Re: plpgsql versus domains

Andrew Gierth <andrew@tao11.riddles.org.uk> writes:
> "Tom" == Tom Lane <tgl@sss.pgh.pa.us> writes:
>  Tom> This is the first attempt at weaponizing the memory context
>  Tom> reset/delete feature, and I'm fairly happy with it, except for one
>  Tom> thing: I had to #include utils/memnodes.h into typcache.h in order
>  Tom> to preserve the intended property that the callback structs could
>  Tom> be included directly into structs using them.  Now, that's not
>  Tom> awful in itself, because typcache.h isn't used everywhere; but if
>  Tom> this feature gets popular we'll find memnodes.h being included
>  Tom> pretty much everywhere, which is not so great.  I'm thinking about
>  Tom> moving struct MemoryContextCallback and the extern for
>  Tom> MemoryContextRegisterResetCallback into palloc.h to avoid this.
>  Tom> Maybe that's too far in the other direction.  Thoughts?
>  Tom> Compromise ideas?

> This was pretty much my first thought on looking at the callback
> patch.  It may seem logical to put the reset callback stuff in
> memutils/memnodes alongside context creation and reset and so on, but in
> fact the places that are going to want to use callbacks are primarily
> _not_ the places that are doing their own context management - since
> those could do their own cleanup - but rather places that are allocating
> things in contexts supplied by others. So palloc.h is the place for it.

That argument sounds good to me ;-).  Moved.
        regards, tom lane