I've been trying to get ANTLR 2.7.4 to run under Cygwin with GCJ. After trying to piece information together, I've finally found a recipe that makes it work.

<p> ANTLR comes with a standard <code>./configure</code> script. Under Cygwin, it executes fine, but during compilation, there is an error related to AWT in directory <code><i>base</i>/antlr/debug/misc</code>. AWT seems to be a JAVA standard library.

<p>A previous <a href="http://www.jguru.com/faq/view.jsp?EID=138">article</a> shows how to work around the error, but it didn’t solve everything for me: I was able to create a <code>libantlr.so</code> file (see step 4 of that page), but step 5 broke down, probably because some things in my path weren't set correctly (I think). Step 5 relies on a file called <code>antlr/Tool.class</code>, which doesn't exist in the <code>./antlr</code> directory but in the <code>antlr.jar</code> file instead (a jar file is a tar-like library in which .class files can be gathered.)
In the end, I got it to work by un-jarring all the files and simply bypassing the creation of a shared library.

<p>Here's what I did: (<code>basedir</code> is the directory where you can find ./configure)
<OL>
<LI>Install gcj (through the setup.exe on <a href="http://www.cygwin.com">cygwin.com</a>)
<LI>Download the tar files of the latest version of ANTLR (2.7.4 in my case) and untar it in a local directory.
<LI>Run <code>./configure --prefix=<i>install dir</i></code>. While we do not need to configure to compile the antlr executable, we <b>do</b> need it to compile the support C++ libraries. Do NOT run make after configure!
<LI>Delete AWT dependent code:
<code>cd <i>basedir</i>/antlr/debug
rm -fr misc</code>
<LI><code>cd <i>basedir</i></code>
<LI>Un-jar all the files in the antlr.jar, back into their original position.
<code>jar xfv antlr.jar</code>
<LI>Compile everything into an executable
<code>gcj --main=antlr.Tool `find antlr -name "*.class"` -o cantlr</code>
If all is well, you will see a bunch of warnings on your screen that, I assume, you can safely ignore. The end result is a file called <code>cantlr.exe</code>).
<LI>Test the executable by running it (<code>./cantlr</code>). You should see a bunch of lines with program information.
<LI>Move this file to a place somewhere in your path (e.g. <code>/usr/local/bin</code>)
</OL>
At this point, we have created that main executable that will convert a .g grammar file into a set of Java, C# or C++ files. These newly generated files, however, rely on base classes that are also part of the ANTLR distribution. For C++, we need to build a library that contains the compiled based classes.

<OL>
<LI><code>cd <i>basedir</i>/lib/cpp</code>
<LI> Build the C++ library. This step won't work if you previously didn't run <code>./configure</code>.
<code>make</code>
<LI> If everything went fine, then the <code>./src</code> directory will contain a file called <code>libantlr.a</code>. Now install the library and include files to the place that was originally indicated during the <code>./configure</code> step.
<code>make install</code>
</OL>
The next step is to test if now have a fully working system. 
<OL>
<LI><code>cd <i>basedir</i>/examples/cpp/calc</code>
<LI>The standard Makefile will not know where to find the antlr executable. As a work around, just generate all files by manually invoking antlr.
<code>cantlr calc.g</code>
A set of C++ files will now be generated.
<LI><code>make</code>
<LI>If everything went fine, you will now see a fresh set of executables!
</OL>
This is how I got the system work for me while using GCJ and Cygwin. I'm pretty sure that the same procedure will also work in a Solaris or Linux environment...

<p>Tom Verbeure