Bug 141058

Incidently, here's how ant.home gets set from the ANT_HOME environment
variable in the classic-ant shell script:

$ grep 'ant\.home' `which classic-ant`
    "$JAVACMD" $ANT_OPTS -classpath "$LOCALCLASSPATH"
-Dant.home="${ANT_HOME}" -Djikes.class.path="$JIKESPATH"
-Dcygwin.user.home="$CYGHOME" org.apache.tools.ant.Main $ANT_ARGS "$@"
    "$JAVACMD" $ANT_OPTS -classpath "$LOCALCLASSPATH"
-Dant.home="${ANT_HOME}" -Dcygwin.user.home="$CYGHOME"
org.apache.tools.ant.Main $ANT_ARGS "$@"
    "$JAVACMD" $ANT_OPTS -classpath "$LOCALCLASSPATH"
-Dant.home="${ANT_HOME}" -Djikes.class.path="$JIKESPATH"
org.apache.tools.ant.Main $ANT_ARGS "$@"
    "$JAVACMD" $ANT_OPTS -classpath "$LOCALCLASSPATH"
-Dant.home="${ANT_HOME}" org.apache.tools.ant.Main $ANT_ARGS "$@"

Fedora Core 2 is now maintained by the Fedora Legacy project for
security updates only. If this problem is a security issue, please
reopen and reassign to the Fedora Legacy product. If it is not a
security issue and hasn't been resolved in the current FC3 updates or
in the FC4 test release, reopen and change the version to match.

Due to the native-compiled version of the ant executable, this is still a
problem in FC3, and will be a problem in FC4 as well.  Please let me know what
additional information I can provide.

Thanks. I'm going to set this to version "devel" for FC4.

This won't be an issue for FC4.  Current rawhide has bytecode ant, and the new
BC ABI in libgcj means that stuff shouldn't break when we roll out the native
stuff (Real Soon Nowâ¢).

Gary, thanks for the feedback!  This bytecode API sounds like a major
improvement which will likely fix a ton of the classloading issues I've seen
with the native-compiled ant binary.  However, I'm unsure how it alone could fix
the setting of ant.home, a java.lang.System property which is set for the JVM by
all the traditional Ant invocation wrapper scripts, and assumed by many Ant
tasks and build.xml files to be available.

To fix this issue, it seems to me that some additional code would need to be
executed by the gcj-compiled binary on startup which performs the additional
operations handled by the traditional wrapper scripts (operations which occur
before any Java bytecode is ever executed).  Missing the ant.home system
property is only once instance of this class of problem; note other system
properties like jikes.class.path and cygwin.user.home which are set by the
snippet of the traditional shell wrapper script which I pasted into comment #1.

For most developer/end users of Ant, Ant is more than the its Java bytecodes. 
It's also that set of wrapper scripts which insulate the user from environment
specific issues which they'd prefer not to have to know about.  :-)

In FC4 we're using the standard ant wrapper scripts -- the gcj-compiled binary
is history.

Created attachment 120903 [details]
bz141058.sh

simple test case

Seems to work in FC4:

|$ curl -o bz141058.sh \
|  'https://bugzilla.redhat.com/bugzilla/attachment.cgi?id=120903'
|  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
|                                 Dload  Upload   Total   Spent    Left  Speed
|100   242  100   242    0     0    290      0 --:--:-- --:--:-- --:--:--  236k
|$ bash bz141058.sh
|Buildfile: /dev/fd/63
|
|test:
|     [echo] ant.home=/usr/share/ant
|
|BUILD SUCCESSFUL
|Total time: 1 second


Closing.

Nice work guys.  Verified on FC4.