[[!comment format=mdwn
 username="joey"
 subject="""comment 1"""
 date="2018-02-08T16:59:16Z"
 content="""
Is the benefit of doing this being able to correlate the error message with
the filename whose processing caused the error?

For that benefit to be realized, the error message would have to be included
in the same json object with the file being processed. If a separate json
object were emitted containing only the error message it would not be clear
what file it was related to (especially when git-annex is running
concurrent jobs), and so that does not seem any better than
output to stderr.

Looks like implementation would just involve making outputError 
check if there is a jsonBuffer, and add the error message to it,
probably using an array since there could be multiple warning messages.

I kind of feel like they should still go to stderr in addition to any json
output, because any existing json consumers may rely on the current stderr
behavior to let the user know what went wrong.
"""]]