@jmchilton would you mind putting eyes to this if you have time? @martenson you might be interested in this as well.

The Good

<?xml version="1.0" encoding="UTF-8"?>
<testsuite name="planemo-update"
           tests="1"
           errors="0"
           failures="1"
           skip="0">
    <testcase classname="seqtk" name="" time="6.95435190201">
            <error type="planemo.FailedUpdate" message="Failed to update repository">
            </error>
        <system-out>
        cd '/home/hxr/work/tools-iuc/tools/seqtk' &amp;&amp; git rev-parse HEAD
        cd '/home/hxr/work/tools-iuc/tools/seqtk' &amp;&amp; git diff --quiet
        </system-out>
        <system-err>
        Could not update seqtk
        Unexpected response from galaxy: 400: {"content_alert": "", "err_msg": "No changes to repository."}
        </system-err>
    </testcase>
</testsuite>

XUnit reports for shed_update and shed_diff will now contain stdout/stderr, and make those available to browse in Jenkins

The Bad

Logs under --report_xunit are no longer pretty colours. They're boring black and white. Seems like an acceptable loss since you gain awesome XUnit reports.

If --report_xunit is not specified, tests are still colourful.

The Ugly

This PR provides a small class called planemo.io.Capturing which lets you capture stdout/stderr for any function calls, and then process that data at your convenience. You don't have to rearchitect any of the existing infrastructure for logging user facing messages, you can just capture it and deal with it as you please. This code smells and it's pretty stinky. Instead of some way to handle message propagation properly and dispatching messages to handlers which care about them, we monkey patch sys.std*. I am not proud of this, but it solves the problem simply and quickly with very few changes.

The whole of cmd_shed_update and cmd_shed_diff are probably in dire need of a refactoring