Re: HOT chain validation in verify_heapam()

On Thu, Feb 9, 2023 at 12:09 PM Himanshu Upadhyaya
<upadhyaya(dot)himanshu(at)gmail(dot)com> wrote:
> Initially while implementing logic to identify the root of the HOT chain
> I was getting crash and regression failure's that time I thought of having
> this check along with a few other changes that were required,
> but you are right, it's unnecessary to add data to the predecessor
> array(in this case) and is not required. I am removing this from the patch.

I finally found time to look at this today -- apologies for the long
delay -- and I don't think that it addresses my objections. When I
proposed lp_valid, I had a very simple idea in mind: it tells you
whether or not the line pointer is, at some basic level, valid. Like,
it contains numbers that could point to a tuple on the page, at least
hypothetically. But that is something that can be determined strictly
by inspecting the line pointer, and yet you have
check_tuple_visibility() changing the value based on the visibility
status of xmin. So it seems that we still don't have a patch where the
value of a variable called lp_valid corresponds to whether or not the
L.P. is valid.

--
Robert Haas
EDB: http://www.enterprisedb.com