Currently the FileImporter applies AbuseFilter checks for the contents of every text revision imported. - This was done to make sure, that abusive content - even if it is not part of the current revision - will not be imported to the target wiki.
From T206486 we got an example where an AbuseFilter -rule on Commons was triggered, because one of the revisions contained empty text and the rule makes sure, that empty text cannot be submitted. When "just" an old revision is affected by having no text this does not really seem to be an issue that should stop the import.
- image that should be imported: https://mr.wikipedia.org/wiki/चित्र:Sabudana.jpg
- filter rule to avoid page blanking https://commons.wikimedia.org/wiki/Special:AbuseFilter/4
The example shows that in some cases it might not make sense to run the AbuseFilter checks for every "old" text revision. Some ideas how to deal with that:
- dispose the checks for old revisions.
- partly dispose the checks (e.g. only check the summary lines of old revisions but not the text)
- find a way to tell the AbuseFilter that we're in the middle an import process and that we're dealing with "old" text revisions
Probably the last point is the best way to go here. Then the maintainers of the filter rules can decide for each rule what to do in our case. We're just not entirely sure yet if this is doable and how it should be done best.