Question regarding scripting support in OO3

[Timothy J. Wood]

On Thursday, May 6, 2004, at 03:31  PM, Daniel Flatin wrote:

> This identifying property for rows should:
>
>    1) not be editable by the user, even in a script

   Correct; this is a read-only property.

>    2) be unique within a document

   Correct.

>    3) be assigned whenever a new row is created

   Correct.

>    4) not have any ordered relationship to the structure of the 
> document

   Correct.  They are essentially random (and then uniqued if there 
happens be a rare collision).

> Additionally, each document should not reuse deleted ids until all
> available representations have been used.  If a 32 bit unsigned
> integer is used as the id, there are 4 billion opportunities before an
> id would have to be reused.

   We considered using incrementing integers but this has bad side 
effects with source control systems (we, at least, store many OO 
documents in CVS).  If two different users add two different rows to 
the document, they'd both end up using id N+1 and would produce an XML 
document with duplicate identifiers.  We handle reading this just fine, 
but one of the rows will then lose its identifier and get reassigned a 
new id.  We also want to preserve identifiers across loads of the 
document so that you could, say, store the id of a row externally 
(possibly as a property in a script).

   So, we ended up using a string encoded random number approach.


> Transferring outline content between documents should cause the unique
> identifiers to be reassigned in the new location.  This way each
> document could be considered a separate database.

   What happens now is that the identifier is preserved if possible 
(i.e., it isn't used in the destination document already).  This makes 
it extremely likely that you can move a row from document A to document 
B and then back again.


> Given
>
>     set MyRow to first child of MyParent
>     set MyID to id of first child of MyParent
>
> these two should be equivalent:
>
>     MyRow
>     row id MyID

   What we actually ended up with is that to get an identifier based 
reference you end up doing "reference to row id MyID".

   Due to the issues noted previously, defaulting to id based references 
makes common operations fail.  This is really a trade off between 
having the ability to do what you need when you need it and making the 
common case (simple hierarchy modifications) easy.  Certainly not a 
perfect solution, but it seems that there isn't a perfect solution.

-tim