This PR brings all the changes from the new Metal APIs. During the
review pay special attention to the changes done in the Protocols in
order to add tvOS support.
The main problem we have had doing this PR is that tvOS was not done on
time before the NET branching, that left us with a lot of memebers that
were NOT added in tvOS that are abstract on dotnet, which has left use
in a pickle.
Lets use the following code as an example.
Code found before this commit:
```csharp
[Mac (11, 0), iOS (14, 0), NoTV]
[MacCatalyst (14, 0)]
#if NET
[Abstract]
#endif
[Export ("accelerationStructureCommandEncoder")]
IMTLAccelerationStructureCommandEncoder CreateAccelerationStructureCommandEncoder ();
```
A naive approach would be to add just the tvOS suppor as follows:
```csharp
[Mac (11, 0), iOS (14, 0), TV (16,0)]
[MacCatalyst (14, 0)]
#if NET
[Abstract]
#endif
[Export ("accelerationStructureCommandEncoder")]
IMTLAccelerationStructureCommandEncoder CreateAccelerationStructureCommandEncoder ();
```
The above change represents and API braking change on the donet tvOS dll
because it adds a new Abstrtact members, so this is no an acceptable
solution.
There is a second naive approach we can take which is as follows:
```csharp
[Mac (11, 0), iOS (14, 0), TV (16,0)]
[MacCatalyst (14, 0)]
#if NET &!TVOS
[Abstract]
#endif
[Export ("accelerationStructureCommandEncoder")]
IMTLAccelerationStructureCommandEncoder CreateAccelerationStructureCommandEncoder ();
```
Yet again, the naive approach has an issue with it. In this case, all
the extension methods that are generated for tvOS (something the
generator writes when methods are not abstract) will be decorated with
availability attributes for all the other platforms, which is incorrect
and will make developers life worse.
That leaves us with the following approach:
```csharp
#if NET
#if !TVOS
[Mac (11, 0), iOS (14, 0), NoTV, MacCatalyst (14, 0)]
[Abstract]
#else
[NoMac, NoiOS, TV (16,0), NoMacCatalyst]
#endif
#else
[Mac (11, 0), iOS (14, 0), TV (16,0), MacCatalyst (14, 0)]
#endif
[Export ("accelerationStructureCommandEncoder")]
IMTLAccelerationStructureCommandEncoder CreateAccelerationStructureCommandEncoder ();
```
With the above change, we do not add an abstract method in tvOS and we
only add the tvOS abailabity attribute to the extension methods, and use
NoiOS etc for all the other platforms.
The change had to be done to ALL methods that added tvOS support. The
good news are that our cecil tests and our introspection tests catch the
two naive approaces :)
---------
Co-authored-by: GitHub Actions Autoformatter <github-actions-autoformatter@xamarin.com>
Co-authored-by: Rolf Bjarne Kvinge <rolf@xamarin.com>
Co-authored-by: Haritha Mohan <harithamohan@microsoft.com>
Co-authored-by: Alex Soto <alex@alexsoto.me>
Change all null checking expressions to use 'is null' and 'is not null'
instead of '== null' and '!= null'.
This was mostly done with sed, so code can probably be improved in many
other ways with manual inspection, but that will come over time.
Also add code to the autoformat script to automatically fix these issues in the future.
Remove Runtime.Arch and ObjCRuntime.Arch from Mac Catalyst, because they don't
apply for a Mac Catalyst app (which is neither a simulator environment, nor a
device environment).
This means that code using these APIs will have to be re-evaluated to
determine what's the correct behavior for Mac Catalyst.
Also update our tests accordingly.
Fixes https://github.com/xamarin/xamarin-macios/issues/10312.
* Change all XAMCORE_4_0 conditions to NET conditions.
* Add numerous Release attributes that xtro started complaining about.
* Misc other minor API changes/updates.
* MTLCopyAllDevices returns a retained object, so we need to release it.
* MTLCopyAllDevicesWithObserver returns an observer (no need to provide one). This
means we can obsolete the 'ref NSObject' overload (the API doesn't make sense),
and instead add an 'out NSObject' overload.
* The returned observer from MTLCopyAllDevicesWithObserver is retained, so we must
release it.
* The returned array from MTLCopyAllDevicesWithObserver is a retained object, so
we need to release it.
* Simpify the supporting block code for the calls to MTLCopyAllDevicesWithObserver.
* Clean up the block we passed to MTLCopyAllDevicesWithObserver.
* [tests] Preserve all test fixtures.
This fixes the wildly diffetent number of tests when running xammac tests with and without linking.
Without linking:
> Tests run: 2446 Passed: 2321 Inconclusive: 9 Failed: 0 Ignored: 125
vs with linking:
> Tests run: 1885 Passed: 1802 Inconclusive: 4 Failed: 0 Ignored: 83
Now we run the same tests either with or without linking.
One test was updated to not fail when linking is enabled.
Also add a test to ensure all future test fixtures are preserved.
* Remove whitespace noise.
* Fix system version checks to work properly on Mac Catalyst (which uses the macOS
version as its system version).
* Add the framework-specific defines to the build for monotouch-test.csproj (using
the generated response files), this way we can use them in the tests.
* Sprinkle conditionals in numerous places - I tried using either framework-specific
or XAMCORE_3_0 whenever that made since (instead of using Mac Catalyst as a condition).
* Updated a few tests to use non-deprecated API (because the deprecated API often
doesn't exist in Mac Catalyst).
* Also a few minor API fixes to make any corresponding tests compile.
We've modified our NUnitLite to special-case the native types, so that they
can easily be compared with the standard types, but that doesn't work anymore
when switching to the official NUnit[Lite], so change all asserts to
explicitly convert whenever needed.
Apple's charts say indirect command buffers are available with MTLGpuFamilyCommon2.
That's not quite so, devices that support MTLGpuFamilyCommon2 may crash when
MTLDevice.CreateIndirectCommandBuffer is called.
So make the conditions for calling CreateIndirectCommandBuffer an intersection
of the previous condition for macOS (MTLFeatureSet.macOS_GPUFamily2_v1) + the
new condition (MTLGpuFamily.Common2) + Xcode 11+ (just to make things
simpler).
I've tested this on all our macOS bots, and it worked on all of them.
If it fails anywhere else (iOS devices), the next patch will remove the entire
test.
Hopefully fixes this assertion:
> 07:42:06.7701360 validateStrideTextureParameters:1512: failed assertion `Linear texture: bytesPerRow (64) must be aligned to 256 bytes'
which doesn't happen on my machine.
The structs MTLQuadTessellationFactorsHalf and
MTLTriangleTessellationFactorsHalf are wrong as per the header
definition:
```c
typedef struct {
/* NOTE: edgeTessellationFactor and insideTessellationFactor are interpreted as half (16-bit floats) */
uint16_t edgeTessellationFactor[4];
uint16_t insideTessellationFactor[2];
} MTLQuadTessellationFactorsHalf;
typedef struct {
/* NOTE: edgeTessellationFactor and insideTessellationFactor are interpreted as half (16-bit floats) */
uint16_t edgeTessellationFactor[3];
uint16_t insideTessellationFactor;
} MTLTriangleTessellationFactorsHalf;
```
We set the arrays size to be the correct one.
Fixes https://github.com/xamarin/xamarin-macios/issues/4611
* Add tests that ensure that the size of the managed structs is the same
as the native ones.
The values used have been checked against a native app compiled on 32
and 64 (iphone 5 and iphone X).
It turns out `MTLHeapDescriptor` does not exists on the simulator.
```Undefined symbols for architecture x86_64:
"_OBJC_CLASS_$_MTLHeapDescriptor", referenced from:
objc-class-ref in AppDelegate.o
ld: symbol(s) not found for architecture x86_64```
A `[DisableDefaultCtor]` was used (by mistake / lack of documentation)
and this was hiding the issue (on our bots).
However on device introspection was not happy:
iOSApiSelectorTest
[FAIL] Selector not found for Metal.MTLHeapDescriptor : cpuCacheMode
[FAIL] Selector not found for Metal.MTLHeapDescriptor : setCpuCacheMode:
[FAIL] Selector not found for Metal.MTLHeapDescriptor : size
[FAIL] Selector not found for Metal.MTLHeapDescriptor : setSize:
[FAIL] Selector not found for Metal.MTLHeapDescriptor : storageMode
[FAIL] Selector not found for Metal.MTLHeapDescriptor : setStorageMode:
[FAIL] iOSApiSelectorTest.ApiSelectorTest.InstanceMethods : 6 errors found in 18339 instance selector validated
Note that these are the properties, not the `init` that's mentioned here.
So first `init` is possible, on devices, from Xcode:
MTLHeapDescriptor *hd = [[MTLHeapDescriptor alloc] init];
NSLog (@"%@", [hd description]);
gives
<MTLHeapDescriptorInternal: 0x17401da50>
{
cpuCacheMode = MTLCPUCacheModeDefaultCache;
size = 0;
storageMode = MTLStorageModePrivate;
}
so we need to remove our `[DisableDefaultCtor]`.
That does not fix the selectors above... but note the *Internal type
returned, they are forwarded and that's generally something that
respondToSelector (that introspection uses) does not cover.
But, to be sure, we add unit tests showing all the properties are
working like expected :-)
Tested with iOS 7.1 on an iPad 3
- Dynamic registrar needs to ignore new CBManager base class;
- CGImageMetadataTag.Value returns a mutable array on iOS10;
- MPMediaItem properties needed 9.2 / 10.0 check;
- MSMessage not available before Xcode 8;
- Metal not available before Xcode 6;
- MetalPerformanceShaders not available before Xcode 7;