> > But won't that generate invalidation for the rel twice in the case
> > (change Replica Identity from Nothing to some index) you mentioned in
> > the previous email?
> >
>
> Oh, I see the point. I think this is okay because
> AddRelcacheInvalidationMessage doesn't allow to add duplicate rel
> invalidation. If that is the case I wonder why not simply register
> invalidation without any check in the for loop as was the case with
> Tang's original patch?
OK, I also think the code in Tang's original patch is fine.
Attach the patch which register invalidation without any check in the for loop.
WFM.
