When I search for "PATH+" in the code base, I noticed the following things...

First, all of the calls are either environment.put("PATH+", ...) or environment.override("PATH+", ...), but the class of "environment" differs in each case. In the first one (the "put" case), the "environment" variable is a regular Java HashMap. Its put() method does not contain special handling for the "+" character, so passing "PATH+" to put() seems wrong.

Second, for the calls that use override(), the "environment" variable is a Hudson EnvVars map. This method does support the "+" notation, but the underlying implementation of override() assumes that the pathsep character (either : or ; ) is for the local environment. So I suspect these locations either need to handle the "PATH+" manually (not relying on the EnvVars.override() method), or this code needs to be moved to executing on the remote slave machine.

We can confirm that we are seeing this same bug at our site. We are successfully executing a sed replace of : with ; at the start of the command execution. This allows us to continue unblocked by this bug.

Shannon: how did you install sed on windows? What does your sed look like?

Good question. I have GNU Sed 4.2.1 installed in Program Files (x86)/GnuWin32/bin/sed.exe

I also have sed in my Cygwin installation, by my PATH sets the GNU sed in Program Files (x86) as my primary sed.

Here is a link to that sed: http://gnuwin32.sourceforge.net/packages/sed.htm

Sorry for the spam, here is the code we are using in Jenkins to fix the PATH:

REM ShiningPanda has an existing bug (https://issues.jenkins-ci.org/browse/JENKINS-16176) where
REM the delimiter used to add the virtual Python location ton the PATH is ':' and not ';',
REM causing the generated Python paths to be inaccessible. The following corrects the PATH.
echo Correcting path for ShiningPanda bug 16176
for /f "delims=" %%a in ('\cygwin\bin\echo "%PATH%" ^| sed "s/:C:/;C:/"') do @set "PATH=%%a"

C Girard: What suggests to you that this code is running on the master not the slave? How would the code be moved to execute on the slave?

Latest diggings on this are here:

https://groups.google.com/forum/#!topic/jenkinsci-dev/VgCqabekqPM

I think the analysis from the prior comment is the same reasoning I had for assuming this was running on the master. The path separator is either coming from the "File" class or from the "platform" instance which appears to only come from Platform.current(). Note that "platform" is a private class member, so this rules out subclasses having overridden this or manually setting prior to copy construction.

In either case, getting a colon back would indicate the code is running on a linux machine. In our setup, we have a linux master and windows slave, so it seems like the code is running on the master.