[lehni]

patchAndFetch(), patchAndFetchById(), updateAndFetch() and updateAndFetchById() merge the fetched row into the model they patch with. Each run works on a copy of the query, but that model is shared, so awaiting the same query builder a second time patched all fetched columns, writing back values that other writes had changed in between (#519, since at least 1.0.0).

A second run of such a query now throws, and the docs of execute() say to create a new query for each run. The query counts as run once the fetched row is merged, so a run that fails before that, e.g. on a deadlock, can be retried. Concurrent runs of the same query aren't detected.

It throws rather than returning a new model, as the issue proposed, because patchAndFetchById() returns the model that was passed in as the patch, which a test pins since 2015. Inserts aren't guarded: they also write their ids into the given models, but re-running them already fails with a unique violation, or works as intended with onConflict().

As this turns a re-awaited patchAndFetch() that mostly worked into an error, it's listed under Changes for 4.2.0.

Closes #519