Hi,
8b38a538c0aa5a432dacd90f10805dc667a3d1a0 changed things so that all
columns are checked for NOT NULL ness. Which is fine in general, but it
currently does make it impossible to have a varlena column (except
OID/INT2 vector...) as a key column in syscache. Even if the column is
factually NOT NUL.
The rule for determining attribute's NOT NULL setting in bootstrap is:/* * Mark as "not null" if type is fixed-width
andprior columns are too. * This corresponds to case where column can be accessed directly via C * struct declaration.
** oidvector and int2vector are also treated as not-nullable, even though * they are no longer fixed-width. */
#define MARKNOTNULL(att) \((att)->attlen > 0 || \ (att)->atttypid == OIDVECTOROID || \ (att)->atttypid ==
INT2VECTOROID)
if (MARKNOTNULL(attrtypes[attnum])){ int i;
for (i = 0; i < attnum; i++) { if (!MARKNOTNULL(attrtypes[i])) break; } if (i == attnum)
attrtypes[attnum]->attnotnull = true;}
(the rule is also encoded in genbki.pl)
Now, you can argue that it's a folly to use the syscache code to access
something via a text column (which is what I want to do). I don't agree,
but even if you're of that position, it seems worthwhile to mark further
catalog columns as NOT NULL in the catalog if that's what the code
expects?
E.g. pg_(sh)?seclabel.provider should certainly be NOT
NULL. pg_extension.extversion as well. There's a couple more I think.
So, how about providing bootstrap infrastructure for marking columns as
NOT NULL?
Greetings,
Andres Freund
-- Andres Freund http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training &
Services