Re: Patch to add a primary key using an existing index

On Fri, Nov 26, 2010 at 05:58, Steve Singer <ssinger(at)ca(dot)afilias(dot)info> wrote:
> The attached version of the patch gets your regression tests to pass.
> I'm going to mark this as ready for a committer.

I think we need more discussions about the syntax:
ALTER TABLE table_name ADD PRIMARY KEY (...) WITH (INDEX='index_name')

Issues:
* WITH (...) is designed for storage parameters. I think treating "INDEX"
as a special keyword in the way might be confusable.
* 'index_name' needs to be single-quoted, but object identifiers should
be double-quoted literals in normal cases.
* The key specifier is a duplicated option because the index has own keys.
Do we need it? It might be for safety, but redundant.
Note that the patch raises a reasonable error on conflict:
ERROR: PRIMARY KEY/UNIQUE constraint definition does not match the index

And, I found a bug:
* USING INDEX TABLESPACE clause is silently ignored, even if the index
uses another tablespace.

After all, do we need a special syntax for the functionality?
Reusing WITH (...) syntax seems to be a trouble for me.
"ADD PRIMARY KEY USING index_name" might be a candidate, but we'd
better reserve USING for non-btree PRIMARY KEY/UNIQUE indexes.
Ideas and suggestions?

--
Itagaki Takahiro