Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
Description
In the branch 2.19.2-experimental we discovered that less tests failed on the platform FreeBSD after we flushed stdout. Our old code did this already in some places but some missed this.
As a negative side effect, closing stream, did not solve anything and stream could not be recovered.
We were not sure if FreeBSD platform read the JAR file completely due to we did not close ZIP entry properly in JarOutputStream. We did all for finishing stream and we flushed FileOutputStream due to close() method does not explicitly flush stream, however flush/close contract is not documented, it's just sort of a known implicit behavior in Oracle's JRE that people generally trust. In specific cases where you know you need to call flush() before close() as been JRE platform independent.