exit %ERRORLEVEL%

Thereby resolving the original issue.

...
1 error
2 warnings
BUILD FAILED
File...... file:/C:/Documents and Settings/user/.maven/plugins/maven-test-plugin-1
.4/
Element... javac
Line...... 34
Column.... 46
Compile failed; see the compiler error output for details.

BUILD FAILED
C:\tmp\build.xml:160: Following error occured while exec
uting this line
C:\tmp\build.xml:103: exec returned: 70

Total time: 8 seconds

Windows batch files are quite different from processes, and are always executed in the context of a command prompt (cmd.exe). Ant worries about the return code of the cmd process, not the executed batch file's ERRORLEVEL.

to avoid terminating cmd.exe if maven.bat was invoked from
the command line.

"exit /b %ERRORLEVEL%" at the end of the batch file propably does nothing, as it seems to be intended to exit the batch somewhere in the middle.

"exit %ERRORLEVEL%" at the end unfortunately terminates the process with the exit code (which is what you want when treating a batch file as a process, but not when you invoke it on the command line).

I've been a staunch proponent of Maven (against much resistance I might add) at my company. This issue was highly visible... and not in a good way. As you might imagine, failed builds that no one knows about are pretty serious.

All I know is that this used to work, the build machine has not been OS upgraded to XP or anything, and the EXIT %ERRORLEVEL% seems to fix the problem (although with the aforementioned nasty side-effects if you are running Maven in a CMD window yourself).

Could someone at least take the time to explain why this isn't a Maven bug? Stuff like this may sound trivial but I can assure you that it is not. The fact of the matter is that most people use and are familiar with Ant, so making Maven seamlessly callable from Ant is a pretty big requirement. The Maven Ant task seems to be outdated (I spent quite a bit of time trying to get it to work but never solved forehead-related classloader issues).

Now, the problem is the shell's exit code. As far as I can tell,
it's either the result of the last call in the batch file, or
the value explicitly set via the "EXIT" call. For example, the
following batch file, when executed via <exec>, fails as it should.

— build.xml
...
<exec executable="some.bat" failonerror="true"/>
...
— some.bat
@echo off

java SomeMissingClass
—
C:\temp>ant
Buildfile: build.xml

all:
[exec] java.lang.NoClassDefFoundError: SomeMissingClass
[exec] Exception in thread "main"

BUILD FAILED
C:\temp\build.xml:4: exec returned: 1
—

Unfortunately, as soon as you add other commands the
status gets lost (at least some commands, eg
SET does not harm the status).

— some.bat
@echo off

java SomeMissingClass

goto end

:end
—
[exec] java.lang.NoClassDefFoundError: SomeMissingClass
[exec] Exception in thread "main"

BUILD SUCCESSFUL
—

Windows also sets a pseudo-environment variable ERRORLEVEL to
the exit status of the last executed process, in the example
above to the exit status of "java," which would be the one
we want.

— some.bat
@echo off

java SomeMissingClass

goto end

:end

exit %ERRORLEVEL%
—
[exec] java.lang.NoClassDefFoundError: SomeMissingClass
[exec] Exception in thread "main"

BUILD FAILED
—

But if the batch file is called by an existing shell (command line),
it is executed in that shell, and the exit call would end it, not
just the batch file.

One could explicitly spawn the shell, call the batch file, and end
it with EXIT.

— build.xml
<exec executable="cmd.exe" failonerror="true">
<arg value="/c"/>
<arg value="call some.bat; EXIT %ERRORLEVEL%"/>
</exec>
—

Note the "/c" and the "call" before some.bat.

I have not yet had time to investigate the specifics of what happened on this particular build, but the additional details are greatly appreciated.