I would say it’s less reinventing and more recreating. Also if you want a red wheel but all of the ones available to you come in shades of blue, then if you make your own you can make sure it’s red.

Depends on the project. If it’s for fun, sure make all the square wheels you want and learn how to make them round. But if you just want your project to work you’d find something to use. Really depends on the developers goals.

Depends on the wheel.

I mostly work in Go when I have a choice, and it’s got a lot in the standard library. (The Go standard library doesn’t count as a dependency… or at least not an optional one.) When I write web (as in JS-in-the-browser) stuff, I don’t use any JS dependencies aside from browser built-ins.

Also, I don’t mean to imply I don’t use dependencies at all. But having dependencies that aren’t pretty much absolutely necessary is the kind of thing that ought to make one hate oneself a little more. Just a little self-flagellation for each dependency can’t hurt either. (Just to be clear, I don’t mean this literally.)

As an example, not long ago, I wrote a web-based virtual tabletop application (the kind of software you’d use to play Dungeons and Dragons remotely) in Go. Aside from the Go standard library, it’s got exactly three Go dependencies: a Sqlite3 driver, a library for minifying HTML/CSS/JS, and a transitive dependency of the minifier for parsing HTML/CSS/JS. The JS has zero dependencies other than browser built-ins.

The “wheels” I could arguably be said to have “reinvented” just off the top of my head:

• Go:
  • Facilities for building static assets into the compiled binary.
    • And serving those static assets, but that’s kindof one thing with the building into the compiled binary thing.
  • Authentication.
  • HTTP session management.
  • Server-side in-application message bus.
• JS/Web:
  • JS dependency management. (Something like RequireJS.)
  • Client-side templating. (Something like Handlebars.)
  • Running code on document ready/loaded.
• Somewhere In Between:
  • CSRF protection.
  • Server push (using SSE).

Now, I could pull in Handlebar and RequireJS and React and jQuery and Underscore and Gorilla and have a build system that depends on NPM and Bower and maybe has a Makefile to coordinate it all. But I really don’t see the benefit. Especially compared to the drawbacks.

And by not pulling in libraries for these features I’m saving:

• Performance.
  • Client-side JS is smaller.
  • The browser isn’t bogged down.
  • Quicker compile times.
  • The back end is more responsive.
• Less cognitive load and fewer moving parts.
• I know much better how these features work.
  • With FOSS dependencies, I could also know how things work, but honestly it’s probably faster to write it myself than look through the source code of a framework.
• Less hard-to-track-down bugs.

More reading relevant to avoiding dependencies and frameworks:

• tonsky.me/blog/disenchantment/
• idlewords.com/talks/website_obesity.htm

.

Yeah sure, better than pulling 1000Gb of node_modules…

Sometimes that’s better than having to learn other people’s interpretations of the same wheel over and over again.

I use it as a learning tool. When I was making my own framework it forced me to learn all the intricacies of the thing I was making the framework off. TBH it never saw any use in a project but the process of making it is a huge learning experience for me.