On 26/01/11 14:57, Chris Dollin wrote:
On Wednesday, January 26, 2011 01:18:36 pm Andy Seaborne wrote:
The test suite is producing warnings:
Should these be warnings at all?
WARN [main] (LiteralLabelImpl.java:129) - inventing a datatype for class
java.sql.Date
Yes, they should be at least warnings. The literal system has been asked
to create a typed literal for which it does not have a registered
datatype object. So it creates one (and re-uses it as necessary).
Good programs won't do this, but it allows less-good [1] programs,
and tests, to carry on.
Making it an error would mean fiddling around with the regression
tests -- or finally dumping them, since they are pure legacy and
their job was taken over by NewRegression. Making it configurable
would be a pain (adding a new global flag is an a-priori bad idea,
and doesn't work if different models would have different expectations).
I'm thinking ... pastels! and dumping the old regression suite. Less
clutter to carry forward.
Chris
[1] ++ungood programs have more serious issues.
Can the output of the warning be suppressed during testing? For
example, beforeClass/afterClass bracketing that changes the logging
levels. org.openjena.atlas.logging.Log has some code to enable/disable
logging for log4j programmatically which might help. The logging
messages aren't checked in the tests are they?
Having a build/test that causes output is problematic. It's less error
prone to look through the output and see no messages, or if they are,to
take action. Having to sift through the messages to notice what's
expected and what isn't is liable to miss things.
(The same is true of javadoc warnings - but these aren't from this code)
Andy
