Is it possible for the same javac compiler to compile the same set of source files but produce class files of different checksums?

Question

I was attempting to compare the results of this: (in ant)

                       <javac 
                           target="1.5"   
                           source="1.5"
                           deprecation="on"
                           fork="yes"
                           optimize="true"
                           debug="true"
                           debuglevel="lines,vars,source">

                            <classpath>
                                <fileset dir="${project.basedir}/../lib">
                                    <include name="**/*.jar" />
                                    <include name="**/*.zip" />
                                </fileset>                        
                            </classpath>     
                        </javac>

...against this: (in maven)

        <plugin>                
            <artifactId>maven-compiler-plugin</artifactId>
            <version>2.3.2</version>
            <executions>
                <execution>
                    <id>compile</id>
                    <phase>compile</phase>
                    <goals>
                        <goal>compile</goal>
                    </goals>                
                    <configuration>
                        <target>1.5</target>
                        <source>1.5</source>
                        <debug>true</debug>
                        <debuglevel>lines,vars,source</debuglevel>
                        <showDeprecation>true</showDeprecation>
                        <optimize>true</optimize>
                        <fork>true</fork>
                        <includes>
                            ...
                        </includes>
                    </configuration>
                </execution>
            </executions>
        </plugin> 

...only to discover that 2 of the resulting class files have slightly different checksums, while the rest are identical. I consider multi-threading to have played a part, but the checksums produced for either option appear to be consistent upon repeated attempts. What can possibly account for this result?

Update:

I have examined one set of the files with different checksums using javap -verbose and noticed the only difference being:

const #16 = class   #108;   //  java/lang/Exception
const #17 = Method  #102.#109;  //  java/io/Writer.close:()V
const #18 = Method  #7.#109;    //  java/io/FileWriter.close:()V

as opposed to:

const #16 = Method  #102.#108;  //  java/io/Writer.close:()V
const #17 = Method  #7.#108;    //  java/io/FileWriter.close:()V
const #18 = class   #109;   //  java/lang/Exception

Answer 1

as you showed, the difference is in the generated constant pool, which is really a non-issue, but troubling as you would expect the same output with the same compiler and options. i would bet that the compiler is called with the java files in different order between the two situations, and the order of compilation is affecting the result.

Answer 2

I'm using javac compiler and I've found it produces different resulting binary .class files depending on the order of the source files passed as parameter. Between maven and ant I found differences. The order of the files were not the same.

Anyway, between the 2 compilations there were differences. I've disassembled the code and I found the javac (optimizer?) was removing the deprecated java jsr/ret assembly instructions from one of the compilations.

I don't know if this behaviour it's the result of "implicit" compilation: http://docs.oracle.com/javase/1.5.0/docs/tooldocs/solaris/javac.html#searching