IO#ready? returns incorrect results

[rjnienaber]

From: http://jira.codehaus.org/browse/JRUBY-5617. Works JRuby 1.7.18 (with discrepancies), fails on JRuby 9000.

IO#ready? returns incorrect results on second and subsequent calls.

a) Write 100 bytes to a sender socket, issue flush, keep socket open
b) Attempt 4 consecutive reads of 25 bytes each. Check IO#ready? before reading

Expected results: 100 bytes are read in 4 consecutive reads
Observed results: The first 25 bytes are read, but then the read socket never shows ready again.

Sample code is attached. Compare JRuby behavior with 'standard' Ruby using this code.

[nirvdrum]

I'm a moron and referenced the wrong issue number in my commit message. It should have been #2338. Sorry for all the resulting noise on this issue as a result.

[headius]

Wasn't a hard fix; I re-ported the logic from MRI but where they used FIONREAD and ioctl to check for available buffered bytes, I had no response. The old implementation used select to accomplish the same thing, so I reinstated that behavior.

I also cleaned up much of io/wait and got the MRI tests running. We only fail two of them, and I'm pretty sure it's due to JVM/JDK abstractions we can't avoid.