JDK 9 RFR : java.util.logging.Formatter#formatMessage() swallows Exceptions
alexander.fomin at oracle.com
Fri Nov 20 17:27:30 UTC 2015
On 20.11.2015 17:47, Jason Mehrens wrote:
> Why not just cache the last exception in the formatter and use getTail to clear it and report it?
It might be a good solution, but according to spec
Formatter#getTail() method is intended to return the tail string for a
set of formatted records. With your approach it would need to modify the
spec any way with notes about possible exception messages in the queue.
Besides that, the tail of the records should be the latest record.
I think that according to current spec it must be just that unformatted
message which had been passed to j.u.l.Formatter#formatMessage() and
failed to be formatted. Therefore the logic in Handler would be go
through all records in order to find Exceptions and report them if
found. Looks a little bit strange.
But what I really care about is possible compatibility issue if
getHead()/Tail() overridden in user's Formatter implementation. One of
the example is j.u.l.XMLFormatter. Even if Exceptions have been cached
in base Formatter, it would have not been reachable in Handler with
current XMLFormatter.getTail() implementation. Well, we can modify
XMLFormatter.getTail() since it's Java code, but I don't think user's
will happy with an idea to modify anything in their app code.
> Since formatter is in the same package as Handler you will have elevated access to the error manager through Handler.reportError.
Well, to which Handler? Formatter don't know anything about Handler(s)
that use(s) it. So, the only way in Formatter to get the reference on
some Handler are methods where it passed getHead()/getTail(). Taking
into account these methods may be overridden in user's implementations
the solution doesn't seem reliable.
And then, with setErrorManager()/getErrorManager() user may set his
Any way, taking into account that at least some spec changes would need
in any case, it would better to extend API for flexibility and reliability.
> That also makes it so you don't have to change the public API of Formatter.
> From: core-libs-dev <core-libs-dev-bounces at openjdk.java.net> on behalf of Alexander Fomin <alexander.fomin at oracle.com>
> Sent: Friday, November 20, 2015 7:48 AM
> To: core-libs-dev at openjdk.java.net; Daniel Fuchs; mandy.chung at oracle.com
> Subject: JDK 9 RFR : java.util.logging.Formatter#formatMessage() swallows Exceptions
> please, review this patch to report errors in
> Bug: https://bugs.openjdk.java.net/browse/JDK-8137005
> Webrev: http://cr.openjdk.java.net/~dfuchs/alexander/8137005/webrev.00
> j.u.logging.Formatter#formatMessage() swallows exceptions that
> happening during formatting of a message. In the result the exceptions
> are lost and users don't know about reasons why the message hasn't been
> formatted as expected. We would avoid to throw any exceptions in
> Formatter#formatMessage() from compatibility stand point. To report an
> error in consistent way we have to pass ErrorManager in Formatter. It's
> require API changes. So, I'm going to file CCC when if the fix approved.
> The suggested fix is to add 2 new methods in j.u.logging.Formatter
> to set/get an ErrorManager, update Formatter#formatMessage() to report
> errors via the ErrorManager and update Handler to pass errorManager to
> A couple of new regression tests have been created:
> test/java/util/logging/Test8137005.java - real case provided by
> test/java/util/logging/NullErrorManagerTest.java - additional
> check to make sure no NPE showed if ErrorManager isn't set. Beside of
> this touched new API methods.
> Logging regression tests have been run:
> All tests passed passed.
> failures in the job are known issues and not related to the fix.
More information about the core-libs-dev