Quote from AntC on October 22, 2019, 9:31 am

Quote from Dave Voorhis on October 22, 2019, 7:30 am

I don't recall if this has been mentioned here before, but I ran across this today and thought it might be of interest to update-through-views proponents:

https://www.cis.upenn.edu/~bcpierce/papers/lenses-toplas-final.pdf

Thanks Dave. There has been mention of Pierce's work on the old forum. I can't remember if it was this particular paper.

• It's working on "tree-structured data" rather than relations.
• Views are defined using lenses, which are pairs of functional mappings from one data structure to another, and vice versa. There's no guarantee those two functions are inverses. So we might delete some content through a view; then insert the same data through the view; we'll find the view's content is same as what we started with; but the 'other end's content not necessarily.
  (There are 'Laws' lenses are supposed to observe, but these apply to put/get the whole content, not components/tuples of the content as we expect with (relational) update-through-views.)
• And this: "[The approach] does not directly address some well-known difficulties with view update in the classical setting of relational databases—such as the difficulty of “inverting” queries involving joins."

The paper discusses "well-behaved" lenses and "very well-behaved" lenses. Plenty of opportunity to rerun all the debates we've had here.