PHP, Python and Ruby all feel to me like last-generation languages. IMO we should be building new frameworks on modern languages that support concurrency & static typing. The most promising new stack I've seen is Kotlin's KTOR framework. Thanks to the expressiveness of Kotlin it's not really any more verbose than something like PHP or Ruby but it's far more powerful and robust.
Alas, I feel the same way. After much thought this weekend, I concluded that it was time to move on from PHP. I decided against the any other scripted language (Ruby, Python), as it felt like more of the same.
Requirements were strict typing, functional programming support, built-in concurrency, decent library support, and real support for both native and Javascript. It pretty much came down to Kotlin or Rust.
I really want to like Rust more, but the library support, and the jobs in my area just aren't there (yet?).
Kotlin is really a fantastic language even without IntelliJ but with IntelliJ it makes these older languages feel archaic. Once you've had this level of explicit typing & null checking, code navigation, and refactoring that really works its very hard to go back.
Kotlin also feels like a pretty safe bet since both Google & Jetbrains are backing it.
No garbage collection, doesn't compile to Javascript[0] and everything on Android is moving to Kotlin anyway.
Just to be clear, my reasons stated above is why I'm moving to Kotlin. I make no claim that other people should or would want the same things. I've realised that I like languages with explicit strict typing, and that allow for a greater level of expressiveness at the cost of a higher intrinsic complexity and the comprehension that requires.
One of the main benefits of PHP for web serving, is the "shared nothing" architecture.
By operating as a non-threaded application, entire categories of bugs and security holes are avoided.
Are there situations where threads could genuinely help? Sure, there is always an exception to any rule, but for the vast majority of projects where PHP is used, I don't believe threading would achieve much in terms of performance gains.
There are very few places where PHP cannot do strict comparisons.
The first example in that article even highlights that `in_array` has a `$strict` parameter.
I'd bet a months salary it's easier to teach developers to use strict comparisons (either through arguments to be passed, or strict operators to be used) than it is to teach developers how to avoid thread-safety bugs.
Though, it might take longer to train them to spot every place where they could/should be using some stricter version of the defaults. That's the problem with opt-in safety.
On the other hand, there are languages that make thread-safety bugs much less likely. Some day those ecosystems will catch up to the languages that have been around "forever" and we'll get our cake and eat it, too.
There really aren't that many places to remember, and there is copious amounts of developer tooling available to identify, and in some cases even fix the issues automatically.
I don't recall ever seeing any kind of dev tooling that can just automatically fix thread safety issues.
What do you mean by concurrency here? JS like async or Java/C# like threads? And why would concurrency help in problems where PHP is involved, in the problems I solve with PHP I don't think I ever needed async. For unning more things in parallel like processing 1000 RSS feeds you can spawn 10 scripts that run in parallel and if one crashes the others still continue to work(I had experienced crashes with some RSS XML files crashing the built-in XML parser).
Anything that will block the server. Fetching data from an external API, doing some kind of expensive computation, sending an email etc. You can either handling this Node style with blocking i/o, throw the work onto a background queue like Sidekiq, or use a language with first-class concurrency support like Golang or Kotlin. To me Kotlin feels like a far nicer and more scalable language than Golang.
I like this about PHP, you don't have a callback hell. I don't disagree with you just sharing my experience here, we might work on different kind of problems.
I work on highly concurrent backends that handle 100k requests per second across a cluster, and PHP will never work in those situations. Any time you have to orchestrate multiple data store or service calls in the background, you don't want to block.
Callbacks are not how you do non-blocking IO. You typically have multiple threads and use constructs like futures or promises to make them easy to manage.
PHP bakes you into a very simple request flow. There's so much more you can build if you get out of that manner of thinking.
Yes, I never disagreed with that, best tool for the job. But your problem is not the usual stuff that web projects are about.
Edit: FYI I have experience with other languages too, so if at my work I use PHP does not imply that I am some inexperienced person that has no idea how other languages work. I do not have experience with your domain of high concurrency, low latency stuff but I worked with threads, and thread pools and multi-processes programs(it happened that some thread would segfault and bring the app down so we had to use background processes instead of threads) .
Just curious what problem are you trying to solve, I never had the need for different threads to communicate in PHP, most of the time in my work a user makes a request, and a PHp scripts does something like update the database or does some searching/working and returns a result.
What if that request was a document upload and your system is calling out some external api or processing for long enough such that you’d rather return now and update later when the job finishes. Maybe through Ajax or a redirect.
The shenanigans that WP plugin devs go through to get around the lack of a queue is ridiculous (chunk the file and keep reloading the page until we’ve processed all the chunks or have an Ajax endpoint that processes a chunk at a time...) and the when I asked a marketing friend of mine who works with Wordpress he said that redis is an “advanced requirement” and I should keep things simple.
In Django land I’ve always set up queues even just to send emails because it keeps things snappy.
For that kind of problem we create a database table. then you schedule there your work tasks and background scripts will fetch jobs from the database and run them. This also works great if the third party is busy or fails, the background jobs are set to attempt a few times before giving up.
What solution would you prefer for this kind of jobs? launching a new thread directly? or have a worker project always running and send to it the job directly?
.... As you identified at the end, the correct solution for this problem is a job queue. There is nothing about PHP that prevents the use of a job queue - in fact the shared-nothing, process-per-request model almost encourages the developer to offload as much work as possible to a job queue..
Using Wordpress (or it's plugins) as any kind of reference for what PHP is capable of, is like using McDonalds as the basis for a cook book.
If putting a job/event into a queue, or writing a log entry is taking so long that you need a separate thread just to enqueue/log that entry, something is dreadfully wrong with your application.
Ironically, the root cause that made me think of that example was AWS.
The example was where each SQS call (there could be as many as five) was taking between 50ms-200ms, and logging took another 75ms. The rest of the request processing time was approx 50ms.
There's a bunch of other architecture options we could have looked at, but due to the share-nothing nature of PHP, they weren't an option.
I do. However, the process would have stuck around until it finished. It wouldn't have changed our overall RPS at all.
Arguably the same problem would exist in other languages, but at least there we would have had the option of using threads, and we could have executed multiple calls to SQS in parallel. Probably not something I'd want to do all the time, but it would be nice to have had that option in this case.
I'm there with you about Kotlin. I started with PHP last "last generation" and was pretty bored with it. Kotlin was a breath of fresh air. Also, if you like some similar to Laravel/Symfony for Kotlin, give Alpas a try: https://alpas.dev/
You can defined what types/interfaces methods needs to return, what types/interfaces variables must have, what types/interfaces the method arguments needs to have.
It also supports class inheritance, abstract classes, interface classes, class traits, etc.
It does, and using PHP in strict mode, in combination with tools like phpstan is what led me to looking for a language that is even better in that regard.
If PHP hadn't made advances in it's type system with the 7.x series, I probably would not have stuck with it this long.
However, while PHP the language supports strict typing, it does it at run time (inherently a limitation of scripted languages, I know), rather than compile time.
Nah. The JVM must go; we have Docker and the universal platform are Linux ELF binaries so the entire "bytecode" concept makes sense for the compiler (i.e. LLVM), but no longer at runtime. It's just bloat.
Bytecode is simple enough to alter at load time. This gives you aspect-oriented programming and instrumentation of everything that statically linked binaries have never offered. And I don't think anything compares to the JVM's GC throughout yet; why roll your own if it's slow?
Kotlin native might eventually make the JVM superfluous but for now there are a lot of benefits to being on the JVM including great performance, a huge ecosystem of Java libraries, and very battle hardened monitoring and debugging tools.
Considering the GP commenter started with an emphasis on concurrency, Elixir's concurrency features (thanks to the Erlang VM) are totally different than what Python can offer, namely isolated processes (scheduled in the VM so like green threads) where you can write blocking or CPU intensive code without worrying about blocking the other processes and where GC is per-process. With Python you (AFAIK) run into the GIL when threading or you fork OS processes that are heavyweight so you can't do that for every request or you'll fall down. Or you have to write async code which can still only saturate one core.
Not saying Elixir is the answer to everything but that's where it differs in the concurrency sector. It doesn't have static typing though, which is a requirement for some.
