On Mon, Mar 9, 2015 at 5:18 PM, Jeff Janes <[email protected]> wrote:
>> It has a right-link (that's the easiest way to tell).
>
>
> Meaning that btpo_next is not zero? Should we say that in the patch in so
> many words? I think it will be hard to explain the page_items more without
> also explaining the page_stats more.
Yes. And my wording above was sloppy: By definition, a non-rightmost
page is a page that has a rightlink (it will also have a highkey, but
that's certainly not how we distinguish them).
Attached patch adds a definition of non-rightmost in parenthesis,
which I think helps.
> So if I understand this correctly, if there is a high key it is itemoffset
> 1, but to know whether itemoffset 1 is a high key I first have to look at
> btpo_next.
Yes. The logic for caring about "minus infinity" items within internal
pages is hardcoded into _bt_compare(), so it's kind of a low-level
detail, even by these standards...but it comes up with pageinspect,
since they consume an IndexTuple. Confusion ensues.
> And if there is a minus infinity, it will either be itemoffset 2 or 1,
> depending on whether there is a high key or not.
That's correct. So "data" can be "(has data), (has no data), (has
data), (has data) .... " on a given page, which could freak users out.
--
Peter Geoghegan
diff --git a/doc/src/sgml/pageinspect.sgml b/doc/src/sgml/pageinspect.sgml
index 6f51dc6..502903c 100644
--- a/doc/src/sgml/pageinspect.sgml
+++ b/doc/src/sgml/pageinspect.sgml
@@ -192,6 +192,17 @@ test=# SELECT * FROM bt_page_items('pg_cast_oid_index', 1);
7 | (0,7) | 12 | f | f | 29 27 00 00
8 | (0,8) | 12 | f | f | 2a 27 00 00
</screen>
+ In a B-tree leaf page, <structfield>ctid</> points to a heap
+ tuple. In an internal page, it points to another page in the
+ index itself, and the offset number part (the second number) of
+ the <structfield>ctid</> field is ignored. Note that the first
+ item on any non-rightmost page (any page lacking a
+ <structfield>btpo_next</> field) is the page <quote>high
+ key</quote>, meaning its <structfield>data</> serves as an upper
+ bound on all items appearing on the page. Also, on non-leaf
+ pages, the first "data item" (the first item that is not a high
+ key) is a <quote>minus infinity</quote> item, implying that its
+ <structfield>data</> field lacks any value.
</para>
</listitem>
</varlistentry>
--
Sent via pgsql-hackers mailing list ([email protected])
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers
