At Thu, 15 Mar 2018 00:33:25 -0400, Tom Lane <tgl@sss.pgh.pa.us> wrote in <22193.1521088405@sss.pgh.pa.us>
> Kyotaro HORIGUCHI <horiguchi.kyotaro@lab.ntt.co.jp> writes:
> > Doesn't it make sense if we provide a buildtime-script that
> > collects the function names and builds a .h file containing a
> > function using the list?
>
> Surely this is a fundamentally misguided approach. How could it
> handle extension GUCs?
I understand it is "out of scope" of this thread (for now). The
starting issue here is "the static list of list-guc's are stale"
so what a static list cannot cope with is still cannot be coped
with by this.
As the discussion upthread, with the dynamic (or on the fly)
approach, pg_dump fails when required extension is not
loaded. Especially plpgsql variables are the candidate stopper of
ordinary pg_dump operation. We might have to implicitly load the
module by any means to make it work. If we treat extensions
properly, we must find the extension that have defined a GUC that
is about to be exported, then load it.
I don't think the automatic stuff is essential but the
check_listvars.h is still effective to reduce the effort needed
to maintain the multiple lists that should have the same set of
names of the list-gucs.
Or, we could cope with this issue if the list-ness of used GUCs
is stored permanently in somewhere. Maybe pg_proc.proconfigislist
or something...
regards,
--
Kyotaro Horiguchi
NTT Open Source Software Center