I got this error from Xcode when attempting to compile this rather simple floating point constant: Float((1.0 / 255.0) * 0.5 - (1.0e-7 * (1.0 / 255.0) * 0.5)) It's a clipping constant for 8 bit per pixel image data in the floating point space.

Experiencing this issue with Apple's own Support Forum from iPadOS 15 beta. Log in with fingerprint id, type a message, immediately get a session timeout asking me to "recover" my message, which asks me to log in again, but discards the message I typed. And attempting to type a new message and submit it results in the same session timeout.

Same for me on 2020 M1 Mac mini and attempting to run ai-benchmark. Defeats the whole point of even benchmarking if the library is busted. Fails on test 17/19, Pixel-RNN, due to the alignment issue with the allocator. The tensorflow-opencl software had a similar issue back in 2017 that had to be solved by enforcing memory alignment in the allocator(s).

Tested this on 12.4, still having troubles. Looks like two games I tested have broken controller support. Both of them are Unity games using the InControl framework. I tried three controllers with each: An Xbox Series X controller, a DualShock 4 controller, and a Stadia Pro controller connected via USB, which appears as a generic controller with vaguely Xinput like features. Games tested: The World Next Door, Turnip Boy Commits Tax Evasion. Unrelated, Barony for macOS seems to require me to double-click for it to register a single left click now. Cool, games get progressively more broken.

All of the above. Rebooting my primary machine to test a beta is not feasible, I am regularly in video calls throughout the day, and I would have to disrupt them every time I reboot. The beta is still not stable enough for me to consider running it for regular use. I currently have no less than 5 outstanding feedback issues filed, none of which have been answered yet. One of them is the ability to actually respond to Feedback questions from Apple, which is currently broken. The "This is fixed" / "This is not fixed" bubble prompts do not appear in the dialog, so it is impossible to send a response to Apple.

I already solved my problem. https://github.com/losnoco/Cog/blob/main/Audio/Output/OutputAVFoundation.m AVSampleBufferAudioRenderer supports this perfectly on macOS Monterey (12.0). Plus, the API, even without spatial audio support, is supported all the way down to High Sierra (10.13). You can see where, in my implementation, that I use a function gated by if(@available(macOS 12.0, *)) to enable spatialization. I also use the timed event reporting observer to keep track of the playback position, since currentTime requires Mojave (10.14).

This was fixed eventually, and is at least working properly by 12.4.

I already solved this. It was meant to support either single short replies, or continuous data streaming at a regular playback rate, for playing streaming audio. It turned out to be rather a poor choice for audio streaming in the end. I ended up borrowing a libcurl solution from the DeaDBeeF audio player project. I also added an option to disable peer verification, in case users must use stream servers with either self-signed or expired certificates. It uses the system's libcurl, so it fits the exemption for not bundling any encryption solutions.

I have a privacy policy for my account now. But it only really applies to my software, as a Sticker Pack can't really be phoning home with telemetry or crash reports.

It must have been a fluke of a particular OS install. This is a long stale and solved issue.

Incidentally, it's also not possible to maintain a beta install on a primary development machine if you develop for the App Store, as the beta does not support stable Xcode, and the beta, like all betas in the past, cannot be used to publish to the App Store, unless you use Xcode Cloud to build using a stable version. And I find I cannot use Xcode Cloud, because it messes with my app's Info.plist bundle version in unpredictable ways. I cannot get it to build consistent versioning. I use a template Info.plist and generate the version number from the Git repository at build time, and Xcode Cloud ignores this and stamps the Info.plist with its own decided version string, even one which is incompatible with the App Store, as it contains letters. (It uses the first five of the Git hash as the version string, ignoring my chosen version template of commit number since a specific tag.)

Okay, I figured out what was wrong. Xcode Cloud build system isn't fetching the tags in my repository when pulling it. So I need to add a git fetch --tags to the post clone script.

Okay, nope, it needs to be in the pre xcodebuild script, since the submodules aren't fetched yet.

Oh, drat. I have to --unshallow the repository to get the commit counts. I hope that works.

For some reason, a function that has barely any local variables incurs a local stack frame difference like this:       sp = 0x000000016ec7eb40       sp = 0x000000016ebfea40