svn loader: CRLF/LF mess in svn history results in hash computations divergence

When ingesting the asf mirror, if some error occurs during the loading, the process stops.

That's expected and dealt with. At the moment, this will create an occurrence targeting a revision which holds the svn revision and update the origin_visit to set its status accordingly.

When starting again the loading of the repository, as it is already known, it will fetch the last known revision (and its corresponding svn revision). Then svn exports at that svn revision on the disk and then compute its hash tree and then the revision hash. Compare that revision hash with the one already stored in the db.

If it matches, it continues to load from that point on. If it mismatches, it logs an error message about altered history and stops.

This is what currently happens.

BUT: The asf mirror in question did not change and this also happens. It's a bug.

This is symptomatic of a divergence in the directory hash tree computation (which cascades to the revision hash computation). This is about investigating further this tree divergence.

---

Migrated from T570 (view on Phabricator)