I wrote:
> The argument for "sending too little" comes from the race condition
> that's described in the function's comments: a variable that has
> source PGC_S_DEFAULT (ie, has never moved off its compiled-in default)
> in the leader could have just been updated in the postmaster, due to
> re-read
In the thread about valgrind leak detection [1], we noticed that
RestoreGUCState(), which is intended to load the leader process's
GUC settings into a parallel worker, was causing visible memory
leaks by invoking InitializeOneGUCOption() on already-set-up GUCs.
I noted that simply removing that cal
