On 20/11/2020 19:18, Uwe Schindler wrote: > Thanks for bringing this again. > > I tend to say: Let us just allow also IndexUpgrader beyodn 2 versions! If > somebody complains about incorrect offsets, oh man - It's their problem.
Is there any particular reason why the IndexUpgrader couldn't simply warn about non-upgradable changes? If you have a data type that's no longer supported and there's no migration path: error out. I understand the whole "x != f(x)" upgrade problem, but many scenarios should be pretty straightforward to upgrade. If a field is stored or has docvalues, f(x) can simply be recomputed, no? Am I missing something glaringly obvious here? - Bram --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
