[pgpool-hackers: 3757] Re: Using volatile property instead of black_function_list

[Tatsuo Ishii]

Hi Hou, Zhijie,

> However, I found a bug in my old patch which will affect your patch.
> 
> Currently, the sql is formatted in function function_volatile_property
> and will be formatted in pool_search_relcache again.
> 
> If arg3 or arg2 contains '%s'(such as : create or REPLACE  function "%s"(integer)), 
> the second format will break the sql.

True. But same thing can be said to all other callers of
pool_search_relcache. no?

> So I try to add a formatted flag in sql comment to avoid this.
> A little change in your patch.

Unfortunately this will not work. If you do not allow snprintf in
pool_search_relcache, it could use wrong SQL. For example we have
following query when we call pool_create_relcache:

/*FORMATTED*/SELECT count(*) FROM pg_catalog.pg_proc AS p, pg_catalog.pg_namespace
AS n WHERE p.proname = '%s' AND n.oid = p.pronamespace AND n.nspname
= 'public' AND p.provolatile = 'v"

Next pool_search_relcache() may be called with
public.f1. pool_search_relcache() will find a cache entry for
public.%s and returns the result for public.f1.

Also my patch was wrong. Suppose we have below at the when relcache is
first created:

SELECT count(*) FROM pg_catalog.pg_proc AS p, pg_catalog.pg_namespace
AS n WHERE p.proname = 'f1' AND n.oid = p.pronamespace AND n.nspname =
'public' AND p.provolatile = 'v"

Next pool_search_relcache() may be called with
public.f2. pool_search_relcache() will find a cache entry for
public.f1 and returns the result for public.f2.

Let me think how to solve these problems.

Best regards,
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese:http://www.sraoss.co.jp