Re: Re-opening ISSUE-22 on vendor-specific SQL

David,

On 1 Jun 2011, at 14:25, David McNeil wrote:
> * the column names projected are as specified in the "column" property (or without this feature... the column names are unknown until runtime and may change on each execution (?) like a stored procedure call?)

But that already fails to address exactly the example I gave in my email: What if the mapping says there should be a column "salary", but the result only contains a column "SALARY"? Is the mapping in error, or do you use the SALARY column? Is the answer the same for every RDBMS vendor? Note that the answer is to use the SALARY column anyways in SQL 2008, because we resolved to treat unquoted columns as case insensitive (like SQL 2008).

> We don't talk about the datatypes of the columns of other logical tables so it seems we can avoid talking about datatypes in the case of vendor specific SQLQueries (?).

I think we do -- we require that certain columns be strings (if we generate IRIs an blank nodes IIRC -- I believe this was one of your issues and you proposed this resolution)

Best,
Richard


> What do you think, would it make sense to put a polished version of this in the R2RML? Or does this bleed too much into the hand waving we need to avoid?
> 
> -David

Received on Wednesday, 1 June 2011 17:09:30 UTC