Hacker News Reader

Comment by AnimalMuppet | Hacker News Reader

I would respectfully submit that you completely missed Paul's point.

He's doing real-time audio processing. So, first, he's got performance issues. In that environment, it can be very helpful to re-use a buffer without having to allocate one (and without a garbage collector). For most functional languages, that's not a place where they shine.

But, second, he's (probably) got problems like building echo cancellers. The hard problem is, how do you determine whether something is an echo? Writing it in an FP language doesn't help much with that, because the hard part is the definition (and tuning) of the algorithm, not the implementation.

View on HN

> He's doing real-time audio processing. So, first, he's got performance issues.

FFTW, long the fastest FFT software available, is a program written in OCaml that generates C. You need to think more outside the box!

Another example: Copilot is a Haskell EDSL that generates a hard real-time C program for avionics.

You can write programs functionally and get your imperative performance too.

>He's doing real-time audio processing. So, first, he's got performance issues. In that environment, it can be very helpful to re-use a buffer without having to allocate one (and without a garbage collector). For most functional languages, that's not a place where they shine.

It can be via streaming. The functional portion of the program does all the map and reduce operations on streams. Buffer allocation (and reuse) is automatically handled by the compiler. It's very similar to the way NodeJS hides the event loop. I'm not familiar with audio stuff but I'm sure there's libraries out there that allow you to do extremely efficient audio processing with high level language primitives.

>But, second, he's (probably) got problems like building echo cancellers. The hard problem is, how do you determine whether something is an echo? Writing it in an FP language doesn't help much with that, because the hard part is the definition (and tuning) of the algorithm, not the implementation.

Neural networks can be employed here to "recognize" echos. NNs are basically functional primitives with deterministic inputs and outputs. FP is remarkably appropriate here given that audio is basically a function.

> I'm not familiar with audio stuff but I'm sure there's libraries out there that allow you to do extremely efficient audio processing with high level language primitives.

I am extremely familiar with audio stuff, and I can assure you that there are no libraries out there that solve the problems you face when writing a digital audio workstation.

Doing non-realtime processing of file-based data? Of course! Oodles of stuff for that, and you can write it in whatever language you want.

Realtime processing of 240 tracks of audio feeding 64 channels of physical output, with hundreds of plugins? Sorry, but no, you can't do this "with a library" or in "a high level language". This, however, is what DAWs do.

Well, there are certainly use cases where purely functional languages (at least the ones that currently exist) aren’t the tool of choice… But in the grand scheme of things, like you yourself pointed out, these types of problems are not what the vast majority of engineers are working on.

I had assumed that initially you were referring to the difficulty of the “domain problems” when I wrote my comment.

I'm a Haskell fanatic, even wrote it professionally recently. I have also done a fair amount of small-scale audio programming, including trying to write a basic "soft" real-time-keeping app in Haskell, and it's not a great fit. I would probably choose Rust for anything that had to interface with hardware or tell time at this point, and that's largely because it would probably be easier for me than C/C++, which seems like the obvious choice based on ecosystem and prior art.

I think these kinds of comments made by FP advocates hurt FP advocacy more than they help, by implying that FP advocates can't see the forest for the trees. We are not in a world where the answer to ever programming problem is "throw FP at it!" We will never be in that world, no matter how awesome FP is, and I'm a huge fanboy.

Go take a look at the Ardour (http://ardour.org/credits) codebase and ask yourself in all seriousness how you could approach a project like that, what the tradeoffs would be, etc. and then ask yourself how serious any of your suggestions sound to someone like who you're responding to.

I've written haskell that interfaces with hardware. I've also done real time video processing with C++ which is magnitudes of more data then audio.

This by the way is 15 color cameras of extremely high resolution video so it's not trivial. At this level of data you can't expect real time performance without a GPU, so I go straight into CUDA for massive parallel processing performance.

You can think of each pixel in a video as 3 channels of sound, multiply that with HD resolutions and you have thousands of channels of sound flowing through the system at the same time in REAL TIME. 15 * 1080p * 3. Our pipeline also involves real time ML model processing and object tracking all written in C++/cuda.

This of course is across hundreds of similar edge devices all which feed relevant data into the cloud for further processing.

I am not an FP extremist. I work in systems programming which is pretty much the opposite of FP. I am simply stating a specific point here in that I think audio programming for FP can work. I respect the authors comment and I left it at that, even though I'm positive it's possible to write an FP language/framework that handles this specific use case of writing daws. It's basically a massively parallel streaming framework not so dissimilar to what I work on.

The actual processing is easily a functional issue, and the low level memory details can definitely be abstracted away. In fact, it's LARGELY what a daw is when you think people who write plugins.

What really pissed me off about your comment is the superior attitude. Like the "learn how to respect the experts NEWBIE". Seriously I hate shit like that. Who the hell are you?

Yeah, I'm not into expertism either. Even if I'm the expert (or maybe especially if I'm the expert).

Anyway ..

> In fact, it's LARGELY what a daw is when you think people who write plugins.

This is the problem though. Yes, that's how people who write plugins think about the host. But it's nothing really like the actual internals.

Comparing video and audio requirements is rarely fair and balanced. I tend to only be interested in realtime processing, not offline. Video involves far more data processing, but also missing deadlines is rarely noticeable by a human (unless you really miss a bunch of them). A single missed audio deadline (even 1 sample) is audible to any human. That's just a reflection of the different ways our eyes and ears work. So it's a bit of complex thing to try to compare the loads.

It's relatively easy to design and implement a parallel streaming framework, especially one without GUI interaction and realtime requirements. Most of the hard stuff comes from the user interaction side and the requirements it imposes on the realtime side. An engine that just takes a set of processing instructions and applies them as fast as possible to as much data as possible is significantly easier in a number of important ways.

Reply on news.ycombinator.com