ExternalTerminal#close does not stop the pump thread

[oehme]

This means that the thread's context classloader is leaked and eventually leads to OutOfMemoryErrors, e.g. when repeatedly running Maven 4 in embedded mode.

[oehme]

Here's what a heap dump from one of our CI agents looks like:

Following the code it looks like it should stop eventually, since "paused" is set to true. But this doesn't seem to happen reliably, as I see over a dozen leftover ExecTerminal pump threads in that heap dump.

[oehme]

Digging deeper into the stacktrace where these threads are stuck, it's

java stack of ExternalTerminal[name='Maven', type='ansi', size='Size[cols=160, rows=50]'] input pump thread
    at java.util.concurrent.locks.LockSupport.park(java.lang.Object) (line: 221)
    at java.util.concurrent.locks.AbstractQueuedSynchronizer.acquire(java.util.concurrent.locks.AbstractQueuedSynchronizer$Node, int, boolean, boolean, boolean, long) (line: 754)
    at java.util.concurrent.locks.AbstractQueuedSynchronizer.acquire(int) (line: 990)
    at java.util.concurrent.locks.ReentrantLock$Sync.lock() (line: 153)
    at java.util.concurrent.locks.ReentrantLock.lock() (line: 322)
    at jdk.internal.misc.InternalLock.lock() (line: 74)
    at java.io.BufferedInputStream.read(byte[ ], int, int) (line: 397)
    at java.io.FilterInputStream.read(byte[ ]) (line: 95)
    at org.jline.terminal.impl.ExternalTerminal.pump() (line: 156)
    at org.jline.terminal.impl.ExternalTerminal$$Lambda+0x0000007002013e00.run()
    at java.lang.Thread.runWith(java.lang.Object, java.lang.Runnable) (line: 1596)
    at java.lang.Thread.run() (line: 1583)

So they are waiting for input indefinitely and never reach the if-statement further down where they check whether they are paused.

[gnodet]

There are two pause methods on ExternalTerminal. The doClose method seems to call pause() instead of pause(true) which would interrupt the reading thread.

[gnodet]

Seems similar to #1162

[oehme]

InputStream#read can't be interrupted, so that unfortunately won't fix it. You could busy-wait on input.available >0 instead and then only read as much as is available. That would ensure you never block.

[oehme]

I managed to work around this in our Maven tests by replacing System.in with a ByteArrayInputStream for each test. That way, the pump thread gets an EOF and shuts down.