about

[35:24] language features at the point they're

[35:26] being removed because they just seem

[35:28] some of them just seem absolutely crazy

[35:29] that they they existed in the first

[35:31] place and I imagine uh if you're quite

[35:34] conservative about your PHP usage and

[35:36] you stick to like uh like just the

[35:39] fundamental main bits, then you're

[35:41] probably going to be okay on the

[35:42] upgrades really. I can never really

[35:44] remember having any difficulty on a a

[35:46] PHP version upgrade, you know. say even

[35:48] from 7 to 8 and 5 to 7.

[35:51] >> I think one of the one of the more

[35:53] frustrating things for um for that

[35:57] ecosystem is of course you you mostly

[35:59] hear from the most vocal people that

[36:02] struggle maybe with a tiny thing that um

[36:04] would take 5 minutes to fix and takes

[36:07] hours to complain about. Um but uh that

[36:11] is just people who might like have a lot

[36:14] of stuff to do on their job and now that

[36:16] also happened and now that is really

[36:17] annoying to them but it's actually not a

[36:19] big change. They might just not be very

[36:21] familiar with with with what they're

[36:22] doing there or they might not stay up to

[36:24] date. So thinking about these people and

[36:26] being respectful of their time I think

[36:27] is as important as people who like come

[36:30] to PHP with an issue and say this broke

[36:33] something also being respectful and like

[36:35] the easiest way to to make sure that

[36:37] your stuff works is of course to test it

[36:39] before we make a new release because

[36:40] then we can discuss that and change

[36:42] things. Uh but yeah just due to the

[36:45] nature and the age of PHP there is a a

[36:47] lot of stuff in there like I think PHP

[36:49] has always um over the years like been a

[36:53] language where if you had a very

[36:54] specific problem especially like in the

[36:56] in the earlier like 2 three four days

[36:58] like you could put that in PHP and you

[37:00] had like your very specific solution to

[37:02] your very specific problem and now like

[37:04] the standard library is not changing as

[37:07] much like new things like the um the new

[37:10] random extension or the the URI passer

[37:13] or something are built into their own

[37:15] name spaces so they fit like very well

[37:16] with um with the rest of the ecosystem

[37:19] but uh the the real core functions

[37:22] aren't touched so much because it's just

[37:25] hard to not break people's things

[37:28] >> um and like also for people who want to

[37:32] now put features in PHP like come from

[37:34] very different backgrounds in how they

[37:35] write applications and like what they

[37:37] want to do and um I think over the years

[37:40] PHP has adapted features from all other

[37:43] programming languages. Um, and like

[37:47] sometimes there I feel like there's a

[37:48] checklist of like, oh, this language has

[37:50] a feature like with a syntax that

[37:52] doesn't fit in PHP. Can we cram that in

[37:54] as well? Um, as a bit of a joke, of

[37:57] course, but like people get inspired by

[37:59] a lot of stuff and are like, why doesn't

[38:00] PHP have that? And because the language

[38:02] is so welcoming to to changes, um, a lot

[38:06] of that stuff then happens. So, uh, I

[38:08] think it's quite fun to see like all the

[38:10] little bits and pieces coming coming

[38:11] together.

[38:14] >> Yeah. Yeah. Well, I found this really

[38:17] interesting actually just finding out

[38:18] what happens uh I suppose from the

[38:20] release manager point and how the

[38:22] language is um put together and and all

[38:25] the additional constraints that you have

[38:26] to work with things that perhaps most

[38:28] developers don't think about because

[38:30] they just think about their own you know

[38:33] sphere of knowledge they know about. So,

[38:35] uh, thank you very much for all the work

[38:37] you do on we've done and for getting PHP

[38:40] 8.5 out. Uh, I'm looking forward to your

[38:42] talk at, uh, PHP UK conference, uh,

[38:45] which is on the February the 20th, uh,

[38:48] of, uh, 2026 in London. Um, so I look

[38:52] forward to to seeing you there and thank

[38:53] you for having this taking this time for

[38:55] this chat.

[38:56] >> Very excited to be there and, uh, yeah,

[38:58] looking forward to it and thank you very

[39:00] much. That was a great conversation, a

[39:02] lot of fun.