On Wed, Apr 29, 2015 at 1:30 PM, Michael Paquier
<michael(dot)paquier(at)gmail(dot)com> wrote:
> On Sun, Apr 26, 2015 at 6:05 AM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> xi(at)resolvent(dot)net writes:
>>> In some circumstances, the comment on a table constraint disappears. Here
>>> is an example:
>>
>> Hm, yeah. The problem is that ATExecAlterColumnType() rebuilds all the
>> affected indexes from scratch, and it isn't doing anything about copying
>> their comments to the new objects (either comments on the constraints, or
>> comments directly on the indexes).
>>
>> The least painful way to fix it might be to charter ATPostAlterTypeCleanup
>> to create COMMENT commands and add those to the appropriate work queue,
>> rather than complicating the data structure initially emitted by
>> ATExecAlterColumnType. But it'd still be a fair amount of new code I'm
>> afraid.
>>
>> Not planning to fix this personally, but maybe someone else would like to
>> take it up.
>
> In order to fix this, an idea would be to add a new routine in
> ruleutils.c that generates the COMMENT query string, and then call it
> directly from tablecmds.c. On master, I imagine that we could even add
> some SQL interface if there is some need.
> Thoughts?
After looking at this problem, I noticed that the test case given
above does not cover everything: primary key indexes, constraints
(CHECK for example) and indexes alone have also their comments removed
after ALTER TABLE when those objects are re-created. I finished with
the patch attached to fix everything, patch that includes a set of
regression tests covering all the code paths added.
--
Michael