> > Ideally, open the database which exhibits the problem and first execute >
> PRAGMA integrity_check; > Integrity check does not report any issue. > If this fails to show any problems run your UPDATE command, repeatedly if > necessary, and see if you can make the shell tool do anything that suggests > it isn't working perfectly. > I cannot reproduce the issue in the shell but that doesn't mean much as this differs in many ways from the real scenario: - the shell uses shared cache - the shell is single threaded - the shell doesn't use transactions - I can't accurately reproduce all the other operations that happen to the database (insert, delete, select, ...) prior to the failing update and may contribute to the issue in subtle ways ... I would be more interested in any ways I can get more verbose logging or some sort of execution trace out of sqlite: re-running tests against a patched sqlite is more practical than trying to extract a minimal reproducer at this point.