Add support for targeting iOS #65

fwcd · 2024-02-03T04:06:44Z

Since iOS is largely based on the same frameworks as macOS, the changes required to support iOS are minor. It does, however, lack some time-related methods, most notably CoreAudio/HostTime.h and the suggested approach appears to be using mach time instead: https://developer.apple.com/library/archive/qa/qa1643/_index.html

With this PR portmidi can now be built for iOS simply by changing the CMAKE_SYSTEM_NAME:

cmake -B build -DCMAKE_SYSTEM_NAME=iOS
cmake --build build

fwcd

Some notes on the implementation

fwcd · 2024-02-03T04:08:13Z

pm_mac/pmmacosxcm.c

+UInt64 current_host_time(void)
+{
+#if TARGET_OS_OSX
+    return AudioGetCurrentHostTime();
+#else
+    return mach_absolute_time();
+#endif
+}


The mach methods are available on macOS too, so we could probably drop the macOS-specific AudioGetCurrentHostTime, AudioGetHostClockFrequency etc. I've kept them around under a TARGET_OS_OSX conditional for now to avoid accidentally introducing a regression if there's a bug in the (albeit quite short) mach-based implementation for iOS. Since it introduces an additional, likely not very heavily used code path, I would also be open to removing the macOS branch again. Thoughts welcome.

Is there some reason to choose mach_absolute_time()?

It seems to be the officially recommended drop-in replacement: https://developer.apple.com/library/archive/qa/qa1643/_index.html

clock_gettime_nsec_np(CLOCK_UPTIME_RAW) seems to be a wrapper around mach_absolute_time:

https://github.com/apple-oss-distributions/Libc/blob/c5a3293354e22262702a3add5b2dfc9bb0b93b85/gen/clock_gettime.c#L117-L119

So it's probably a question of style which one to choose? mach_absolute_time seems pretty widely used, so I'm sure they won't deprecate that anytime soon.

Interesting, because documentation for mach_absolute_time says use clock_gettime_nsec_np(CLOCK_UPTIME_RAW) (!). But OK, if Apple can't decide, I'll go with your choice.

Thanks for the link, I must have missed that note. Interesting that they mention fingerprinting. But regardless, mach_absolute_time is the lower level API and almost all projects I've seen that use Core Audio/Core MIDI on iOS went with mach_absolute_time too.

porttime/ptmacosx_mach.c

rbdannenberg · 2024-02-10T20:13:32Z

pm_mac/pmmacosxcm.c

+#include <CoreAudio/HostTime.h>
+#else
+#include <mach/mach_time.h>
+#endif



This seems to be doing what PortTime was intended to do: provide a system-independent way to access time functions.
There are actually 2 porttime implementations for Mac: ptmacosx_cf.c based on core foundation calls and ptmacosx_mach.c based on Mach calls, but it looks like ptmacosx_mach.c still uses core foundation calls for time functions.
I'd much prefer to add system-independent functions in ptmacosx_mach.c and porttime.h than here for getting and translating nanos to portmidi (milliseconds).
A critical question though, is have you established that the nanos you get from Mach calls are the same nanos you get from CoreAudio? It's important because CoreMidi uses CoreAudio timestamps (undocumented but they match in tests) and it won't work to use just any nanosecond timer. I have no idea if mach gets the same nano values.

Hence #65 (comment):

Ideally, we should probably share these utility methods between porttime and pm_mac. The right place for that would likely be porttime, but we'd also create a leaky abstraction by exposing these implementation details, even if only to the (also platform-specific) pm_mac. Ideas welcome.

The question would be how pmmacosxcm.c would consume those functions (or whether there's a way to model these conversions in a fully platform-agnostic way, so they can be part of the main porttime interface. Haven't dug deep enough into porttime to understand the architectural decisions there though).

A critical question though, is have you established that the nanos you get from Mach calls are the same nanos you get from CoreAudio? It's important because CoreMidi uses CoreAudio timestamps (undocumented but they match in tests) and it won't work to use just any nanosecond timer. I have no idea if mach gets the same nano values.

That would indeed be interesting to check, I haven't done so yet. I would be surprised though if they would differ.

I think your additional functions to get host time and convert host time to micros and nanos is fine. At this point, only OSX and IOS would use them, and I think it's fine to make them Apple-only functions.

I'm working to integrate a lot of changes into PortMidi. Returning to this, the whole reason for incorporating CoreAudio time into PortMidi was that the absolute time values are important because they are used as Midi timestamps by MacOS (I don't think this is documented). I believe many PortMidi tests would work fine with bad timestamps, so it's important to look at some MIDI packets in iOS and compare their timestamps to time sources to see what time source is being used (or find it in documentation if it is stated anywhere). Similarly, output timing should be tested to see if absolute timestamps based on mach_absolute_time() or whatever can be used to produce correctly timed output. If you can verify the time source is correct (at least usable for timing and seems correct), then this would be a nice addition to PortMidi.

rbdannenberg · 2024-02-10T20:33:15Z

pm_mac/pmmacosxcm.c

+static UInt64 current_host_time(void);
+static UInt64 nanos_to_host_time(UInt64 nanos);
+static UInt64 host_time_to_nanos(UInt64 host_time);
+static UInt64 micros_per_host_tick(void);



portmidi may be overoptimized to use host_time when possible rather than always using nanos (which would give a simpler, smaller set of time functions), but OK, let's extend porttime. These additional functions can exist for apple only (since they are not called from other platforms). Names should be, e.g. Pt_Current_Host_Time -- now I kind of hate this style, but it was originally supposed to be consistent with PortAudio, and better to be consistent that not (or to break everything by changing naming conventions now).

Names should be, e.g. Pt_Current_Host_Time

Pt_CurrentHostTime? That would be consistent with the PortMidi naming style:

portmidi/pm_common/portmidi.h

Line 184 in 928520f

PMEXPORT int Pm_HasHostError(PortMidiStream * stream);

I've moved these functions to a new header, ptmacosx.h, since they are Apple-specific and so they don't leak into the porttime interface on other platforms:

portmidi/porttime/ptmacosx.h

Lines 9 to 15 in 15ddc24

UInt64 Pt_CurrentHostTime(void);

UInt64 Pt_NanosToHostTime(UInt64 nanos);

UInt64 Pt_HostTimeToNanos(UInt64 host_time);

UInt64 Pt_MicrosPerHostTick(void);

i.e. replace UInt32 -> uint32_t SInt32 -> int32_t UInt64 -> uint64_t SInt64 -> int64_t

The patches are based on PortMidi/portmidi#65

fwcd added 3 commits February 3, 2024 04:39

Replace EndianS32_BtoN with Core Foundation equivalent

a96f40b

Add drop-in replacements for host time methods in pm_mac

ba325d4

Add drop-in-replacements to porttime too

b14c729

fwcd commented Feb 3, 2024

View reviewed changes

This was referenced Feb 3, 2024

Port Mixxx to iOS fwcd/m1xxx#16

Open

Add arm64-ios-release CI mixxxdj/vcpkg#116

Draft

rbdannenberg reviewed Feb 10, 2024

View reviewed changes

fwcd added 3 commits February 10, 2024 22:35

Migrate host time conversion to Apple-specific porttime header

15ddc24

Remove no longer needed includes from pmmacosxcm

8772481

Replace UInt and SInt spellings with stdint.h

d1219cf

i.e. replace UInt32 -> uint32_t SInt32 -> int32_t UInt64 -> uint64_t SInt64 -> int64_t

fwcd added a commit to fwcd/vcpkg that referenced this pull request Feb 29, 2024

[portmidi] Use updated iOS patches based on latest master

3a30970

The patches are based on PortMidi/portmidi#65

fwcd mentioned this pull request Feb 29, 2024

[portmidi] Add overlay with iOS patches mixxxdj/vcpkg#128

Merged

fwcd requested a review from rbdannenberg March 3, 2024 20:42

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add support for targeting iOS #65

Add support for targeting iOS #65

fwcd commented Feb 3, 2024

fwcd left a comment

fwcd Feb 3, 2024

fwcd Feb 10, 2024

fwcd Feb 10, 2024

rbdannenberg Feb 10, 2024

fwcd Feb 10, 2024

rbdannenberg Feb 10, 2024

fwcd Feb 10, 2024

rbdannenberg Feb 10, 2024

rbdannenberg Mar 23, 2024

rbdannenberg Feb 10, 2024

fwcd Feb 10, 2024

fwcd Feb 10, 2024

	UInt64 Pt_CurrentHostTime(void);

	UInt64 Pt_NanosToHostTime(UInt64 nanos);

	UInt64 Pt_HostTimeToNanos(UInt64 host_time);

	UInt64 Pt_MicrosPerHostTick(void);

Add support for targeting iOS #65

Are you sure you want to change the base?

Add support for targeting iOS #65

Conversation

fwcd commented Feb 3, 2024

fwcd left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment