[Glass] Queries in Gemstone ...

Dale Henrichs via Glass glass at lists.gemtalksystems.com
Sun Feb 8 13:32:15 PST 2015


Okay, I was looking for examples of the actual api, I think I understand 
what you want, but I was looking for suggestions as to how you would 
express what you want.

If I look at your original message, again, I guess you were actually 
making a feature request asking that we support a more complicated query 
syntax? In other words, you were wondering if we would expand the query 
syntax to support something like this:

    select x from y ordered by x.q where x.z = .... limit n

So are you thinking about something along the lines of OQL (Chapter 4 of 
The Object Data Standard book[1]), SQL++[2], one of the other languages 
mentioned in the Object-Oriented Query Languages survey[3] or something 

I think I understand the theoretical advantage of having an 
Object-Oriented Query Language, but from a practical perspective, I am 
wondering about the incremental advantage of having a query language 
beyond being able to write simple Smalltalk code to augment the existing 
GemStone query language.

It seems to me that often the right answer is to explicitly use a 
completely different object representation than to provide a (limited) 
palette of standard data structures accessed via a standard query language.

The GemStone query language basically replaces the select/reject/detect 
expression by providing an underlying representation that is more 
efficient for doing equality/range comparisons.

The "ordered by" clause implies that a sort block be used and given an 
arbitrary "where" clause, the "ordered by" cannot be optimized by 
pre-calculation (the trick applied by GemStone indexes)...so there's not 
much that we can do here ...

I think there might be areas where we could "do better", for example if 
you included a "in range of s to t by u" in a query:

   where x.z = .... in range of s to t by u

I can see where some significant optimizations could be made, but I 
don't know if such corner cases exist in the wild.

Soooo, in a nutshell, I am wondering about the actual opportunity for 
being able to "do better than a good programmer" by expanding the query 
language ...


[1] http://www.odbms.org/odmg-standard/odmg-book/odmg-2-0-book-extract/
[3] http://www.cs.cornell.edu/home/ulfar/oodbms/ooqlang.html

On 2/8/15 12:14 PM, itlists at schrievkrom.de wrote:
> Well, the simplest stuff would be: execute a query and give me back the
> results - if indices can be used: use them (coding perspective).
> If its too slow the programmer must have some hints from the query
> executing machine like it is on the relational database - then the
> programmer can add indices as wanted to improve the situation
> (performance perspective).
> That's the idea I would like to see ... and which is not trivial.
> Marten
> Am 08.02.2015 um 20:22 schrieb Dale Henrichs:
>> On 2/8/15 10:21 AM, itlists at schrievkrom.de wrote:
>>> Yes, I know ... but for a simple question this seems to be lots of work
>>> and with the introduction of the new query manager in 3.2.3 I hope that
>>> Gemstone does more work possible in this direction ...
>> Marten,
>> Since you appear to be disappointed with my "put an index on x.z, query
>> on x.z = ??? and then sort the result set on x.q" answer, I am curious
>> what answer you were looking for?
>> You used the word "efficient" in your question and I assumed that you
>> meant "efficient" from a performance perspective, but perhaps you meant
>> "efficient" from a coding perspective?
>> If so, do you have suggestions for an API that would satisfy you?
>> Dale

More information about the Glass mailing list