If select returns false, try again

We poll after each failed select to make sure we're not supposed
to wake up and die or propagate an exception. If we are supposed
to block, we should keep blocking here until interrupted or the
stream has data.

If we proceed from here to a blocking read when data is
unavailable, we may end up inside NIO code that closes the stream
upon any interrupt. Selects are interrupt, so we work harder to
stay in select logic when doing a blocking read that might be
interrupted.

core/src/main/java/org/jruby/util/io/OpenFile.java

@@ -1571,7 +1571,11 @@ private static void selectForRead(ThreadContext context, OpenFile fptr, ChannelF
             if (fd.chSelect != null
                     && fd.chNative == null // MRI does not select for rb_read_internal on native descriptors
                     && !fptr.nonblock) {
-                context.getThread().select(fd.chSelect, fptr, SelectionKey.OP_READ);
+
+                // keep selecting for read until ready, polling each time we wake up
+                while (!context.getThread().select(fd.chSelect, fptr, SelectionKey.OP_READ)) {
+                    context.pollThreadEvents();
+                }
             }
         } finally {
             fptr.lock();