but it's limited and unstable

I thought the problem was that you didn't have working audio drivers

problem is the kernel side of things

the audio stack is complex and very picky

which would be easily solved by porting UniAud from OS/2 (it's a port of Linux ALSA and it's drivers to OS/2)

hmm

the effort needed to convert UniAud into an NT Driver would probably outweight the effort of fixing the existing audio stack

if you know both how audio systems work, and how kernels work

I don't really think so... OS/2's audio API is quite similar to the Windows API model

I understand, to a certain degree, how kernels work

audio systems, not so much, except from the userland side

well my knowledge is rudimentary so I can't really give you a proper description

About 10 years ago, I worked with a friend to write a rather crappy multitasking kernel for PCs

but think of it as: an audio device has a series of "pins", which can be inputs or outputs

and they support a number of configurations (sample formats, sample rates, etc)

heh

and then have some buffers around

sounds like the crap I put up with when writing a crappy ALSA output "driver" for PsyMP3 years ago... which I scrapped for having SDL just handle everything

then an audio stream is composed of a bunch of buffers, a pin, and a configuration

the audio stack has to manage who owns which pins, has to maintain the buffers full, etc

SDL's audio model involves a buffer and a configuration

while also tracking the hardware state, timings and such

the pin bit is handled by the library

also, it's callback-based, so the timing is also handled by the library... was a hell of a lot simpler than trying to use ALSA directly

then on top of that, add the mixer component, which has to keep track of virtual "pins", and which configuration they are currently using, so that it can do the necessary conversions, before adding the data together into one output

i've also played with OSS on Linux and FreeBSD, and Sun AudioIO on NetBSD and Solaris

yep, in windows

there's more than one api

I know

I played with Windows waveOut stuff a little

like 8 years ago

ah high level, there's waveOut, which uses synchronization objects (sortof a semaphore) to signal when a buffer is empty and ready to be filled in

thought it was brain-damaged

at low-level, I think both KS and WASAPI work with callbacks instead

hmm

DSound is mostly a library that interfaces with kernel streams

(it has a kernel-side server)

and XAudio2 is purely high-level:

what little I saw of kernel streams looked more like ALSA/OSS

in NT5 it wraps DSound, and in NT6 is uses WASAPI

i.e. less brain-damaged than waveOut

yep

but at thesame time

you get brain damage when trying to configure the stream

since all the low-level configuration details have to be negotiated by hand

This is true

which is why WASAPI is so awesome and reactosshould implement it even if it's not NT6 ;P

This is why, for PsyMP3, I simply decided to use SDL's audio API

even on Linux, you have to negotiate by hand

especially if you're running on an old Pentium 2 box with an ancient Sound Blaster that can only do 22.050kHz sampling with 8-bit samples

SDL lets you just declare what you're producing and it'll resample as necessary

(if you're wondering what's with the ancient P2 box, well, I had a bunch of old machines around when I was writing PsyMP3 and was using as many of them as possible for testing because I was, and still am, a shit programmer)

yepseems like ALSA is on a similar level of complexity

https://github.com/PCSX2/pcsx2/blob/master/3rdparty/portaudio/src/hostapi/alsa/pa_linux_alsa.c — 4000loc

https://github.com/PCSX2/pcsx2/blob/master/3rdparty/portaudio/src/hostapi/wasapi/pa_win_wasapi.c — 4300loc

https://github.com/PCSX2/pcsx2/blob/master/3rdparty/portaudio/src/hostapi/wdmks/pa_win_wdmks.c — 5800loc

My friend Mattis helped me quite a bit

KS is definitely WAY more complex ;P

also

LOC count is meaningless

this is lines excluding comments, required to perform the same functions

so?

it's not quite pointless

for determing actual code complexity, it is.

if you need over 1500 lines more in order to maintain and configure a stream

sure but I can't measure it in "wtf/minute" ;P

although, even if the complexity wasn't higher, you can't deny it's messier ;P

sure but I can't measure it in "wtf/minute" ;P

XKCD? :D

no, I don't think this originated in xkcd, even if it seems to be attributed to him

https://blog.codinghorror.com/content/images/uploads/2009/02/6a0120a85dcdae970b012877707a45970c-pi.png

http://i.imgur.com/J1svNp7.jpg

oh, yeah, that

:D

http://www.osnews.com/story/19266/WTFs_m

this seems to be the source

you should see the Bézier curve code I wrote when I was completely fucked up on a lot of drugs

It does work, and works perfectly

but I have no idea how it works

I wrote a quadratic interpolation curve before i knew it had a name ;P

in my VB6 days