On Mon, Aug 30, 2010 at 18:21, David Holmes <[email protected]> wrote:
> Hi Martin, > > My only minor concern with this is the potential change in timing. Is it > possible that a test will assume the streams are closed when waitFor > returns? If so I guess we need to update the test. > > This change reverts to the historical behavior. One cannot really close any user-visible stream before waitFor returns, because there may be buffered data remaining to be read, at both the java and os pipe level. (this stuff is all surprisingly tricky... further improvements are possible, but are going to be much more work) Martin Cheers, > David > > Martin Buchholz said the following on 08/31/10 04:38: > >> This is a bug report with fix. >> >> My code in UNIXProcess.java.linux introduced a hang if the subprocess >> terminates, but it has remaining descendants that keep the pipe file >> descriptors open. >> >> Here's a fix: >> >> http://cr.openjdk.java.net/~martin/webrevs/openjdk7/ProcessHang/<http://cr.openjdk.java.net/%7Emartin/webrevs/openjdk7/ProcessHang/>< >> http://cr.openjdk.java.net/%7Emartin/webrevs/openjdk7/ProcessHang/> >> >> >> Please review. >> >> Alan, there are remaining (pre-existing) thorny issues with proper >> handling of process streams and asynchronous close, that we might want to >> discuss someday. >> >> >> >> Martin >> >
