1. There is a wide agreement on implemented feature - nothing changed from previous review - it is not necessary comment it again.
2. v6 patch: patching cleanly, compilation without errors and warnings, all regress tests passed
Tom's objections was related to GUC part. It is redesigned as Tom proposed.
The code is good - and I don't see any problem there.
I have only one objection - What I remember - more usual is using a list instead a bitmap for these purposes - typical is DefElem struct. Isn't it better?
A using DefElem will be longer, but it is typical pattern for this case in Postgres.
TBH, if I thought this specific warning was the only one that would ever be there, I'd probably be arguing to reject this patch altogether.
Of course, nobody assumes that it will be the only one.
Also, adding GUC_LIST_INPUT later is not really cool since it changes the parsing behavior for the GUC. If it's going to be a list, it should be one from day zero.
Actually it does not since it all has to be handled in check/assign hook anyway.
But nevertheless, I made V6 with doc change suggested by Alvaro and also added this list handling framework for the GUC params. In the end it is probably less confusing now that the implementation uses bitmask instead of bool when the user facing functionality talks about list...
This obviously needs code review again (I haven't changed tests since nothing changed from user perspective).
