"I prefer a lack of magic for my programming [production systems]" the dev says while they write in a language for a compiler they don't understand, with algorithms in the standard library they couldn't implement, performing processor specific optimiza...

[gerbrand@fosstodon.org]

@samir @hrefna The problem with magic is that it leaks, spills over and doesn't compose.

[samir@mastodon.functional.computer]

@gerbrand @hrefna Yeah, it’s all leaky eventually (Spectre is a great example), but when we’re complaining about magic, we’re almost never complaining about `malloc` lying to us about available memory, are we?

[hrefna@hachyderm.io]

@samir

Spectre and malloc are both great examples.

Really at some point you just accept that everything you are writing is pseudocode that mostly runs in a way that approximates what you expect and is _good enough_. Most of the time.

Until it isn't.

@gerbrand

[hrefna@hachyderm.io]

@samir

You also don't have to scratch very hard in compiler optimizations either before you start running into it causing actual real problems or unpredictable behavior in production code.

Not even in a JIT either. In a standard AOT compiler.

That's without getting into Tomasulo's algorithm or how modern branch predictors work.

@gerbrand

[structuralist@mastodon.social]

@hrefna

It is all a rube Goldberg device.

[dontony@hachyderm.io]

@hrefna i like my footguns to stare at me in the face, not be an abstract concept that is nondeterministic

[hrefna@hachyderm.io]

@dontony Cool, enjoy programming on paper with pi calculus.

[lunch@tilde.zone]

@hrefna i mean idk about them but i just understand all that stuff there's no magic to me sounds like a knowledge issue ig

[hrefna@hachyderm.io]

@lunch Is your name Dunning-Kruger?

[hrefna@hachyderm.io]

@Structuralist worse, because a good half or more of the complexity is necessary complexity.

Rube Goldberg devices you can simplify and get the same output. The ball can still be moved.

No such luxuries here, where most of these things exist for a reason. This is less Rube Goldberg and more Chesterton's Fence.

[recursive@hachyderm.io]

@hrefna I like all the reasonably durable abstractions we can manage, but with a bias towards ones we can prove some things about

Although on some level things like this I consider a challenge

Like maybe I should learn all about Infiniband, just because

[lcannell@hachyderm.io]

@hrefna You may be interested in reading this Turing Award acceptance essay by Ken Thompson, published in the Communications of the ACM in August, 1984:

Reflections on Trusting Trust

https://www.cs.cmu.edu/~rdriley/487/papers/Thompson_1984_ReflectionsonTrustingTrust.pdf