fyi, I found that _get_revision_number()'s -1 revision for added nodes would leak at least into an error message from 'svn log'.
Checking that out I noticed that -1 is the revnum returned for _revision_unspecified; 'svn log' in that case assumes HEAD. Probably other callers do the same or similar things, not checking _unspecified explicitly. The problem is that an added node returning SVN_INVALID_REVNUM for _revision_base would look like it was _revision_unspecified. Instead, svn_client__get_revision_number() asked for an added node's _base should probably rather throw an error. ~Neels