←back to thread

116 points doekenorg | 1 comments | | HN request time: 0.232s | source
1. zackmorris ◴[] No.44502882[source]
Wow, it's cool to finally get closure (:-P) on a question I asked in 2012:

https://stackoverflow.com/questions/12939319/coroutines-in-p...

Looks like generators were released in php 5.5 in 2013:

https://versionlog.com/php/5.5/

I was interested in coroutines because most backend server logic eventually devolves into a sea of state machines where each request/response advances the state by updating database rows. This becomes unmanageable by humans, which is why server code can only reach a certain level of complexity, perhaps 1 million lines, before it becomes "enterprise" and triaging overtakes architecting as the main mode of operation for developers.

This is akin to how in the 1990s, object-oriented programming (OOP) limited the size of most desktop programs to around 1 million lines, due to similar state management limits under imperative programming.

I had hoped to replace the state machine soup of backend API endpoints with coroutines that guided users through stuff like their onboarding steps along one-shot functions made up of mainstream conditional logic and higher-order methods.

This history explains why most websites and apps today have so little actual business logic. Most would be considered entry-level or semester projects for desktop developers in the olden days, who were forced to wrangle the complexities of C++ and Java.

Today, most time is lost to the idiosyncrasies of managing build pipelines, version updates, boilerplate, etc. Meaning that we're too mired in babysitting our tools to see how old school approaches like using spreadsheets and batch files in office environments would make a mockery of our work. Now it's mostly all a waste of time, and we can feel it, but I digress.

Anyway, I had attempted to make this state machine <-> coroutine bridge by saving php functions and their state using the jeremeamia/super_closure package:

https://packagist.org/packages/jeremeamia/superclosure

https://github.com/jeremeamia/super_closure

Which gave way to opis/closure:

https://packagist.org/packages/opis/closure

https://github.com/opis/closure

Which gave way to laravel/serializable-closure:

https://packagist.org/packages/laravel/serializable-closure

https://github.com/laravel/serializable-closure

That way the coroutine would get resurrected during each user request and proceed through its logic. Thereby removing the mental load complexity limit imposed by state machines and allowing 1 or 2 developers to compete with larger enterprise teams at big companies.

Since then, I've abandoned these types of approaches and moved towards pure functional programming (less state management and fewer side effects), declarative programming (repeatable processes that eventually meet constraints imposed by integration tests), and data-driven development (higher-order methods on trees and graphs). So lots of work with spreadsheets, Terraform, Firebase, etc. I find that programming languages mostly get in the way now.

After a career mostly spent hacking on legacy code and tearing my hair out, I yearn to be free to get real work done. This would look like abandoning most approaches people are pursuing today. For example, most of the async/await stuff in Javascript is an evolutionary dead end, because we already went down the cooperative threading road in the 1990s and discovered that there was no there, there. Async is today's goto. Same with absurdities like the "final" keyword, which is a self-flagellation habit born from difficulties around name-mangling when exporting C++ methods in object files. The compiler should reorder structures and classes to match the constraints of the runtime, not humans. Otherwise we break Postel's Law:

https://martinfowler.com/bliki/Seal.html

https://martinfowler.com/bliki/SoftwareDevelopmentAttitude.h...

https://martinfowler.com/bliki/DesignedInheritance.html

https://martinfowler.com/bliki/OpenInheritance.html

https://martinfowler.com/bliki/TolerantReader.html

With this context, we can see how large powerful companies have doubled down on the Directing Attitude and Designed Inheritance to the point that we're handcuffed to our tools. They've done little or nothing to advance the state of the art of our languages and frameworks from first principles to be more freeing by providing more leverage. For every revelation like Erlang and Go, there are countless AWSs and Reacts. Forcing us to focus on specifics like regions and edges, side effects and performance, etc. Because nobody did the real work of designing distributed systems that "just work" via techniques like true multiprocessing, memoization.. I could rant forever. It's all bare hands work now, by us serfs under neofeudalism.

Sorry this got long, it's a passion project for me, a dream I may never have time to live if the rest of my life gets lost to making rent.