|
It's not as if Python or Ruby have set out to be deliberately difficult. They've grown difficulty because they're used to solve real problems, and real problems are hard. Implied inside your question is the idea that we can, in advance, choose the right level of abstraction. That we have a library / language / set of verbs that is neither too high level (like Automator) nor too low-level (like your bash script example) to solve the problem. While I think we can improve a little bit on the set of abstractions that we have, I think a broad solution is naive. As a point example, I used to be frustrated that so-called 3D engines tended to grow until they included the kitchen sink. Why do I need a menu system as part of my 3D engine? Why do I need a sound engine? Can't I just use system libraries like a sane person? etc. All I want is a set of primitives that let me render models to the screen. How hard can it be? It turns out that the problem of rendering graphics to a screen quickly is so hard that it actually infects the hardness of other problems. Doing elastic collisions is a trivial thing that any first-year physics student can do. Until, that is, your game loop skips six frames because it's busy rendering graphics code, at which time you need to retroactively handle collisions that have occurred in the past. Now we're doing rocket science, and so your graphics library needs its own physics system, because you're not going to derive that stuff. And that, basically, is the story of every feature added to every serious game engine that exists today. You see this problem everywhere you look, for example, at one point people thought POSIX and C were reasonable building blocks that everyone would use, and at another point people thought that about Java. At first, BSD sockets was the primitive, now we're starting to think about HTTP connections as the primitive, maybe we should work to make REST requests in Python not suck so damn much? But at this point SPDY looks like it might be turning into a thing, so maybe we should just build primitives for that instead? If you look at this staggering tower of abstractions you start to see things that seem crazy, like SPDY including its own flow control even though TCP already has it, because TCP has some bad assumptions that should be really be fixed at the network level, but will never be fixed there as a practical matter. I hope you followed Inception, because if not you're gonna have a bad time. I'm doing a bad job of explaining this, but I had this insight first after reading an article by Kamp, the author of Varnish: https://www.varnish-cache.org/trac/wiki/ArchitectNotes. He makes the argument with a specific example that as we have moved to higher-level primitives, we've built an overly simplistic mental model of how the lower-level ones like OS and hardware work. As a result, we implement our own caching code, because we've forgotten that OSes were designed to do that for us. So in a sense we have more power because we have this nice set of high-level primitives, but in practice we've forgotten all the things that those primitives wrap, we've started to believe that they are simple because they look simple. And so, we muddy up our application code with all this stuff that has already been solved. I doubt very much that there is any software program that people actually use that is not either undergoing a big refactor right now or has one scheduled, either to make things "simpler" (less abstraction) or "more powerful" (more abstraction). Just thumbing through the projects that I work on, I am adding features to a mental backlog for a future refactor for every single one of them. This seems to be like a fundamental problem that all software projects have, and creating a new set of primitives and saying "no seriously, that should do the trick" is at best going to yield a nice little language like Python or Ruby that somebody will post articles about on HN saying why it's so hard for casual use. |
The problem of getting from the East coast to the West coast has not changed in 100 years (move body until done), yet when was the last time you had to set the spark and choke and hand-crank a car? Cars solve real problems, and today's cars are sophisticated enough that essentially anyone can use one. To solve real problems.
Certainly the problems solved by a car are in a more narrow domain than programming languages. But cars themselves are in a narrower domain than internal combustion engines.
Edit: stealth.