On 20.01.2019 2:45, Alexander Korotkov wrote:

3) How do we calculate the "id" property returned by keyvalue()
function?  It's not documented.  Even presence of "id" columns isn't
documented.  Standard stands that it's implementation-depended
indetifier of object holding key-value pair.  The way of its
calculation is also not clear from the code.  Why do we need constant
of 10000000000?

                 id = jb->type != jbvBinary ? 0 :
                     (int64)((char *) jb->val.binary.data -
                             (char *) cxt->baseObject.jbc);
                 id += (int64) cxt->baseObject.id * INT64CONST(10000000000);

I decided to construct object id from the two parts: base object id and its
binary offset in its base object's jsonb:

object_id = 10000000000 * base_object_id + object_offset_in_base_object


10000000000 (10^10) -- is a first round decimal number greater than 2^32
(maximal offset in jsonb).  Decimal multiplier is used here to improve the
readability of identifiers.


Base object is usually a root object of the path: context item '$' or path
variable '$var', literals can't produce objects for now.  But if the path
contains generated objects (.keyvalue() itself, for example), then they become
base object for the subsequent .keyvalue().  See example:

'$.a.b.keyvalue().value.keyvalue()' :
 - base for the first .keyvalue() is '$'
 - base for the second .keyvalue() is '$.a.b.keyvalue()'


Id of '$' is 0.
Id of '$var' is its ordinal (positive) number in the list of variables.
Ids for generated objects are assigned using global counter
'JsonPathExecContext.generatedObjectId' starting from 'number_of_vars + 1'.


Corresponding comments will be added in the upcoming version of the patches.

--
Nikita Glukhov
Postgres Professional: http://www.postgrespro.com
The Russian Postgres Company

Reply via email to