Re: Rel 7.0beta5: view on table* crashes backend

This has been fixed in 7.0.


> "Oliver Elphick" <olly@lfix.co.uk> writes:
> > The following commands invariably crash the backend:
> >   morejunk=# create table test_t(id int4);
> >   CREATE
> >   morejunk=# create table test_d() inherits(test_t);
> >   CREATE
> >   morejunk=# create view test_v as select * from test_t*;
> >   CREATE 135069 1
> >   morejunk=# select * from test_v;
> >   pqReadData() -- backend closed the channel unexpectedly.
> >           This probably means the backend terminated abnormally
> >           before or while processing the request.
> >   The connection to the server was lost. Attempting reset: Failed.
> 
> The problem appears to be in this routine in optimizer/prep/prepunion.c:
> 
> static RangeTblEntry *
> new_rangetable_entry(Oid new_relid, RangeTblEntry *old_entry)
> {
>     RangeTblEntry *new_entry = copyObject(old_entry);
> 
>     /* ??? someone tell me what the following is doing! - ay 11/94 */
>     if (!strcmp(new_entry->eref->relname, "*CURRENT*") ||
>         !strcmp(new_entry->eref->relname, "*NEW*"))
>         new_entry->ref->relname = get_rel_name(new_relid);
>     else
>         new_entry->relname = get_rel_name(new_relid);
> 
>     new_entry->relid = new_relid;
>     return new_entry;
> }
> 
> It bombs on a RangeTblEntry that's been inserted by rule expansion,
> because the entry's eref is NULL.  The reason it's NULL is that we
> deliberately decided not to save eref data in stored rule strings.
> So RTEs coming from rule expansion have NULL eref fields.
> 
> This routine is the only one outside the parser that makes use of
> eref as opposed to ref in RTEs.  I am thinking we can fix the problem
> by making it look at ref->relname instead of eref->relname.  But I'd
> like to get Thomas to look at the issue, since according to the CVS
> logs he was the one who changed it from ref->relname to eref->relname
> in the first place.  Thomas, is there a semantic difference between
> ref->relname and eref->relname?
> 
> Wait a minute ... now that I look at it, it's arguable that the entire
> chunk of code is misguided in the first place.  What we're doing is
> making an RTE point at some inheritance child of a table that was named
> (with *) in the original query.  It seems to me that we should leave the
> ref name alone and substitute for the true relation name regardless of
> whether the ref name is *CURRENT* or not.  That is, if you write
>     ... FROM foo* ref
> then the reference name ref applies to all of foo's children as well as
> foo.  I don't see why it would be different inside a view.  In other
> words the code probably should just be
> 
>     new_entry->relname = get_rel_name(new_relid);
>     new_entry->relid = new_relid;
>     return new_entry;
> 
> Jan, Thomas, any comments?
> 
>             regards, tom lane
> 


--  Bruce Momjian                        |  http://www.op.net/~candle pgman@candle.pha.pa.us               |  (610)
853-3000+  If your life is a hard drive,     |  830 Blythe Avenue +  Christ can be your backup.        |  Drexel Hill,
Pennsylvania19026