Logic Pro's AU "validation" nonsense

[Super Piano Hater 64]

Well - to add to the frustration - I've asked some new beta testers to try the plugin in Logic on their M1 machines and - like here - it works! Validates and runs. No issues.

So now the bigger question - why does it PASS validation on some M1's but FAIL on others?

This strongly suggests undefined behavior. If you don't understand what kind of threat that poses, you should learn immediately before writing another line of code.

[Fender19]

If you don't understand what kind of threat that poses, you should learn immediately before writing another line of code.

That’s a little dramatic. I’m not writing code for defibrillators here. Logic itself has had magnitudes more issues than my plugins have ever had yet that hasn’t stopped Apple from”writing another line of code” and selling their latest whatever.

Maybe Apple should use some of that Developer Account fee I pay them every year to fix auval so it’s checks are reliable - and publish some info on what it’s error codes mean.

[Super Piano Hater 64]

That’s a little dramatic. I’m not writing code for defibrillators here.

Me either. This isn't about some cyberheist scenario where Tom Cruise crawls through your ceiling. It's not even about crashes. All I'm talking about is whether your code does what it's "obviously" supposed to do, or some other seemingly random thing you definitely didn't intend.

Other replies in this thread have offered you some thoughtful advice about how to try to solve your problem (EDIT: Removed the rest of this comment. I was being a jerk. That wasn't helpful and I'm sorry.)

[Markus Krause]

you should learn immediately before writing another line of code.

What an over-aggressive and toxic comment...
The issue is not necessarily the flaw fo Fender19.
Please do note that he is facing many problems that he is not resonsible for:

1) Broken development tools. It is not possible to debug auval since Catalina. Apple is resonsibe for this.
Reference: https://forums.steinberg.net/t/unable-t ... val/693455

2) A bug in AudioComponentRegistrar. Apple is responsible for this and did not fix things that are broken since 7 years (High Sierra). It affects hundrets of developers and tenthounsands (!) of end-users that had to reboot their Macs unnecessarily.
Reference: viewtopic.php?t=531117&hilit=killall

3) Missing/incomplete documentation. Apple (and Steinberg if he uses the VST SDK) is resonsible for this.

4) Possible bugs in the SDK (in this case Steinberg or the JUCE team would be responsible).

[Markus Krause]

I've also been facing the same (?) problem as Fender19 two years ago. In my case I had to change the code for the setBusAttangeMents function to make my plugin pass Auval.

Old code that did not work (VST 3.7 SDK). The plugin did not pass auval without a reasonable error code:

tresult PLUGIN_API T2Audio::setBusArrangements(SpeakerArrangement* inputs, int32 numIns,
SpeakerArrangement* outputs, int32 numOuts)
{
if (numIns == 0 && numOuts == 1 && outputs[0] == SpeakerArr::kStereo) return SingleComponentEffect::setBusArrangements(inputs, numIns, outputs, numOuts);

return kResultFalse;
}

Fixed code:

tresult PLUGIN_API T2Audio::setBusArrangements(SpeakerArrangement* inputs, int32 numIns,
SpeakerArrangement* outputs, int32 numOuts)
{
return SingleComponentEffect::setBusArrangements(inputs, numIns, outputs, numOuts);
}

[mystran]

1) Broken development tools. It is not possible to debug auval since Catalina. Apple is resonsibe for this.
Reference: https://forums.steinberg.net/t/unable-t ... val/693455

This one is arguably complete nonsense for a tool that serves no purpose other than to help devs.

Code: Select all

(lldb) target create "/usr/bin/auval"
Current executable set to '/usr/bin/auval' (arm64e).
(lldb) run
error: process exited with status -1 (attach failed (Not allowed to attach to process.  Look in the console messages (Console.app), near the debugserver entries, when the attach failed.  The subsystem that denied the attach permission will likely have logged an informative message about why it was denied.))

[Super Piano Hater 64]

Sorry folks. I was definitely way over the line. I've had my own frustrations with Apple and I know first-hand that expecting them to fix anything is like trying to lift up the sun. Bug reports just disappear. Too often, suffering and working around their mistakes is the only option, if there's any option at all.

I've also had loads of arguments with people who were flippant about UB, but that's no excuse to jump down someone's throat and I shouldn't have talked like that.

[Zaphod (giancarlo)]

nevermind

[machinesworking]

The issue is known to Apple. I had a conversation with one of their developers a few years ago and he said they won't fix it as AUv2 is "deprecated".

As a fly on the wall here, this makes me think we're looking at some sort of push from Apple sometime in the near future to make AUv3 the standard like Steinberg did with VST3. The end result being almost no difference at all between iOS and Mac OS. Makes me again wish CLAP had come around 15 odd years ago, by now all DAWs (barring Pro Tools and Logic) would support it.

[kerfuffle]

With Logic being on iPad now, Apple might hope this encourages developers to use AUv3 on Mac too, since that makes projects portable between devices. We'll probably get some news about this at the upcoming WWDC.

[Zaphod (giancarlo)]

With Logic being on iPad now, Apple might hope this encourages developers to use AUv3 on Mac too, since that makes projects portable between devices. We'll probably get some news about this at the upcoming WWDC.

Side question: a problem I'm not entirely sure about because I haven't had time to investigate, it seems that on iOS you can't run libraries dynamically (runtime loading, i.e., not statically linked like a framework). We need this feature. Does anyone have clear ideas or solutions on this topic? For consistency with the Windows platform, we absolutely need to load signed dylibs at runtime variably (we can also notarize, that's not an issue). We can't statically link frameworks or executables. Is this a problem?

[Fender19]

FYI - we have since found and fixed the problem reported in this thread. Turns out it had NOTHING to do with the error reported by auval on M1 ("Problem with initial Channel layout state"). It was caused by a call to clear an FIR filter that - for some reason - was instantiated at a different time on M1/Arm Mac than on x86 Mac. That FIR filter is internal to our DSP code and does not interface with channel I/O. The "bug" did not cause any operational issues but, as some suggested here, could have been a latent failure in wait.

Auval on x86 Mac did not report ANY issues with the plugin but Tracktion's "Pluginval" tool did (although without much direction on possible cause). We sent an email commending and thanking Tracktion for their superior error detection as it at least gave us a "red flag". (We initially thought Pluginval was giving us bad info but it was actually auval giving us bad info).

We worked backwards - commenting out lines of code - until we found the bad spot, again checking with Tracktion's Pluginval tool. We moved that line of code elsewhere and problem solved. It now validates everywhere including Logic on M1/Arm.

Thanks to everyone here for your inputs and suggestions. Debugging this stuff - especially on Macs - can be a real bear.

[Fender19]

2) A bug in AudioComponentRegistrar. Apple is responsible for this and did not fix things that are broken since 7 years (High Sierra). It affects hundrets of developers and tenthounsands (!) of end-users that had to reboot their Macs unnecessarily.
Reference: viewtopic.php?t=531117&hilit=killall

It appears Apple has monkeyed with this once again. Plugins we published just a few months ago immediately showed up in Logic after being installed. The plugin being discussed in this thread - that we just compiled and released a week ago - requires REBOOT before it shows up in Logic.

WE did not change anything. This plugin was cloned from a previous plugin and used all the same compiler settings, signing and notarization settings, etc. So, it appears Apple changed/broke something again in AU registration (via recent updates to Xcode or OSs).

[hugoderwolf]

FYI - we have since found and fixed the problem reported in this thread. Turns out it had NOTHING to do with the error reported by auval on M1 ("Problem with initial Channel layout state"). It was caused by a call to clear an FIR filter that - for some reason - was instantiated at a different time on M1/Arm Mac than on x86 Mac. That FIR filter is internal to our DSP code and does not interface with channel I/O. The "bug" did not cause any operational issues but, as some suggested here, could have been a latent failure in wait.

You're describing the very problem with undefined behavior: making assumptions not guaranteed by the standard about the behavior of the program (e.g. initialization order), which can cause all kinds of (often unrelated and hard to debug) failures. You should consider it a fully qualified bug, even when it doesn't cause operational issues. These things tend to fall on your feet unexpectedly (unrelated code change, compiler change, change of API behavior in line with specs).

[Fender19]

making assumptions not guaranteed by the standard

Nothing is guaranteed by the standards. DAWs all have their own interpretation/implementation of the various plugin specs, sometimes even within brands. For example, Steinberg's Wavelab handles VST plugins differently than Steinberg's Cubase and Nuendo; Cakewalk and FL Studio load/display the plugin UI and DSP in different order than most other DAWs; Reaper is quite robust and will run plugins that crash other DAWs; Pro Tools is it's own "thing" and so is Logic Pro/Apple AU.

Out of all the formats, the plugin issue that inspired this thread ONLY occurred in Logic Pro and ONLY on M1. It was literally one "bad Apple" out of hundreds of other good combinations. When a situation like that arises, yes, we jump to an assumption as to the source of the problem. Did we ultimately fix a global issue? Maybe. Did any other DAW care? No.