I'm writing a devrant like site, so a kind of forum that supports live chat under every article. Login will be just username and password to stay anonymous. Email is optional for password reset. Also it won't have password requirements. Who cares if user uses insecure password. I do like the devrant avatar thing. I will use the ducky generator instead. So everyone on the site is a custom duck. K-SASS prolly never expected his generator to be used anywhere. The requirement of this site is that it scales very well. I have db calls of 0.006s, this is for persistent data only and will be used by all site instances. I expect that it can handle many clients concurrent as long I do not return more than 30 rows or so. Events get handled by a self written pubsub server.

All sounds great and development goes fine. But why is this a rant? Because the same thing as always is biting me, I can't design a site at all. I know how but I don't have any feeling for design at all making me almost incapable of building an attractive site. The only thing I can 'design' is an application in bootstrap or smth. I spend so much time one design while I don't like to do it ironically. But looks of site is almost as important as an good working site. Good working site doesn't get used if looks bad in many casee. This is since the start of my career an issue and it sucks that I appearantly can't deliver a whole site on my own meeting my standards.

My backend work is top notch tho. Btw, this application is not to be an alternative for devrant. I do not think I can attract more users than it already has and I've seen two communities disappearing once because someone decided to make a new one, took half of community with him and both communities died after short while.

End product of this project is a working project, not a live site hosted somewhere. It's pure about mixing mostly self written tech to get the best performance. Reinventing wheel on many levels. I wanted maybe to do the site in C but decided that it's way to much work for the value. I change the site so rapid since I don't have decent plan that python aiohttp is the best choice in amount of writing it yourself and fast. It's very lightweight.

More a story than a rant, sorry

About 4-5 hours ago I wanted to make simple websocket to get input from textarea and parse it on server and somehow got myself into developing in asyncio -> aiohttp -> graphql-core -> graphql-ws -> aioredis
and svelte-> typescript
I still didn’t make the stuff I wanted but I’m very close on backend at least.
I have some frontend part somewhere in my old prototypes so it will be faster if I figure out svelte.

Still don’t understand what the fuck just happened.
Maybe it’s because I wanted try those frameworks for a long time.

All ‘simple’ examples I found have around 20-30 files for backend and same amount for frontend so more then 50 files to get this shit working.
They’re called oh irony “simple chat”.
Now I see why no one fucking understands this shit.

I’m trying to cut mine to 5 files.

I thought developers are lazy bastards who don’t like write code.
But now after this they’re all looking like adhd coders.

Looks like Monday won’t be my best day.

I was so bored with work in the end. It was more administration than programming. So, i kinda quitted full application development. The thing is, it's expected that you use some existing framework. First of all - they never work how you want and the programming part of your work is mainly solving the limitations the framework brings without hacking too much. You keep within the boundaries of the framework. Besides that - since all fun stuff is already done by the framework builders all you have to do left is kinda administration. Field here, field there, rest call here. Extremely boring. When you've setup the base good, there's no challenge anymore, just producing windows and input forms.

Now, a few days ago, I started to make a clone of rocket chat. I use minimalist http framework (aiohttp) and you have to build most features yourself on top. Same for the ORM, i use dataset which does schema synchronisation for you but doesn't come with models. So i made a complete model / mapper entity framework on top of that. I made one single validation system that applies on models, forms and frontend validation. There's only one truth of valid data. Within the models, services, mappers and forms there's always the services variable available making it possible to fetch any data from any object. Never weird exceptions has to be done to get data. The implemented global LRU cache system is super in auto synchronizing the objects, don't have to do anything manually.

Finally software development of a full product is fun again. If you know how to do it - making your own framework is way easier than an existing one. On top of that, it's more advanced. I do understand that frameworks are aiming to be a bit minimal to be multi purpose, but with that attitude in mind, they still achieve to make it annoying as fuck.

Regarding time, it's just a few days of development. That's nothing for something that does exactly what you want. We have to drop the use-a-framework-because-it-is-stupid-to-do-yourself mentality. We should be programmers again! Not administrators! It's not weird that chatGPT can do so much of our jobs, our projects became lame.

Sooo. My team and I have module we're supposed to be porting to async code and aiohttp will not work. The server keeps rejecting the byte payload, but if we use synch code like the requests library, it works fine. The code is like identical, the only difference is async. It's been really frustrating because another drop in async version of requests (httpx) works just fine! I don't want to use httpx, the rest of our codebase is already using aiohttp! We think the problem is with gzip encoding being handled incorrectly by aiohttp. I've reported the issue.