But do you also consider the opposite use-case: editing some code in the wiki, and then export a XAR containing the metadata in an XML and the content in a plain text file? Otherwise, it makes it harder to fix some code in the wiki directly before committing it in the repository. No. Not for this case. I guess it could be implemented by detecting which things are outside and which are inside but it's not really trivial.
As far as I know, fetching the xar form the wiki really is a first step which you do not do on a regular basis, you then push adjustments to the wiki which you have edited in the source directory (e.g. using copy and paste or attachments' uploader), then do an upload as a xar to make sure it is clean. The problem with fetching from the wiki is that it always inserts more information than wanted. E.g. a creator name, comments, ... Whereas creating the xar from source can be tailored to not contain any of this information (and part of that is checked by the xar mojo). Am I wrong? Paul