Use poll in readBytes to fix segfaults with high fd numbers

[hiddenalpha]

• Replace 'select' by 'poll' in readBytes to fix segfaults when fd >= 1024
• Log a few error cases which we concealed in previous code.
• Fix some concurrency issues.

HINT: I did not add recompiled libs as I only could provide/test the linux lib
anyway. Not sure how this gets handled in this git repo. Let me know.

Edit: (@tresf)

• Supersedes Better waitEvents implementation for *nix-like systems #78
• Closes Better waitEvents implementation for *nix-like systems #53

[tresf]

Thanks. I'd also ask you to read the code and suggestions here: #78. Note, we're short on volunteers here to help merge, so I do apologize for any delay before this hits our master branch.

[hiddenalpha]

Thx for the hint 👍 Seems as we need a few more tweaks then. Will give it a try.

[graynk]

Is this part still needed? It seems like in this implementation we wait for data before returning in native code (either by poll or by select, even on Mac OS), so this sleep can be removed, if I understand everything correctly

jssc/src/main/java/jssc/SerialPort.java

Lines 1304 to 1310 in 597d06a

	//Need to sleep some time
	try {
	Thread.sleep(0, 100);
	}
	catch (Exception ex) {
	//Do nothing
	}

[tresf]

Maybe I should open individual issues to document all my findings (currently only visible here and there in my scratch-book) so we could discuss what we want do about them.

Fine by me. @GMKennedy and I will be making a new release this week, so I'd like to get this merged in asap if possible. For the record @GMKennedy did compile and test the code on MacOS and Linux with no observable regressions.

[tresf]

@hiddenalpha, Just checking in on the Sleep(...) comments. I'd like to have a decision on this so we can make this part of the next release.

[GMKennedy]

Code changes implement 'synchronized' function (ie monitor lock) which replaces thread sleep function for handling of contention/concurrency. SerialPort.java has been updated accordingly.

[tresf]

AFAIK, @GMKennedy's speaking specifically about this branch: https://github.com/java-native/jssc/commits/Poll-Update

We'll add to this PR in the next day and merge.

[hiddenalpha]

@hiddenalpha, Just checking in on the Sleep(...) comments. I'd like to have a decision on this so we can make this part of the next release.

Sorry for not responding. But had no access to my account because I was not at work a few weeks. AFAIK this decision is superfluous meanwhile :) But honestly I even could not say if it should be removed or not.

[factoritbv]

During our thorough testing (mac/linux/raspberry) of our improvements suggested in #126, we have never run into the issue of file descriptors with a value above 1024. However, it seems to be an issue on (some?) linux systems.

Maybe you could have a look and improve our suggested code with your polling mechanism. I'm not sure if it is as portable as select(), which is basically available on every UNIX system. But if it makes it more stable / flexible, I think the code in #126 can be improved with use of the poll() function.

[hiddenalpha]

... I'm not sure if it is as portable as select(), ...

Yeah :( unfortunately poll is not as portable as select. This was the reason we kept select as a fallback.

... we have never run into the issue of file descriptors with a value above 1024. However, it seems to be an issue on (some?) linux systems.

Unluckily I do not know the exact reason why we ran into such FDs. I only saw that we had them. There were some suspicions that it could come for example from docker or that we use the lib in a jetty where half a dozen apps are running performing lots of file/network IO.