Suppose I’ve created a number of normal relations.... Then I use only
relational algebra operations on these relations to get new relational
values which are possible presented in virtual relations. Also in all
these operations I use RENAME operator on attribute names to rename
them according to some special rules. The new attribute names of the
virtual relations are complex enough (e.g. “details of bills”) and can
contain a dot inside (e.g. “bills.details”)

Is this system the relational one?

Evgeniy Grigoriev wrote:

Based (only) on your description it can't not be.

--
Roy

Thank you. I think the same.

But is a way of how I create the normal relations important? Suppose a partof system exists which creates the relations according to some external non-normar expression descrided some data. In other words this part of systemtranslates external expression into commands which create a set of normal relations. So user of such system gives to the system some non-normal data description expressions but operates with the data presented in form of normal virtual relations only.

(once again - all data are presented for user in form of normal relations)

The attribute names of these relations are complex enough and can contain adot inside (e.g. “details.amount”). These complex attribute names arebuild according to income data description.

Is this system the relational one?

On 5/4/2012 2:01 AM, Roy Hann wrote:
Be careful: "some special rules" is handwavy enough that it could mean
anything--including exposing physical details.

пÑ�тница, 4 маÑ� 2012Â*г., 23:28:21 UTC+4 пользователь Bob Badour напиÑ�ал:
I'm not sure if this rule includes exposing physical details.

Suppose there are two normal relation in a system, which contain headers and lines of some documents (don't ask me about dots in front of name of attributes... I can name the attributes as I wish).

(1) DOC ( ID , .No , .Date , ...) //ID is pkey, No is a pkey too

and

(2) DOC.Lines ( ID , .Item , .Qty , ...) //Pair ID, Item is pkey.

I can create next virtual relation JOINed (1) and (2) by ID and RENAMEd sone attributes

(1+2) 'DOC(ID, .No, .Lines.Item , .Items,Qty)

At that all combinations (header_name + attribute_name) are the same in (1),(2) and (1+2) (...almost same , e.g. combination which gives name sequence Doc.Lines.Item. I keep a name sequences unchangeble.

This is my main rule.

Is such system relational?

I'm asking about this because it seems that I use reference constructions and all these dots seem to be the "physical details" you are speaking about.But in fact it's just play with names which is performed by RENAME operation.

On Friday, August 17, 2012 8:37:23 AM UTC-7, Evgeniy Grigoriev wrote:
When you started this thread I looked at some papers of yours in the arxiv and if that's the technique you're using, you are not being relational. Youare just using multiple relation variables as a data structure and you areincluding values that act as pointers into the data structure by encoding locations in it.

To be relational a DBA must associate with every relation variable a predicate about the world that each of its present tuples makes true and each of its absent tuples makes false. It must map a tuple and a world situation toa truth value. With those pointers being there, if you tried to write predicates you would find that you have to mention relation variable names as further parameters. So you would not be giving predicates. You would be giving simultaneous equations whose unknowns were relation variable names, to be solved for valid relation variable values.

These equations would happen to form constraints for valid data structure values. But you can't use(update or query) the data structure relationally because you don't have predicates to tell you what the relation variable names in queries mean, and thus what the queries mean. Of course you can use relation operators to manipulate relation variables of your data structure just like you can use any operators to manipulate variables of any types. But that's not querying about the world.

Indeed you won't have told the user how to use (update and query) your datastructure to describe a world.

And you need special operators to manipulate your data structure reasonably.. These will be partly describable/implementable in terms of relation operators but also pointer encoding/decoding operators. Yes occasionally a special operator on some arguments can be implemented exactly as a relation operator on the same arguments but this is rare and incidental. When you are using relation operators in expressions that should be special operator callsthat is the implementation/representation/"physical" leaking out. (Ironically this lower level is itself relations. Though not relational.)

As to your particular messages in this thread, you are not describing what you are doing clearly enough for us to give you an answer. You can't just write something that is consistent with your system. You have to write something that defines it to us.

philip

пÑ�тница, 17 авгуÑ�та 2012Â*г., 23:34:50 UTC+4 пользователь (неизвеÑ�тно) напиÑ�ал:
Thank you for answer. Let me start from the end of your message.

1) About this thread. Here I just want to show that I can use such complex names to name relational attributes.

2) I don't understand your remark about "special" operator. As I understandI can use any copmplex name to denote relation and any copmplex names to denote its attributes. This complexity means nothing for how this relation can be used. From the viewpoint of the formal relational data model, all these complex names are equal to names used to denote abstract relation R(a1, a2…). So usual relational opertion can be used, nothing special.

3) About predicates and what I must associate with relation. Relational data model is abstract mathematics. Relation is a formal and meaningless data structure. If a data structure is equal to formal definition of a relation (I prefer to use the one given in "The Theory Of Relational Database" by David Maier) it is a relation. Relation doesn't have to tell something to DBAabout the world to be relation. It's not only my viewpoint (http://www.dbdebunk.blogspot.com/201...ment.html#more)

4) About my papers. I have another ones on ODBMS.ORG where the experimentalsystem is described where this simple rule is implemented fully. But in this thread I want to show that dot in the name is just a part of name. Just a character. Nothing more. (Anyway, thank you very much that you read my paper )

By the way, from my viewpoint the relation

(1+2) 'DOC(ID, .No, .Lines.Item , .Items,Qty)

is full of sence.

compdb (AT) hotmail (DOT) com wrote:


Just for the record, someone's newsreader is attributing comments to me
when posting follow-ups.

I do remember making some comment on this thread months ago but nothing
quoted here was ever written by me.

--
Roy

On Friday, August 17, 2012 5:04:58 PM UTC-7, Roy Hann wrote:
My apologies, the quote is from Bob Badour. (I repeated a misattribution to you.)
philip

On 8/17/2012 4:14 PM, Evgeniy Grigoriev wrote:
Define "complex".


Since you never spelled out what operators you are defining, whatever
you do is "special." If you are using a relational language, upon which
relational primitives is the language defined?


See Codd's 1972 paper. A relation is a predicate.


Maybe it is, and maybe it isn't. You have not given us enough
information by which to judge.