Re: Alternatives to a unique indexes with NULL

Hi John

On 17/01/15 12:39, John McKown wrote:
> ​I read the above. As I understand it, you can have a unique index on 
> a column which is NULL-able. That will guarantee that all the non-NULL 
> values are unique. What it will not guarantee is that there will be at 
> most one NULL value in the indexed column. Are you saying that what 
> you want is a column with a unique index where you cannot have two or 
> more rows with NULL in the indexed column?
That's correct - the application reads in a very old-format of 
fixed-length file and, if any field in there is just a single space, 
replaces it with a NULL, since a single space implies a null. However, 
only one of the records needs a constraint on one of these fields.
> ​If so, then you will need to have a value to indicate the equivalent 
> of NULL. Personally, I use a zero length string "" instead of a single 
> blank ' '. This is value since you say this column is a "character 
> varying(1)". Which seems a bit strange to me, but I don't know your 
> application.
OK, that makes sense and it was more-or-less along the lines of what I 
expected.  I like the idea of a zero-length string versus a single 
space, so I'll go implement that.

I believe the column type a Rails-ism, which creates 'string' fields 
with a length constraint of 1 as 'character varying(1)'.  Probably not 
ideal, but there's usually a trade-off somewhere.

Thanks very much for your help and quick response!


Peter