I was still playing with classloaders and namespaces and antlib import with the
AntDSL, and I encountered an unexpected limitation.
What I found: an antlib which is referencing itself in its definition cannot be
loaded dynamically via an uri other than its antlib one.
This is the case for antunit:
<antlib xmlns:au="antlib:org.apache.ant.antunit">
[...]
<taskdef name="fail"
classname="org.apache.ant.antunit.AssertTask"/>
[...]
<macrodef name="assertTrue" backtrace="false">
<attribute name="message" default="Assertion failed"/>
<element name="assertion" implicit="true"/>
<sequential>
<au:fail message="@{message}">
<assertion/>
</au:fail>
</sequential>
</macrodef>
And the following doesn't work:
<taskdef classpath="antunit.jar" uri="urn:antunit"
resource="org/apache/ant/antunit/antlib.xml" />
(actually, this precise line works, but trying to use any antunit macro will
fail)
I think that the issue is how Ant interpret the namespace declaration in the
antlib definition. In Antunit:
<antlib xmlns:au="antlib:org.apache.ant.antunit">
When Ant is parsing it, it considers it as an antlib to be resolved later,
whereas it should detect that this is the canonical uri of the antlib it is
parsing, or sort of "this", and should use the uri actual used rather than the
antlib one.
The behavior only affect cases where somebody is trying to load two different
versions of the same antlib. This is some quite edge case, but do we agree this
is an issue ?
I have committed a test case to see it in action. See r1385269. To run it,
rename first broken_testURI to testURI in
src/tests/antunit/taskdefs/taskdef-antlib-test.xml.
I had a quick look to how to fix it. As far I can tell, the best way to do it
is to:
* in Definer.loadAntlib(ClassLoader, URL), make the createAntlib aware of the
"resource" it was loaded from, and try to build an antlib uri from that.
* in Antlib.createAntlib(), let the projecthelper parse the file to an unknown
element with the 'incorrect' uri. But then make a deep lookup in the tree of
UnknownElements and change everything that needs to be changed.
It seems not trivial.
Let me know what you think.
Nicolas
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
