WriteBytes does not return how many bytes were not written

[danielgruber]

Hello,
At the moment we encountered that writeBytes does not return how many bytes were written. For a CTS / RTS flow, this is absolutly necessary. Is there a reason why this is not needed / or how we use this library correctly with hardware CTS / RTS without losing data in the middle?

Thanks a lot!

Best

Daniel

[tresf]

Hi,

I don't believe this is possible currently. To support this without breaking backwards compatibility, a new function (e.g. write(bytes[])) would have to be written and the old boolean functions would have to be update to check the return length.

This would require Windows and Unix C++ code to be updated to return jint and the current functions would be wrapped to still return a boolean.

Disclaimer, this library is mostly in maintenance mode, so any features would have to be authored by the vested/interested parties.

Relevant functions:

Unix:

jssc/src/main/cpp/_nix_based/jssc.cpp

Lines 528 to 535 in 6886b44

	JNIEXPORT jboolean JNICALL Java_jssc_SerialNativeInterface_writeBytes
	(JNIEnv *env, jobject, jlong portHandle, jbyteArray buffer){
	jbyte* jBuffer = env->GetByteArrayElements(buffer, JNI_FALSE);
	jint bufferSize = env->GetArrayLength(buffer);
	jint result = write(portHandle, jBuffer, (size_t)bufferSize);
	env->ReleaseByteArrayElements(buffer, jBuffer, 0);
	return result == bufferSize ? JNI_TRUE : JNI_FALSE;
	}

Windows:

jssc/src/main/cpp/windows/jssc.cpp

Lines 234 to 257 in 6886b44

	JNIEXPORT jboolean JNICALL Java_jssc_SerialNativeInterface_writeBytes
	(JNIEnv *env, jobject, jlong portHandle, jbyteArray buffer){
	HANDLE hComm = (HANDLE)portHandle;
	DWORD lpNumberOfBytesTransferred;
	DWORD lpNumberOfBytesWritten;
	OVERLAPPED *overlapped = new OVERLAPPED();
	jboolean returnValue = JNI_FALSE;
	jbyte* jBuffer = env->GetByteArrayElements(buffer, JNI_FALSE);
	overlapped->hEvent = CreateEventA(NULL, true, false, NULL);
	if(WriteFile(hComm, jBuffer, (DWORD)env->GetArrayLength(buffer), &lpNumberOfBytesWritten, overlapped)){
	returnValue = JNI_TRUE;
	}
	else if(GetLastError() == ERROR_IO_PENDING){
	if(WaitForSingleObject(overlapped->hEvent, INFINITE) == WAIT_OBJECT_0){
	if(GetOverlappedResult(hComm, overlapped, &lpNumberOfBytesTransferred, false)){
	returnValue = JNI_TRUE;
	}
	}
	}
	env->ReleaseByteArrayElements(buffer, jBuffer, 0);
	CloseHandle(overlapped->hEvent);
	delete overlapped;
	return returnValue;
	}

[danielgruber]

Hi @tresf,

Thanks for your update! I’m happy to implement the changes, still I would first want to know if there was a reason behind it not to implement it like this from the beginning?
So is this at all possible with the C stack being used? Or do I have to try?
Thanks!
Best
Daniel

[tresf]

I would first want to know if there was a reason behind it not to implement it like this from the beginning?

Hi, this project is a fork of the original from https://github.com/scream3r/java-simple-serial-connector, so the history behind this decision likely lies there.

So is this at all possible with the C stack being used? Or do I have to try?

I think the Unix code makes it relatively easy, but I'm not sure about the win32 API. Perhaps the answer lies in GetOverlappedResult returning a zero? I'm not very familiar with hardware flow control and only partially familiar with the C++ portions of the codebase. We do have some other volunteers here as well as a Slack channel if interested. @hiddenalpha has been very helpful with the C++ portions as well.

[hiddenalpha]

Looks related to issue #96