|
Patch which fixes the problem; thanks to Philip for finding this and trying the patch.
It now works on Windows 2003. I also ran derbyall on Solaris 9, JDK 1.4.2 (no errors) and Windows XP to check against regression. The latter has several errors, but run correctly for this test. I don't think bugs should be marked as fixed until their patch is committed.
Committed revision 219265.
Any thought of porting this patch to 10.1? I just ran derbyall on the 10.1 branch and this came up.
Back-port patch to the 10.1 branch
uploaded a svn merge-command which will port the patch for this issue.
(assumed current directory is 10.1). I have run derby-all with no errors. correcting the "Fixed in" to include both trunk and 10.1
updated Fix Version to show bug has been fixed in both 10.1.2.0 and trunk
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
System.err in this case) by redirecting it to
System.out., it seems the underlying file isnt closed immediately.
When the file is subsequently checked, the check for non-empty
races ahead of the actual closing of the file (notice in Philip's log the
file is indeed not empty, it is just the assert which fails).
An explicit close solved the problem on Philips machine. Posting the patch
as soon as i have run some more tests.