Llogiq on stuff

Yeah, but what *is* "modern" programming?

Recently Prof. D. Lemire wrote a blog post about his programming history and how – in his opinion – “modern” programming is different than what we did in ye olde days. I really like his writing, and completely agree with the article. Still, I have some additional ideas on what constitutes “modern” programming I want to share here.

Like Prof. Lemire, I started out with BASIC and assembly (what a coincidence), then settled on Turbo Pascal (before being shanghaied into Java during my CS degree).

My first computer had a Z80 CPU and 32 kilobytes of RAM. My first PC had one whopping Megabyte of RAM (of which 640 kilobytes were usable directly and the rest by playing some tricks with extended/expanded memory).

Current CPUs are not only much faster, but also a lot more complex than the 8MHz 80186 I wrote my first Pascal code on. Where we had a few scarce 16-bit registers, we now have 64-bit wide general purpose registers, plus a number of even wider SIMD ones, complete with their own opcodes to do operations in parallel. Where ye olde 16-bit CPU took 4 cycles for an add instruction (and many more for a mul or even div), current CPUs can sometimes do more than one instructions per cycle, thanks to pipelining. I also have two cores that can each execute two threads in my CPU, so I can do four things in parallel, on a low-end CPU. High-end desktop CPUs now have eight, ten or even more cores.

Even better, our computers now include GPUs that offer even broader parallelization opportunities (for those able to program them), so our code can do massive computations that would have been infeasible even on early 90’s supercomputers.

For many of us, that doesn’t matter much, because a good portion of their time they don’t really use a desktop or notebook, but a smaller, mobile personal device called Smartphone. Those now have CPUs and RAM that rival the contemporary notebook specs. Heat and power draw are the chief limiting factors. But I digress.

Turbo Pascal really was a wonderful language and a great development environment. Though I rarely used the debugger, I liked using the IDE a lot. Yet the other IDE features we take for granted (syntax highlighting, context-sensitive content assist, call hierarchy, refactorings, quick fixes to name a few) were missing. TP also had very little in the way of optimizations, making one go down to assembly level (which was available via asm { .. } syntax) for maximum performance.

Pascal didn’t offer those features because they wouldn’t have been usable with early ’90s CPUs and memory – either too slow due to scarcity of CPU cycles per second or even infeasible to implement due to scarcity of bits in RAM.

Contrast with a contemporary optimizing compiler, which – even if it would have run at all – would take hours, nay, days to compile a medium-size project (I have no hard numbers on how much optimizations benefit runtime, but a Rust program compiled with cargo build --release usually runs one or two orders of magnitude faster than the unoptimized version). The compilers can make use of the increased complexity to make our code run faster.

Not only are our compilers more complex, our langauges are, too (well, with the possible exception of Go, but that’s intentional). Even Java now has some form of lambdas and streams, so partial functional programming should now be considered mainstream (hint: it wasn’t in the days of LISP machines). If I choose a VM environment, I can get a garbage collector to deal with the problem of cleaning up memory after my program is done with it.

Many of our programming language use their powerful type systems to allow us to reuse code across different types while checking a good number of invariants at compile time. All without requiring us to write a proof of those invariants – it’s all implicit.

And if something goes amiss, the error messages we get are fabulous! Look into Elm or Rust for the best examples, but even gcc nowadays has some good examples.

Not only can we unit-test, we write documentation (well, the better of us do) that includes examples that are actually tested during our build (for those of us that use python or Rust, or Java with one of the javadoc extensions, I also wrote a doctest.lua at one point). With Rust documentation, the examples even include a playground link, so we can execute them online!

In our unit tests, we can make use of properties-based testing methods like quickcheck. We even invented techniques to test coupled instances (stubbing and/or mocking), although to be fair, many consider those a code smell.

When our early 90’s code crashed, we got strange patterns on the screen, or maybe the occasional corrupted file. Nowadays, we get DDoS botnets, crypto-trojans, banking scams and all sorts of nasty things. To paraphrase Neal Stephenson’s “Snow Crash”, this is no longer a safe place.

To counter this, we have built bespoke static code analysis tools. Code deemed security-relevant is also now heavily fuzz-tested, a technique that has only recently become feasible thanks to the explosion of available CPU cycles we can throw at the problem.