I am trying to introduce groovy into a fairly sizable java project (~300k SLOC) and one of the hurdles I have run into is that the groovyc ant task is very slow compared to its javac counterpart. I understand it has to do some extra work for cross compilation but it is slow even when all my source files are java. As a comparison, javac was taking ~1 minute, while groovyc was taking ~5 minutes to compile the same set of files. This was not acceptable to us, but I didn't want to give up so easily so I did a little digging.

What I found is the groovyc task relies on another class called JavaAwareCompilationUnit, which was ultimately calling the following method when collecting the source files:

private void addJavaSource(File file) { String path = file.getAbsolutePath(); for (Iterator iter = javaSources.iterator(); iter.hasNext();) { String su = (String) iter.next(); if (path.equals(su)) return; } javaSources.add(path); }

As you can see this is extremely inefficient as the number of source files gets large. I made a change to this class to change the javaSources variable from a LinkedList to a LinkedHashSet and modified this method to do the following:

private void addJavaSource(File file) { String path = file.getAbsolutePath(); javaSources.add(path); }

Presto! My build time went back down to ~1 minute, almost exactly as long as the javac task was taking. Note that this was all done in non-forked mode (due to issues getting forked mode to work with a long classpath). I may do some investigation into forked mode as well since it looks like there may be a fix in the works for that, but just thought I'd share this so it could be put back into the groovy project and others could benefit.

One other thing I noticed is that the exact same issue appears to be present when adding groovy sources (in the method org.codehaus.groovy.control.CompilationUnit#addSource(SourceUnit source)).