Skip to main content

The same file, the same mistake, found by CI a second time

11. The same file, the same mistake, found by CI a second time​

What happened: after patches 0001–0005 shipped and were believed complete, the next CI run failed with:

error: call to undeclared function 'AudioConvertHostTimeToNanos'
error: call to undeclared function 'AudioGetCurrentHostTime'

Root cause: ao_coreaudio_utils.c's ca_get_latency() function had its own, separate #if HAVE_COREAUDIO || HAVE_AVFOUNDATION guard, calling two functions declared in <CoreAudio/HostTime.h> — a header that patch 0002 had already narrowed this file's own #include of to HAVE_COREAUDIO only. We had fixed the include but missed that this function's own guard condition needed the identical narrowing, since it called functions from that now-conditionally-included header.

Fix: patch 0006 narrows ca_get_latency's guard to HAVE_COREAUDIO only, so an avfoundation-only build correctly falls into the function's existing #else branch (a mach_absolute_time-based equivalent that's already there, already correct, and needs no CoreAudio API at all — it just wasn't being selected due to the too-broad guard).

What we did afterward to reduce the chance of a third repeat: re-scanned every remaining HAVE_COREAUDIO || HAVE_AVFOUNDATION guard across all four touched files for any other hidden AudioDeviceID/AudioStreamID/HostTime-API reference, rather than stopping at the one instance the compiler happened to report first.

Lesson, stated directly in patches/mpv/README.md now: when a file has multiple guarded sections sharing the same macro condition, fixing the one instance a compiler error points at doesn't mean the others are safe — the same file had two separate instances of essentially the same mistake (a guard covering more than its body actually needs), and only a full-file scan catches the second one before CI does.