Alan Hodgson wrote:
> On Tuesday 27 March 2007 17:45, Steve Singer <ssinger_pg at sympatico.ca> 
> wrote:
>   
>> I don't know what workarounds are going to be options for you compared
>> with running a patched version of slony but if you
>> need a fix right away you could consider patching your version of slony.
>>
>> The patched attached to
>> http://lists.slony.info/pipermail/slony1-patches/2007-March/000005.html
>>
>> should address the problem your describing.  I assume it will be
>> included with the next 1.2.x release.
>>     
>
> The patch referenced above, with the very small fix I posted earlier 
> to -patches, completely fixes this problem.  Hopefully it will get reviewed 
> and make its way into 1.2.10.
>
> As I suspected, my problem was due to ddlScript_prepare_int not running 
> alterTableRestore against tables (because it checked for set membership), 
> but ddlScript_complete_int trying to run alterTableForReplication (because 
> it doesn't). However, I misunderstood what should be the normal contents of 
> sl_table. As it turns out, there is no need for changes to the SQL 
> functions if used with Steve's patch. There are some checks in 
> ddlScript_prepare_int that are made redundant and could be removed, but 
> they no longer hurt anything.
>
> Steve - thank you very much for your help.
>
>   
I have committed a (significantly altered) version of this patch, along
with a change to the DDL test script to exercise this, so we don't miss
this in future.

If people can take a look at this, it would be greatly appreciated.

I made one "big, manifest assumption" for the "EXECUTE ONLY ON" case,
namely that we alter back *all* tables that are involved in replication
on the local node.  (And there's already a bug in this for the 'restore
to replication' code...  Another patch goes in momentarily...)