Re: patch for type privileges

On lör, 2011-12-10 at 16:16 +0100, Yeb Havinga wrote:
> * ExecGrant_type() prevents 'grant usage on domain' on a type, but the
> converse is possible.
>
> postgres=# create domain myint as int2;
> CREATE DOMAIN
> postgres=# grant usage on type myint to public;
> GRANT

This is the same as how we handle types vs. domains elsewhere. For
example, you can use DROP TYPE to drop a domain, but you can't use DROP
DOMAIN to drop a type.

> * Cannot restrict access to array types. After revoking usage from the
> element type, the error is perhaps a bit misleading. (smallint[] vs
> smallint)
>
> postgres=> create table a (a int2[]);
> ERROR: permission denied for type smallint[]

OK, that error message should be improved.

> * The patch adds the following text explaining the USAGE privilege on types.
>
> For types and domains, this privilege allow the use of the type or
> domain in the definition of tables, functions, and other schema objects.
>
> Since other paragraphs in USAGE use the word 'creation' instead of
> 'definition', I believe here the word 'creation' should be used too.
> IMHO it would also be good to describe what the USAGE privilege is not,
> but might be expected since it is such a generic term. USAGE on type:
> use of the type while creating new dependencies to the type, not usage
> in the sense of instantiating values of the type. If there are existing
> dependencies, revoking usage privileges will not return any warning and
> the dependencies still exist. Also other kinds of exceptions could be
> noted, such as the exception for array types and casts. The example you
> gave in the top mail about why restricting access to types can be
> useful, such as preventing that owners are prevented changing their
> types because others have 'blocked' them by their usage, is something
> that could also help readers of the documentation understand why
> privileges on types are useful for them (or not).

Good suggestions. I'll review the text.

> * The information schema view 'attributes' has this additional condition:
> AND (pg_has_role(t.typowner, 'USAGE')
> OR has_type_privilege(t.oid, 'USAGE'));
>
> What happens is that attributes in a composite type are shown, or not,
> if the current user has USAGE rights. The strange thing here, is that
> the attribute in the type being show or not, doesn't match being able to
> use it (in the creation of e.g. a table).

Yeah, that's a bug. That should be something like

AND (pg_has_role(c.relowner, 'USAGE')
OR has_type_privilege(c.reltype, 'USAGE'));

I'll produce a new patch for these issues in a bit.