[FIXED JENKINS-8420] recording 3c5c2c5c

changelog.html

@@ -58,6 +58,9 @@ Upcoming changes</a>
 <!-- Record your changes in the trunk here. -->
 <div id="trunk" style="display:none"><!--=TRUNK-BEGIN=-->
 <ul class=image>
+  <li class='major bug'>
+    On IBM JDKs, Jenkins incorrectly ended up closing stdout to read from forked processes.
+    (<a href="http://issues.jenkins-ci.org/browse/JENKINS-8420">issue 8420</a>)
   <li class=bug>
     Jenkins was unable to kill/list up native processses on 64bit Mac JVMs.
   <li class=bug>

core/src/main/java/hudson/Proc.java

@@ -202,12 +202,12 @@ public abstract class Proc {
                 copier2 = new StreamCopyThread(name+": stderr copier", procErrorStream, err);
                 copier2.start();
             } else {
-                // On IBM JDK5, returned input and error streams might be the same. 
-                // So, in this case, closing one isn't a good idea since it will close both.
+                // the javadoc is unclear about what getErrorStream() returns when ProcessBuilder.redirectErrorStream(true),
+                //
+                // according to the source code, Sun JREs still still returns a distinct reader end of a pipe that needs to be closed.
+                // but apparently at least on some IBM JDK5, returned input and error streams are the same.
+                // so try to close them smartly
                 if (procErrorStream!=procInputStream) {
-                    // while this is not discussed in javadoc, even with ProcessBuilder.redirectErrorStream(true),
-                    // Process.getErrorStream() still returns a distinct reader end of a pipe that needs to be closed.
-                    // this is according to the source code of HotSpot JVM
                     procErrorStream.close();
                 }
                 copier2 = null;