Re: constraints and sql92 information_schema compliance

On Sun, 26 Feb 2006, Clark C. Evans wrote:

> Stephen,
>
> So, a quick re-cap of the questions/concerns I had:
>
>   * Making the default constraint names include the table
>
>     -> This was implemented in 8.x, thank you!
>
>   * Forbidding the creation of a foreign key constraint where
>     the column list for the referenced table doesn't *exactly*
>     match a canidate key on that table.
>
>     -> I think you've agreed to something like this, or am
>        I mis-understanding?

Well, SQL03 requires it to match exactly (I haven't checked 99). SQL92
explicitly requires us to support not matching exactly and we can't
really remove it for some amount of time due to compatibility. About the
best we're likely to be able to do is change pg_dump to dump it in the 03
order and possibly give an optional way to turn on an exact check (default
off) for the next version, probably changing the default 1 or 2 versions
after that.

Personally, I disagree with the 03 requirement and think that it's more an
example of them misdesigning the information schema, but we should
probably move in that direction for compatibility with more recent
versions of spec.

>   * Issue a warning when creating a constraint who's name is
>     not unique within its (the constraint's) schema.
>
>     -> This request seems to have gotten lost in the
>        vigorous discussion ;)

I don't have a problem with it (once, I argued for following the spec
constraint on this way back when), however I think this was proposed and
rejected before as excess noise.  You might want to look back through the
archives.