Notes on metaschema additions to replace hcg_struct data structures.

RJLRef:  $PH/COOL-BDE/metadataExtensions060822.2300 - RJL060822

Invarant  metadata is now pre-computed by chgen -mweta. For bde, 
this metadata is saved by chgen to bde/schema/94sde_schema.msdat, 
but not yet utilized.

The SV-->TT-->TA metaschema and its instantiation for a given 
.sch schemaafile (e.g. bde/schema/94sbde_schema.sch) 
can be augmented with static (but not invariant) properties 
of bde tables if this data is dynamically updated by the CRUD 
operations that are auto-generated by chgen/gencpp 
in bde/pr_util_*log).

The 'horrible' access expression in get_rcount below shows that  
rcount (and maxkey) is currently located in what I call hcg_structs):

In dprint.h:
------------------
/***********************************************************************/
/* Macro get_rcount returns the number of rows currently loaded in the */
/* specified table, iff in the specified view/version). [was pr_rcount]*/
/* If tbl is not a ttabbr in the view, 0 is returned.                   */
/* Side-effects: find_*_idx() updates [hcg_?]tbl_idx  and hcg_view_id()x.*/
/***********************************************************************/
/* Added '&&find_tbl_idx(#tbl)' to condition. - RJL060806 */
/* find_tbl_idx  updates hcg_tbl_idx, not tbl##_idx, or returns 0 -RJL060809 */
#define get_rcount(viewname,tbl) \
  ((find_view_idx(viewname) && (find_tbl_idx(#tbl)) ) ? \
     hcg_ts_list[hcg_tbl_idx].ts_list[hcg_view_list.view_list[hcg_view_idx].version_list[hcg_tbl_idx]].rcount : \
        0)
-------------------

A new rcount data member in table TT can be initialized to 0 
in metaschema file bdeischema.msdat by chgen or gencpp.
Whenever pr_add_row allocates a new table-row it requires a new pkey.
This is created by pr_gen_key and maxkey and rcount are incremented. 
(Since pkeys are never re-used, pr_delete() will decrement rcount 
but not maxkey.)

Table TT has an index tbl_idx map(tbl) for each table abbrev XX.
Since TTrows have fixed length and MAXROWS is known,
they can be are pre-allocated in an array (TBD). Then
a lookup table indexed by ttidx can map it to its address TTcurr:
	TTcurr = TTloc[tt_idx] = TTloc[0] + (tt_idx * (sizeof(struct TT)) 
Another map can produce bl_idzx from the table abbrev:
	ttidx = TTmap(ttabbrev);
Then get_rcount(viewname,XX) would be simply expressed 
and efficiently computed without the find_*_idx searches as 

int get_rcount(char* viewname, char* tbl)
{
  ttidx = TTmap(ttabbrev);
  TTcurr = TTbegin + (sizeof(struct TT))*(ttidx));
  assert((tblInView(ttabbrev, viewname) ;
  return (TTcurr->rcount);
}
//tblInView checks the associative relation TV between TT and VV
//to see if table ttabbrev is contained in view viewname.
//It can be used to return an error if tbl is not in viewname.
//It can use a pre-computed view_idx = VVmap(viewname).

Background:  Moving table statistics to tables TT and TV 
from the hcg_structsxa was always a (deferred) TBD. 

Two over-complex features made this change more complex 
than it needs to be now:
(1) Version numbers were mantained in pkeys and manipulated
according to constraints in views. 
(2) pr_init() guaranteed that pkeys are unique over a 
given list of .dat files.

I want to remmove version numbers from pkeys and remove pr_init()'s
scan of a list of files. pr_load can update maxkey and rcount
for the single-file design DB that it loads:

(1) Without version numbers in pkeys, CVS can handle version control.
(2) Without pr_init(), each .dat file is an independent database;
(there are no constraints on pkey overlap between two .dat files).

(Namespaces add a top level of table type partitioning,
There are no conflicts among table types in different namespaces.)