Обсуждение: default namespace (schema) confusion

I have been looking forward to schemas (namespaces) for sometime.

I had not been able to decipher the schema symantics necessary for a
default schema, until I hacked the source a bit.

Now I know that the rules to get a default schema usingdb_user_namespace = truesearch_path = '$user,public'
in postgresql.conf


the user logs in aspsql testdb  testuser

but the userid is reallytestuser@testdb

if a schema named "testuser@testdb" exists
the user gets it as a default schema
otherwise the default schema is public;

and if the user is a global user
the login ispsql testdb globuser@
and the required schema is "globuser"

this use of "@" in the default schema is a bit counter intuitive


so I offer the following patch against CVS

it tries the userid as found in pg_user,
then strips off the @db stuff if possible

--- src/backend/catalog/namespace.c    Thu Oct 17 21:23:34 2002
+++ src/backend/catalog/namespace_2.c    Thu Oct 17 21:21:52 2002
@@ -1386,8 +1386,10 @@             if (HeapTupleIsValid(tuple))             {                 char       *uname;
+                char       *uname_local;                  uname = NameStr(((Form_pg_shadow) 
GETSTRUCT(tuple))->usename);                 namespaceId = 
GetSysCacheOid(NAMESPACENAME,                                              
CStringGetDatum(uname),                                              
0, 0, 0);
@@ -1396,7 +1398,25 @@                     !intMember(namespaceId, 
oidlist) &&                      
pg_namespace_aclcheck(namespaceId, userId,                                           
ACL_USAGE) == ACLCHECK_OK)
+                {                     oidlist = lappendi(oidlist, 
namespaceId);
+                }
+                else
+                {     
+                    uname_local = (char *) 
malloc(strlen(uname)); +                    strcpy 
(uname_local,uname);
+                    uname_local = 
strtok(uname_local,"@");
+                    namespaceId = 
GetSysCacheOid(NAMESPACENAME,
+                                 
CStringGetDatum(uname_local),
+                                     
0, 0, 0);
+                    if (OidIsValid(namespaceId) &&
+                        !intMember(namespaceId, 
oidlist) &&
+                         
pg_namespace_aclcheck(namespaceId, userId,
+                                          
ACL_USAGE) == ACLCHECK_OK)
+                        oidlist = 
lappendi(oidlist, namespaceId);
+                    free(uname_local);
+                }
+             }         }         else

Carl Anderson <candrsn@mindspring.com> writes:
> this use of "@" in the default schema is a bit counter intuitive
> so I offer the following patch against CVS

Hmm, this seems like a wart, but then the db_user_namespace feature
is an acknowledged wart already.

I think I'd be willing to hold still for this if it happens only when
db_user_namespace is on.  Comments anyone?

> +                    uname_local = (char *) 
> malloc(strlen(uname)); +                    strcpy 
> (uname_local,uname);

Use pstrdup, please.
        regards, tom lane

Tom Lane wrote:
> Carl Anderson <candrsn@mindspring.com> writes:
> > this use of "@" in the default schema is a bit counter intuitive
> > so I offer the following patch against CVS
> 
> Hmm, this seems like a wart, but then the db_user_namespace feature
> is an acknowledged wart already.
> 
> I think I'd be willing to hold still for this if it happens only when
> db_user_namespace is on.  Comments anyone?

I dislike double-testing the username in schema areas but not other
places.  Seems if we do it, we should do it consistently for all
username references, or not at all.

--  Bruce Momjian                        |  http://candle.pha.pa.us pgman@candle.pha.pa.us               |  (610)
359-1001+  If your life is a hard drive,     |  13 Roberts Road +  Christ can be your backup.        |  Newtown Square,
Pennsylvania19073
 

Bruce Momjian <pgman@candle.pha.pa.us> writes:
> I dislike double-testing the username in schema areas but not other
> places.  Seems if we do it, we should do it consistently for all
> username references, or not at all.

What other places do we have an explicit dependence on the username?
        regards, tom lane

Tom Lane wrote:
> Bruce Momjian <pgman@candle.pha.pa.us> writes:
> > I dislike double-testing the username in schema areas but not other
> > places.  Seems if we do it, we should do it consistently for all
> > username references, or not at all.
> 
> What other places do we have an explicit dependence on the username?

Uh, what about CREATE/ALTER/DROP USER, and pg_hba.conf.  Of course,
those are more admin, but isn't the schema also sort of admin too?

I am also concerned about opening cases where a CURRENT_USER test
doesn't match the user's schema.

--  Bruce Momjian                        |  http://candle.pha.pa.us pgman@candle.pha.pa.us               |  (610)
359-1001+  If your life is a hard drive,     |  13 Roberts Road +  Christ can be your backup.        |  Newtown Square,
Pennsylvania19073