http://gnats.netbsd.org/32591 I have the impression that while writing the db in a separate file avoids part of the issue, there is still a problem where the old db can be read as the source file is parsed. So I think this means either: 1) Remove the db before writing, which can result in no db 2) Add a source-only flag to cgetnext, to ignore the binary databases. 2 involves ABI changes 1 has the downside that if the db creation fails, we can be left with no binary db. But presumably cap_mkdb is being run because the source file is newer, and it seems better to end up with the correct answer more slowly than to get the wrong answer. So basically I suggest unlinking the capdb file foo.db as the first thing, just before opening foo.db.tmp for writing.
Description: PGP signature