Need to devise a way to get the platform specific headers from IKVM.Java. This requires new IKVM.Net.SDK support, and to split up IKVM.Java into platform specific versions. Which introduces our ref-assembly problem. Needs to be solved.

…ded since they don't exactly match the headers generated by IKVM.Java. Some important utiltiies are available, however.

For instance, Net.socket0, for creating sockets, can now be used.
FileDispatcherImpl will be an important one to get building, but it doesn't build just yet.
ServerSocketchannelImpl does, SocketDispatcher, SocketChannelImpl, LinuxWatchService.c builds. UnixNativeDispatch
Trying to figure out how to materialize a .NET framework socket out of a handle.

…ends can be used to set the FD from the call to socket0(). Core makes this easy, as it will take an existing handle. Framework needs to invoke the SocketInformation ctor, with information provided by a duplicate socket. Mono is unimplemented. But, if on Unix, we can probably pack some stuff into SocketInformation. Leave that up to the future.

Fix a few JNI calls.

Update Clang version to release.

…s technically still Windows machines with Framework out there with ARM32.... but, I think the chances of them using IKVM are zero.

…d not available in musl.

Always use LoadLibraryEx on Windows. NativeLibrary seems to have some uncertainty about when it tries to do path searches and stuff.

…plicated things needlessly between Framework/Core.

Rename libjava to libikvm. I hate this. But, .NET Core produces a java.dll file, which is loaded into the process using LoadLibrary, even though it's not native. And screws up dependency resolution for net.dll and nio.dll. dotnet/runtime#90839
Will research the single file thing. Need to figure out how to apply it to everything in files to publish.

Just call Load a second time. Same result.
libnio needed a blank JNI_OnLoad. Not sure on this one. Doesn't exist in OpenJDK.
Wrap dlopen() in libjvm. P/Invoke wasn't capable of resolving symbol properly on Linux, with libdl being deprecated in Ubuntu > 22.

Enable native initIDs method.
Add a datagram test to the Java lang tests.
Add a test to check that adhoc send fails when connected.
Add some tests for bind with null values.

Add a test for joining IPv6 multicast groups.
Skip unaligned long compare and swap tests on ARM[64]. Crashes with SIGBUS or similar. ARM doesn't support unaligned atomics.
Add missing socket functions ot jvm.cpp.

…de the same functionality.

Fix test for osx-arm64.

…rgetName).dylib, which should allow linking libs to resolve from relative paths? Maybe.

Use native FileDispatcher. We had to make a local copy of FileDispatcher.c in order to use lock0, since the signatures differ between solaris/windows versions. This is a minor fork, and gets locking working... maybe on OS X too?
Remove lock0/release0 implementation from C#, use C version.

…or non-java related IKVM native functionality apparently.

…ndencies on OpenJDK related things. dl functions moved here. This is to prevent the bootstrapping issue with being forced to load libjvm itself using DllImport: the path of libjvm has to follow Java library load rules. libikvm follows .NET loading rules.

…the native OpenJDK libraries. Unified the logic for building .runtime images into IKVM.Image/IKVM.Image.runtime.targets. We reuse IkvmImageItem processing, but still pack.

This puts the native libraries exclusively in the image. Which will let us eliminate our search path processing for runtimes/native/*. The sun boot path can now be set to the image path, while the user lib path can be set to runtimes/ or the application dir. Need to do that still.

NativeLibrary once again using NativeLibrary where available. This makes more sense now that we ensure full resolution before invoking.
NativeLibrary has to bootstrap ikvm.dll on Windows without Mono by calling LoadLibrary directly, since the .NET SDK has the potential for putting the DLL files under runtimes/rid even under Framework. But sometimes it doesn't.
boot path replaced with Image path.
user path replaced with .NET paths.
DestinationSubDirectory still needed for files without extensions.
We preload jvm.dll through System.loadLibrary, so that it resolves out of JVM system paths. But, then we also load it by hand in JNI to get the JNI trampoline functions.
Ensure we finalize all the manually loaded native libraries.

…uplicates.

Add CopyToOutput on a number of PublishProjectReferences.
Remove conditional IncludeProjectReferences. Should not be required.

Allow loader a first chance on PreLoad.

Make Tools tests project tolerate duplicate items.
Try to load existing libjvm by checking ClassLoader.

…e, which prevents accessing non initialized fields.

MSBuild.Tasks.Tests ignore duplicates.

Fix native lib name unmapping.

Call System.loadLibrary in JniTests. Unsure about this one.
Native.cs no longer needed in tests. Should use default loading logic.

…C_PATH to @loader_path, and names itself @rpath/libjvm.dylib.

We preload libjvm using NativeLibrary (or libikvm), against a full path.

… Regardless, DllImport couldn't find it, even after a successful preload. Linux behaves same as OSX in that regard. Just use out NativeLibrary to pull out symbols.

…rt for initializating based on file based FD or handle (using libikvm to detect if a handle is a file or socket). Add initIDs back to our local fork. We might be pretty close to discarding this fork and adding fields/properties through map.xml.

Replace RandomAccessFile.open0 with native code. Maybe this will impact OS X locking?
Ensure libjava (libiava) is loaded before any JNI call. This isn't loaded by any OpenJDK code that I can find. Seems to be built-in/internal. But, we need it to be loaded to bind JNI methods, and we don't wnat to edit OpenJDK Java files.
isnanf() doesn't exist in MUSL. Manually define to isnan().
Move LibIkvm to separate file.

…Loader. In here we prepopulate nativeLibraries with libjava (libiava). We do that on the boostrap class loader, instead of just in time, to avoid recursion with string comparison stuff that also relies on libjava.

libjava itself needed to be loaded through libjvm (it is a JNI library).

… implement it using mJNI, intercept VFS calls, and then forward to native open0 implementation using JNIFrame by hand. Works fine.

Add Flush wherever FileDispatcherImpl calls write, since that seems to actually matter.

…dle to the loaded library, and handles proper finalization. Ensures finalizers run after everything else.

Might have to do something similar for libJVM's LoadLibrary call.

Throw IOException properly on unauthorized exception.

Importer tests should have runtime libs.
Update Core, which fixes PublishProjectReferences to not ignore nested directories.

Dispose in proper order in tests.