Hi,
On 2026-01-03 18:05:19 +0100, Peter Eisentraut wrote:
> On 02.01.26 20:17, Andres Freund wrote:
> > On 2025-01-20 15:01:08 +0100, Peter Eisentraut wrote:
> > > This has been committed.
> >
> > I don't like the const markings in PageGetItem():
> >
> >
> > /*
> > * PageGetItem
> > * Retrieves an item on the given page.
> > *
> > * Note:
> > * This does not change the status of any of the resources passed.
> > * The semantics may change in the future.
> > */
> > static inline void *
> > PageGetItem(const PageData *page, const ItemIdData *itemId)
> > {
> > Assert(page);
> > Assert(ItemIdHasStorage(itemId));
> >
> > return (void *) (((const char *) page) + ItemIdGetOffset(itemId));
> > }
> >
> > The const for PageData seems like a lie to me, because we cast it away. And
> > indeed, we often then use the returned value to set hint bits etc.
>
> I agree this is incorrect. Since no callers appear to rely on the const
> qualification of the argument, the easiest solution would be to just remove
> it. See attached patch.
LGTM
> (In the future, this might be a candidate for a qualifier-preserving
> polymorphic function like the C23 string functions, but that's for another
> day.)
Makes sense. I doubt this is the place I would start with that though, the
amount of effort that'd take around all the functions dealing with heap tuples
seems too large...
Greetings,
Andres Freund
