Re: [HACKERS] It would be nice if this could be fixed... - Mailing list pgsql-hackers

From Tom Lane
Subject Re: [HACKERS] It would be nice if this could be fixed...
Date
Msg-id 9009.926007133@sss.pgh.pa.us
Whole thread Raw
In response to Re: [HACKERS] It would be nice if this could be fixed...  (Thomas Lockhart <lockhart@alumni.caltech.edu>)
Responses Re: [HACKERS] It would be nice if this could be fixed...  (Bruce Momjian <maillist@candle.pha.pa.us>)
List pgsql-hackers
Thomas Lockhart <lockhart@alumni.caltech.edu> writes:
>> COALESCE sql function causes postgres to CRASH!
>> httpd=> SELECT story.title, COALESCE(story.image, mfr.image)
>> httpd-> FROM story, mfr where story.category= mfr.oid;

> The problem is in combining columns from multiple tables in the
> COALESCE result.

I see at least part of the problem: flatten_tlistentry forgets to
recurse into the 'expr' part of a CaseWhen node.  There may be some
other contributing bugs in setrefs.c.

There are dozens of routines in the backend that know all about how to
walk a parse tree --- or, in some cases like this one, not quite all
about how to walk a parse tree :-(.  I just spent some time yesterday
teaching a couple of other routines about ArrayRef nodes, for example,
and I've seen way too many other bugs of exactly this ilk.

I think it'd be a good idea to try to centralize this knowledge so that
there are fewer places to change to add a new node type.  For example,
a routine that wants to examine all the Var nodes in a tree should be
able to look something like this:
if (IsA(node, Var)){    process var node;}else    standard_tree_walker(node, myself, ...);

rather than having another copy of a bunch of error-prone boilerplate.
        regards, tom lane


pgsql-hackers by date:

Previous
From: Tom Lane
Date:
Subject: Hashjoin status report
Next
From: The Hermit Hacker
Date:
Subject: Re: [HACKERS] Hashjoin status report