table_name syntax

[Dave Morris]

On 2015-05-13 10:52, Mark Taylor wrote:
> 
> On Tue, 28 Apr 2015, Patrick Dowler wrote:
> 
>> In Sybase, the convention semes to be that schema support is sketchy 
>> and
>> people generally use the catalog as both allocation and namespace. So, 
>> if
>> someone implements a TAP service on Sybase, they would quite likely 
>> want to
>> expose tablenames line foo..bar (implicit schema between the two 
>> dots). Yeah,
>> that's all kinds of ugly but this is what people do.
> 
> I don't think that foo..bar is a legal ADQL <table_name>.
> 
> (I'm not 100% certain, because it depends what the symbol "..." means
> in the BNF from Appendix A of ADQL 2.0, i.e. whether it works
> like a regex "*" or "+"; I don't see that symbol discussed in e.g.
> the wikipedia BNF page.  However, if foo..bar is legal, then
> zero-length <identifier>s must be legal too, which sounds unlikely
> and bad in equal measures.)
> 

I would not expect foo..bar to be valid.

However, I don't think there is anything in the current specification 
that explicitly states whether zero-length identifiers are allowed or 
not.

Internally we run SQLServer, which like Sybase, does use foo..bar

However, our ADQL service hides all the implementation details, 
providing an abstract layer which publishes our tables as foo.bar rather 
than foo..bar or foo.dbo.bar which is what they are underneath.

We would not expose foo..bar as part of a TAP service in TAP_SCHEMA or 
in VOSI.

Currently our HTML web form does allow foo..bar, but it is provided as 
an undocumented feature purely to enable our researchers to use existing 
queries they have in their notes and in published papers.

We would not recommend including foo..bar as part of the ADQL standard.



Dave

--------
Dave Morris
Senior Troll
Wide Field Astronomy Unit
Institute for Astronomy
University of Edinburgh
--------