Testing with NB 3.5 post-beta (clean English build
from trunk), clean userdir.
Tryed to connect to stable & beta AU servers - OK.
Trying to connect to alpha AU server - Error.
Shortened ide.log attached
Recently the alpha AU server files have been truncated. dev_1.6_.xml
is empty. 35_1.6_.xml ends suddenly halfway through. Guilty logs:
date: 2003/04/26 03:06:12; author: rbalada; state: Exp; lines:
Automatically updated during NetBeans release35 200304252350 build
date: 2003/04/26 04:10:47; author: rbalada; state: Exp; lines: +0 -8478
Automatically updated during NetBeans dev 200304260100 build
Once again, I would suggest that issue #31961 be implemented - it
should be rather easy - as that would cause the build to signal an
immediate error in case bogus or ill-formed update server XML files
are being created, rather than waiting for someone to discover the
problem manually. Or for the short term (3.5), even without changes to
the task, it should be simple to run <validatexml> on the generated files.
This is my bug. I'm testing new infrastructure and it looks like
conflict between the infrastructures.
The very first investigation discovered, that 35_1.6_.xml file is
90112 bytes long, what is (64 + 24) * 1024. Maybe some
buffer/filestream is not properly flushed before close.
Related fix done in trunk.
nbcvs ci -m "Issue 33264. Added OutputStream and PrintWriter flush ()
right before their close()"
Checking in antsrc/org/netbeans/nbbuild/MakeUpdateDesc.java;
new revision: 1.15; previous revision: 1.14
Processing log script arguments...
Mailing the commit message to email@example.com (from
pw.flush() is probably useless. Writer's generally are documented that
close() implies a flush() first, and in fact PW.close calls here
OutputStreamWriter.close, which ultimately calls OS.close, after
writing all remaining converted characters; and FileOutputStream.flush
according to JDK 1.4 sources does nothing (no buffering).
So calling os.flush is probably useless as well. The Javadoc for
OutputStream.close does not explicitly state that there is an implied
flush first (as Writer does), but I think that is the intent - e.g.
FilteredOutputStream's close (used also by BufferedOutputStream, which
*does* implement flush) does an automatic flush before closing.
Probably the calls to os.close() are gratuitous as well - just calling
pw.close() should suffice. Calling os.close() *afterwards* should be
So I am not sure the attached patch will do anything at all - the code
may have been correct without it. Were you able to reproduce the
creation of the truncated files and confirm that the patched
<makeupdatedesc> works correctly? If not, there may be some other problem.
Fixed for trunk builds. The trunk builds are run on another build
infrastructure which was changed a little bit from the one used for
1) L10N modules have been put into separate module groups (folders)
by language (no more "garbage" in the root)
2) dev_1.6_.xml is actually 435737 bytes long
I was able to successfully work with dev/release35 alpha update
centres on J2SDK 1.4.1, from NetBeans dev build 200305050100 and
release35 2003292350 (RC1).