Most Linux systems provide an 'ant' but the jdk Makefiles have not allowed it to be used easily at times. This may be due to the way ant is bundled and installed on the system. Most ant's seem to work well if 'ant' and the 'java' needed are in the PATH, so if we can make this work better, fewer people would need to install their own ant just for jdk building. Granted, some people have worked around this, but my experince indicates that we could do better here.

Currently the 'ant -version' is found by the makefiles by trying to point ant at the BOOTDIR jdk as the java to use, and requiring that ANT_HOME be set. The ANT_HOME being required to be set is a Windows thing due to the ant startup script on Windows not being smart enough to know where it's home is, probably a tricky task with shells on Windows. And setting JAVACMD or JAVA_HOME before running 'ant' seems to not work in all cases. The ant in CentOS 5.2 and Fedora 9 seem to behave in strange ways when using this trick, so it's not a universal trick.

The proposal is to just use 'ant -version' and let 'ant' get 'java' from the PATH, where the BOOTDIR java should be. Anything else seems to be problmeatic. This allows the use of the /usr/bin/ant on many Linux systems so you don't have to go find and install your own ant, and also means that an appropriate ANT_HOME need not be found and set, using /usr/share/ant like what is in the makefiles now, just doesn't work in all cases.

So people building the jdk would need to have 'ant' in their PATH and the BOOT java in their PATH, and ANT_HOME can be set if they want but would not be required (however 'ant' will likely require ANT_HOME is set on Windows).

If 'ant -version' does not work, or doesn't give up it's version, this will result in a build failure.