MULTICS TECHNICAL BULLETIN                                MTB-641

To:       MTB Distribution

From:     Paul W. Benjamin <Benjamin @ System-M>,
          Ronald B. Harvey <Harvey @ System-M>,
          L. A. Dupuis <Dupuis @ System-M>

Date:     12/20/83

Subject:  The View Manager Facility

Forum Meeting:  >udd>m>mtgs>End_User_Data_Access (euda)

ABSTRACT

This paper  describes the View Manager  Facility, a comprehensive
new package  for information management.  It  was first discussed
in MTB-621, An  Architecture for End User Data  Access.  It is an
overview, supported by 4 more detailed papers:

    MTB-642 The View Manager Facility:  Subroutine Interfaces
    MTB-643 The View Manager Facility:  View Master Subsystem
    MTB-644 The View Manager Facility:  SQL Parser
    MTB-645 The View Manager Facility:  Data Dictionary Interface

This set of papers represents a major portion of the workplan for
the End User Support Unit for the  next 2 years, with MR12 as the
target release.

_________________________________________________________________

Multics  Project  internal  working  documentation.   Not  to  be
reproduced or distributed outside the Multics Project.


MTB-641                                              View Manager

Background

This  proposal  emerged in  early to  mid-1983 from  a discussion
initiated  by  Roger  Lackey   about  something  called  a  'join
manager'.  The  idea was that  an intermediary could  sit between
MRDS  and the  user that could  request retrievals  and do joins.
The  retrievals  could be  from separate  databases or  even from
non-MRDS  tables.  An  informal group was  established to discuss
the idea  further.  That group  soon determined that  another key
idea in the design should be  the replacement of LINUS with a new
end-user facility, and that a  new selection languange more along
the  lines  of  the  IBM (and  potential  industry  standard) SQL
language  should be  used.  Those discussions  continued and were
expanded  to  include  several  more  people.   The  project  was
internally  dubbed PITS,  an acronym for  "Pie in  the Sky".  The
result  of those  meetings was  MTB-621, An  Architecture for End
User  Data Access,  a crude manifesto  that listed  9 authors and
detailed the concept  as it existed at that  time.  That paper is
recommended to the reader.  The  information therein that has not
been  substantially altered  is not  reproduced here  and it will
provide an historical perspective on this topic.

After the publication of that  early paper, PITS development went
on the  back burner in  favor of MR10.2  activities (MRDS, LINUS,
GES),  but  weekly  discussion  sessions  were  held  to progress
towards a more concrete design and  to keep the idea alive.  Much
evolved during that  period, as will become apparent  as you read
these  papers.   When MR10.2  was  in place,  full  attention was
turned to  the PITS implementation,  and a lot  more changes took
place.

Introduction

These MTBs are the result of  the effort discussed above and will
serve as  a basis for  the development work to  follow.  They are
intended  as  living  documents,  will evolve  first  into design
specifications for  the planned prototype effort,  then serve the
same purpose  for the actual implementation,  and then finally be
kept  up  to  date  as  the  product  goes  into  production  and
maintenance.  It should be  understood, therefore, that these are
the first  versions of many, that  there are unanswered questions
and inconsistencies,  but that they are  being published in hopes
of obtaining feedback from the Multics development community.

As  this  is  a  complex  subject,  introducing  many  terms  and
concepts, 3 appendices are supplied to the reader, that may be of
use throughout  the reading of these  papers:  Appendix A, Module
Organization; Appendix B, View  Manager Terminology; and Appendix
C, Permissions on View Manager  Objects.  They are located at the
end of this document.


View Manager                                              MTB-641

The SQL Language

The language  used to do  relational operations (both  in the end
user  facility  and via  the programming  interfaces) is  SQL, as
defined  by  IBM  literature,  followed as  closely  as possible.
There  was lengthy  discussion within the  group on  how close to
follow  the  language,  whether  or  not  to  multicize  it.  The
consensus was not to be tempted  to "improve" on an language that
is  a)  commercially  successful  and  b)  close  to  becoming an
industry  standard.   The  language  is  described  in exhausting
detail in MTB-644.

The parser  for these statements will  be implemented using LALR.
This  raises  some  concerns  about  LALR  being  an  unsupported
product.  It  is a given in  this proposal that it  will become a
supported product, although the questions of who and when are not
resolved.

MTB-621  discussed  the  "uniform translator  proposal".   We had
proposed  a  common parser  that  could accept  not only  SQL but
existing LINUS and MRDS selection  language as well.  We have now
backed  off  from that  aspect  of it,  for reasons  discussed in
MTB-644.

The  parser  will produce  what is  being called  Canonical Query
Language.  MTB-621 discussed CQL in some detail, particularly the
Honeywell  efforts to  produce a  common language  across various
product  lines.   As  of this  writing  there is  no  such common
language.  An investigation of the GCOS  CQL proved it to be less
than optimal for  these purposes.  The option is  still left open
however,  to   converting  to  such  standards   should  they  be
established.

The View Master Subsystem

The proposed end user facility is called View Master.  It will be
an interactive subsystem  using ssu_.  There are 2  ways to enter
SQL  statements,  either at  request  level or  via  an input_sql
request.  The former approach  involves the command processor and
may  require quoting,  while the latter  allows the  user to type
into an input buffer.  The  standard ssu_ requests are available,
as are the  requests associated with the LINUS  Report Writer.  A
useful by-product  of excising the  the Report Writer  from LINUS
and making  it available to View  Master will be that  it will be
made  generally  available  as  a  standalone  offering.   It  is
described in detail in MTB-643.


MTB-641                                              View Manager

View Manager Subroutine Interfaces

At the  core of things  lives the view manager  itself (vm_).  It
accepts input in  the form of CQL and returns  data.  To do this,
it communicates with the various file managers (MRDS, Lister, the
"native  file" manager)  through Table  Interface Modules (TIMs).
These TIMs are a veneer that have a standard calling sequence and
communicate with the appropriate  specialized interfaces that the
file managers use.  In some cases (MRDS for example) the TIM will
likely  be  a  separate  module  that  calls  the  file manager's
standard  subroutine interface  (dsl_ in  this case).   In others
(native files, perhaps  Lister), the TIM may be  an integral part
of the package and the  only available subroutine interface.  The
vm_ module also does a great  deal of communicating with the data
dictionary  interface,  described  below.    The  vm_  module  is
described in MTB-642.

While the parser may be identified as a vm_ entrypoint, it should
be stressed that it is not an integral part of vm_.  It shares no
data  structures,  other than  those  passed (or  pointed  to) as
parameters,  with  vm_.  It  is  a primary  design goal  that any
parser or  other program capable  of producing CQL  is sufficient
for the View Manager's purposes.   The SQL parser just happens to
be the one that is being provided, for now.

Data Dictionary Interface

The term "Data Dictionary" is  much abused.  Everybody wants one.
Many  systems  claim  to  have one.   One  approach  to  the term
revolves around  the "meta-data" that describes  a data structure
or  database.   In that  sense, the  MRDS model  is a  crude data
dictionary.  On the other end of the spectrum is the "active data
dictionary",  an all  encompassing facility  that keeps  track of
databases, programs, data items,  etc., anything the site chooses
to  add, and  adjusts everything  involved when  something listed
therein changes.  The latter is  a hot button with customers.  It
is also far beyond our resources  to provide.  Our goal, then, is
to  provide the  facility needed for  meta-data handling, without
precluding  the   ultimate  implementation  of   such  an  active
facility.   Again, the  IBM implementation  is taken  as a model.
The meta-data is treated as  relational data, organized in tables
and manipulated  by the View  Manager software.  See  MTB-645 for
details.

Native Files, Integrity, Security

This is  an area that  was not addressed in  MTB-621.  No thought
was given to  a native file type.  The  original design goals for
the  product  included cross-database  retrievals, a  better EUF,
support  of non-MRDS  tables, etc.,  but nothing  on integrity or


View Manager                                              MTB-641

security.  There was a spirited  debate among the group about the
issues of integrity and security.   The end result was that there
was a requirement  for a level of integrity  and security no less
than that currently  offered with MRDS, and that,  as a desirable
item, much more was needed.

It soon became apparent that something must be chosen as a native
file  type.  What  would be created  when the  user typed "create
table"?  If  temporary tables were handled,  what form would they
take?   Should it  use MRDS?   This seemed  unfortunate from many
standpoints, not  the least of  which was that  a single relation
MRDS database seemed overkill for  such purposes.  Lister was out
of the  question.  How about something  new?  The discussion soon
turned to  the relation manager interface.   FAMIS was determined
to be the  vehicle of choice.  This is an  attractive route as it
deals with the integrity issue.  It also provides transaction and
rollback   capabilities,   an   integral    part   of   the   IBM
implementation.     Further,    it   is    plug-compatible   with
vfile_relmgr_ which  can be used,  as needed, in  the development
effort.

Note that this  decision about native files was  a turning point.
After starting  out to create  a package that  would tie together
MRDS, Lister,  etc., what was  under discussion had  become a new
database manager, more or less.

What  of security,  though?  Even  though FAMIS  files are ring-2
entities, if you know  the calling sequence of relation_manager_,
you can  do what you  please, disregarding any  security that was
built into the  view manager.  This is essentially  what is wrong
with MRDS today, even with  secure submodels.  The IBM model that
is  being emulated  exists in  a very  different environment, one
that   would   require  domains   to  be   implemented  properly.
Discussion of  domains with Olin Sibert  and others has convinced
the group that domains are  not something easily or appropriately
implemented on Multics.   So, what else?  What is  desired is the
ability to control access through  a data dictionary, to restrict
users  to looking  at data through  views, views  that might hide
portions of a physical file from a user while allowing the update
of  other  parts of  it.  This  has  to be  enforced by  the view
manager  software itself,  cannot be accomplished  via acls.  The
answer lies in FAMIS pseudo-rings.   Stored in the FAMIS file are
pseudo-rings.   That portion  of the  view manager  that controls
access resides in  ring-2.  FAMIS compares the rings  in the file
to the validation level of the calling program.  The view manager
files will  require the caller  to be executing  in ring-2.  Only
view  manager,  therefore,  will  be able  to  modify  the files.
Knowing  the calling  sequence of  relation_manager_ will  not be
enough.


MTB-641                                              View Manager

Security and Usage Considerations

One of the goals of the view_manager design is to be able to have
data content security for any given (native) table.  At least the
same `security' offered by MRDS is  also a goal.  This is thought
to be most easily accomplished  with views.  Properly used, views
can provide data content security and column (MRDS submodel-type)
security.  Consider the following view definitions:

          CREATE VIEW DEPT_512 AS
               SELECT * FROM EMPLOYEES WHERE DEPT_NO = 512

          CREATE VIEW EMP_VACATION AS
               SELECT NAME, DEPT_NO, MAIL_STATION, HIRE_DATE,
               VACATION_DAYS_USED FROM EMPLOYEES

If a user (say, the manager  of department 512) were given access
to the table EMPLOYEES and to the view DEPT_512, then there would
be  no way  to keep the  data in EMPLOYEES  secure.  If, however,
access is not GRANTed to the table EMPLOYEES for the manager, but
only to the view DEPT_512, then the manager can only see the data
selected by  the view.  The  access to a  given table or  view is
only relevant  when that table/view  name is specified  in a FROM
clause.  Similarly, a user given  access to only the EMP_VACATION
view would not be able to determine salary data.

To this  end, the permissions on  a view are kept  with the view,
and the  permissions on the  built-upon tables are  only relevant
when  the  view  is  created  and  when  the  permissions  on the
built-upon  tables  are  changed  to modify  the  access  for the
creator of the view.  In this way, a user does not need access to
a  table (cannot  say `FROM  FOO') in  order to  use a  view that
references it.

As  mentioned previously,  when a  view is  created the creator's
permissions on  the tables referenced  in the FROM  clause of the
definition are used to form the  permissions on the view for that
user.   No one  is then allowed  to put more  permissions on that
view than  already exist.  That  is, the user could  not change a
SELECT permission on a given column to SELECT, UPDATE permission.
Also, if the  user does not have GRANT access  to all of the base
tables involved, no permissions for other users will be permitted
on the view.  In this way,  the view actually has the permissions
on the lower  level tables.  Users have access  only if they have
access  to  the view.   If  the creator  of  the view  ever loses
permissions on  one or more  of the tables built  upon, then that
change  in   permission  should  "ripple"  and   cause  the  view
permissions to either change (as in the case of removing UPDATE),
or cause  the view to be  deleted (as when all  access to a table
disappears).  Where more permissions  are given (adding UPDATE to
a  column),  nothing  automatic  should happen.   This  will keep
accidents  from happening.   The GRANT  request will  probably be


View Manager                                              MTB-641

able  to  be  used  on the  view  to  add the  permission  if the
appropriate permission is available on the built-upon table.

Data Security and Multics Security

While  all of  this automagical access  determination sounds very
nice (and is similar to  competitors' offerings), it places a big
burden on  the view manager  and data dictionary  software.  In a
Multics  environment,  a  user  needs modify  permissions  on the
containing directory of  a file system object in  order to change
an acl  on that object.   Also, the user  must have access  to an
object  in order  to use it  - even  if, as in  the DEPT_512 case
above, the user may not know  the object (the EMPLOYEES table) is
being  used.   These Multics  realities  can cause  problems when
views,  assertions  and triggers  are  created in  an environment
where not  all users have  permissions on all of  the base tables
involved.  Also,  a user (other  than the creator)  attempting to
GRANT permissions  to other users  may be limited by  the lack of
modify permission on the containing directory of the object.

There  are  several  considerations  involved  in  determing  how
security and Multics  acls should be handled in  this product.  A
goal  is  that  it  should  be   as  close  as  possible  to  the
competitors' products.   The easiest way to  assure that this can
happen would be to put all objects in ring 2 with regular Multics
acls of (rw  sma) *.*.* and enforce the  security by checking the
appropriate data  dictionary (which would also  have similar real
Multics acls).   While this is  the easiest, it  also leaves open
the door to more accidents (any  code in ring 2 going amuck could
cause damage) or misuse of ring  2 access by users not associated
in any manner with the data involved.

In order to  limit this possibility, the current  plan is to have
tablespaces for  storing native tables.  A  tablespace would be a
ring 2 directory (ring brackets of  2,2) and an acl entry of `sma
*.*.*'.   This  will  allow  any  user  authorized  by  the  data
dictionary  to create  tables, delete  tables and  change acls on
tables.   Although it  may take some  work to  determine what acl
should  be on  a given table  for a  given user at  any time, the
problem  is  thought  to  be solvable.   This  is  still  open to
problems, however.  Today, there is a  bug in v1 forum which will
allow a user to create a meeting  in a ring 2 directory - such as
a tablespace.

Another argument against the `rw  *.*.*' scheme is that one whole
dimension of Multics  security (in fact, the one  most visible to
normal  users) would  disappear.  There  is precedence  for this,
however, as mailboxes and message segments (where daemon requests
are  placed) are  implemented in this  fashion.  It  would be the
complete  and total  responsibility of  the view  manager and the
data dictionary to always be right about who is allowed to access


MTB-641                                              View Manager

what.   This could  open the  door in  the database  software for
security holes  which don't exist  in the Multics  acl mechanism.
These holes may be difficult to detect.  Hopefully, they will not
be difficult to plug.

Another  consideration  is  that   unless  you  know  which  data
dictionary owns a file system object, it would not be possible to
determine what set of users had access (or even owns) the object.
There is a  similar concern even with specific  acls.  Given that
views (and  links to other  data dictionaries) can  place acls on
file system objects, the owner of a  table may be able to see who
has an acl entry.  But it may be impossible for a normal query of
the  owning data  dictionary to  determine WHY  those people have
access (or even differences between INSERT and UPDATE).

It  is possible  that the  view_manager should  not get  into the
business  of automagically  setting acls  at all  and that native
tables  should be  created with acl  terms for  the creator only.
Unfortunately, this would mean that  special code (or cql!)  will
have to be created to set acls on native tables, since they would
be in ring  2 (and have pseudo-acls of ring  2 as well).  This is
also  pretty  far  away  from the  competitors'  world  of access
control   and  table   management  because  the   user  would  be
responsible  for doing  everything by  hand -  including removing
acls after permissions have been revoked.

Ripples: Their Meaning and Some Implications

Ripples  are actions  that are  performed by  the data dictionary
and/or  the view  manager as a  result of some  other action.  An
example would  be that DROPping  a base table would  cause a view
that referenced  it to be  DROPped as well.  Of  course, if there
were a view built upon this  first view, then it would be dropped
as well.  With the exception of dropping a tablespace, all of the
`drop type' ripples are controlled by and completely performed by
the data  dictionary software.  In the  tablespace case, the view
manager will  probably have to drop  each table individually, and
then drop the tablespace.

A second type of ripple (and  one which can cause more headaches)
is one  in which Multics  acls need to  be changed on  an object.
These can be caused by permissions being granted on views, by the
creation of  assertions and triggers, and,  possibly, by dropping
views, triggers or views, as  well as, of course, simply revoking
permission  on  a  base table.   If  a  Multics acl  needs  to be
changed, the data dictionary will have to inform the view manager
so that it can do the deed.

There is a possiblity that synonyms will not be involved in `drop
type' ripples.  This  would allow base tables to  be switched out
from underneath  them so that a  production data dictionary could


View Manager                                              MTB-641

be switched  into `test mode'  for a period  of time.  If  a base
table looked  different from the  one that was  previously in its
place,  many   problems  could  result.   This   is  still  under
investigation.

With the concept of links  being used to enable data dictionaries
to share  tables and views,  rippling can get  very complex.  The
data   dictionary  software   may  have  to   lock  several  data
dictionaries in order to do a simple task like adding a column to
a base table.   It is thought that this can  be a potential cause
of  deadlocks.  Also,  with links and  automatic updating through
them,  a  user must  have some  way of  modifying the  other data
dictionaries (under data dictionary control, of course) that have
links  to  the data  dictionary being  used.  This  is especially
important in the case where a view on a table in the current data
dictionary is  defined in some other  data dictionary.  Since the
access  on a  view is  kept with  a view,  access changes  on the
underlying tables MUST be able to trickle to all views.

Application Programmer's Interface

The  vm_   interface  is  designed   specifically  for  subsystem
programming in a PL/1 environment.   One of its biggest customers
will be the View Master subsystem.   There will be no pains taken
to  accomodate  the  applications   programmer  or  the  non-PL/1
programmer.  There may be arguments  required that use data types
(pointers, for  example) unavailable outside of  PL/1.  There may
be  approaches  taken  that  are  confusing  or  awkward  for the
applications  programmer.  For  example, in  order to  have a SQL
statement executed via  the vm_ interface, you must,  at the very
least, call convert_sql_to_cql_  and then vm_$execute_cql.  There
are  very good  reasons for  having 2  distinct calls  here, very
necessary to the subsystems programmer,  but from a human factors
standpoint, this is less than deluxe.

It  is  recognized that  the  optimal interface  for applications
programming  is  via a  compile-time  preprocessor.  It  would be
necessary to  write such interfaces for  COBOL, Fortran, PL/1 and
perhaps Pascal.  There are simply not enough resources for this.

So, proposed, but not described in any detail in these papers, is
the  Applications Programmer's  Interface (API).   In the example
above, a single call,  for sake of argument, vm_api_$execute_sql,
would be made.  No pointers would be used, to accomodate users of
COBOL or Fortran.

Other End User Interfaces

One  of  the original  goals of  this project  was to  provide an
architecture that would facilitate the creation of additional end


MTB-641                                              View Manager

user  interfaces.   The design  is  such that  a Query-by-Example
package  that produced  CQL could  be implemented  with "relative
ease".   The information  management market  is evolving rapidly,
and  with  it, the  demand  for menu-  and  forms-driven packages
increases.  Further, the latest intelligence out of IBM indicates
that they are  going in the direction of  integrating SQL and QBE
into a single package.  It  is recognized that, to be competitive
in  the marketplace,  a foray into  the video  arena is critical.
However, given the ambitious nature of the overall design and the
resources available, a complete answer to this problem may not be
available.  Further,  the lack of a  generalized forms package on
Multics  would   make  any  forms-oriented   implementation  very
difficult.  There  will, therefore, be no  attempt to implement a
separate  QBE-style  forms  interface   in  the  MR12  timeframe.
However, the View Master  subsystem will provide a menu/prompting
video-oriented   interface   in  addition   to   the  traditional
request-oriented  interface.   This  interface   is  taken  as  a
desirable feature  or "stretch goal" and  will be investigated in
the prototype effort.

Risks, Benefits and Conclusions

The proposed implementation of the View Manager represents one of
the  largest  single undertakings  the  Multics project  has ever
attempted.   While  once  a  pioneer of  sorts  in  the  field of
relational database  technology, we have fallen  behind.  This is
an  attempt to  make a quantum  leap into the  next generation of
software, from database management to information management.  It
represents a  large and costly  investment in the  future and can
only succeed if  we are willing to commit  ourselves fully to it.
It should, therefore, not be entered lightly.

To start with the negatives, what are the risks?

     1) Chasing IBM,  in any area, can  be a risky business.
        Halfway  through  the early  planning stage  of this
        project, they announced DB2 and QMF, a departure, on
        some  levels,   from  the  earlier   things  we  had
        analyzed.   There  was much  scrambling to  make the
        design reflect the changes discovered.  It will take
        a  couple  of years  to get  this product  out.  The
        competition will not stand  still.  Perhaps a better
        approach  is to  mimic the  Japanese and  attempt an
        end-run  around  the   competition  and  investigate
        knowledge-based systems.

     2) FAMIS.   We are  relying on FAMIS  for integrity and
        for  transaction capabilities.   As of  this writing
        FAMIS  has  significant  performance  problems.   As
        FAMIS is not yet a product, our reliance on it could
        be  considered a  risk.  Further,  this design calls


View Manager                                              MTB-641

        for rollbacks  to checkpoints within  a transaction,
        something currently not implemented (albeit planned)
        in FAMIS.

     3) Effect  on  other  products.  Part  of  the scenario
        presented  is that  MRDS and  LINUS development will
        effectively be  frozen during the  MR11/MR12 period,
        except  for  those  changes necessary  for  the View
        Manager.  Our customers are not likely to appreciate
        this.

     4) Overall  complexity.   This is  a very  big project.
        MDC's track record with big projects is subpar.

Having typed  in 1) so  convincingly, it might  be appropriate to
attempt to speak to the  arguments presented.  The intent here is
not to compete  head-on with Big Blue.  The  intent is to conform
to what appears to be an  industry standard, and therein open up,
to some  extent, the arenas in  which we can compete.   So far as
still being  behind once the implementation  is complete, this is
true  to  some extent,  but  the alternative  is  standing still.
Looking  at  what  is or  may  be the  competition  (DEC, INGRES,
Oracle, et  al) rather than  IBM, the comparison might  not be so
unflattering.   The  bottom  line  is that  we  must  go  in this
direction to survive.

So, on to the benefits.

      1) Users   will   be    presented   with   a   single,
         industry-standard  selection language,  SQL, usable
         at both the end  user and programming levels.  This
         language  will offer  many capabilities  beyond the
         current  MRDS repertoire,  including cross-database
         retrievals,   outer   joins,   sorting,   triggers,
         assertions, etc.
      2) There will  be a single end-user  facility that can
         access various types of  tables, including, but not
         restricted to, MRDS databases, Lister files, native
         View Manager files and flat files.
      3) Native  files  will  be  truly secure  in  that you
         cannot circumvent the  access mechanisms by calling
         the underlying file manipulation routines directly.
      4) Checkpoint/rollback services will be available.
      5) A  platform  for   building  alternative  end  user
         facilities will be in place.
      6) By  putting  our development  resources into  a new
         product  rather  than   continually  enhancing  and
         maintaining MRDS,  we can end  up in a  much better
         posture from a maintenance standpoint.  Maintenance
         is a resource sink, and it is proven that the older
         a  product is  and the more  it is enhanced/"fixed"


MTB-641                                              View Manager

         the   incidence   of   bugs   and   expenditure  of
         maintenance dollars increase.

To summarize, the View Manager, while not an attempt to boldly go
where no man has gone before, is an attempt to bring Multics into
the 1980s in  the field of relational technology.   It is a lofty
proposal, one that  should be given the closest  degree of review
by all parties involved.  I strongly  urge the reader to read the
accompanying documents.


View Manager                                              MTB-641

                           Appendix A.

                       Module Organization

        _________________   _________________   __________
        |               |   |               |   |        |
        | Report Writer |<--|  View Master  |-->| parser |
        |_______________|   |_______________|   |________|
                                   |   |______      ^
                                   |  ________|_____|
                                   | |        |
                                   v v        v
                                _______    _______
                                |     |    |     |
                                | vm_ |<-->| DDI |
                                |_____|    |_____|
                         ________| | |__________
                        |          |            |
                        |          |            |
                        v          v            v
                     _______    _______      _______
                     |     |    |     |      |     |
                     | TIM |    | TIM |      | TIM |
                     |_____|    |_____|      |_____|
                        |          |            |
                        v          v            v
                    ________  ____________  __________
                    |      |  |          |  |        |
                    | MRDS |  | Relation |  | Lister |
                    |______|  | Manager  |  |        |
                              |__________|  |________|


View Manager                                              MTB-641

                           Appendix B.

                    View Manager Terminology

API
       See Application Programmer's Interface.

Application Programmer's Interface
       The standard interface to  the View Manager Facility (more
       specifically to vm_, the SQL parser and the DDI), designed
       for usage by applications programmers of languages such as
       PL/1, COBOL and Fortran.

assertion
       A  condition  defined on  a  table or  set of  tables that
       provides  for  data   integrity  rule  enforcement  during
       delete, insert, and update operations.

base table
       A table that exists as  a real, physical entity, i.e.  not
       a view.   A base table can  be either a native  table or a
       non-native table.

Canonical Query Language.
       The   translated   internal  representation   of   an  SQL
       statement.  Other translators can  produce CQL from inputs
       other than SQL.

checkpoint
       A  point within  a transaction  where all  changes made to
       tables since the last checkpoint,  or the beginning of the
       transaction if  there is no last  checkpoint, are applied.
       The  SQL save  statement is  used to  create a checkpoint.
       The  SQL  restore  statement  is  used  to  rollback  to a
       checkpoint.   The SQL  rollback work statement  is used to
       rollback  past   checkpoints  to  the   beginning  of  the
       transaction.

column
       the  set  of like  data  elements that  comprise  that Nth
       element of each row of a table.

commit
       The  final point  within a  transaction where  all changes
       made  to tables  are applied  and made  permanent, and the
       transaction is complete.

CQL
       See Canonical Query Language.

data dictionary


MTB-641                                              View Manager

       The collection of system tables.  (MTB-645)

Data Dictionary Interface
       That   portion   of   the  View   Manager   Facility  that
       specifically  handles and  manipulates the  system tables.
       (MTB-645)

DDI
       See Data Dictionary Interface.

link
       An entry  in a data  dictionary that refers to  an item in
       the table namespace of another data dictionary.  (MTB-645)

native table
       A table that was created by the View Manager Facility.

non-native table
       A table created outside of  the View Manager Facility, for
       example, a  Lister database or  a MRDS relation,  that has
       been 'registered' for use with the View Manager Facility.

parser
       A  software program  that accepts  some sort  of input and
       produces CQL  as output.  Often  used to refer  to the SQL
       Parser.

permissions
       The allowed actions that an  individual or group of people
       can  perform  on  various  objects.   The  objects include
       tables, views, links,  tablespaces, and data dictionaries.
       The actions  are the SQL statements  that manipulate these
       objects, and  permissions are expressed in  terms of these
       statements.

privileges
       See "permissions".

request
       A  View  Master  character  string read  by  the subsystem
       utilities  listener which  results in  the execution  of a
       procedure to perform some pre-defined operation.

ripple
       One or more actions that  happen as a result of performing
       some relational operation.  The  deletion of a table would
       have the ripple affect  of deleting all indexes associated
       with that table, all views which reference that table, all
       links  which  link to  that  table, and  all  synonyms and
       permissions defined on that table.

rollback


View Manager                                              MTB-641

       The  final point  within a  transaction where  all changes
       made to tables since the  beginning of the transaction are
       backed out and the tables return to the same state as when
       the transaction began.

row
       The collection of  one or more column values  that make up
       one record of a table.

SQL
       See Structured Query Language.

SQL Parser
       That portion of the View Manager Facility that accepts SQL
       statements as input and produces CQL as output.  (MTB-644)

statement
       An  SQL   character  string  which   specifies  that  some
       relational operation be performed on behalf of a user.

Structured Query Language.
       A collection of statements which permit data manipulation,
       data definition, and data control.  (MTB-644)

synonym
       An alternate name for a table or view.

system provided keys
       A key provided by the system to uniquely identify a row in
       a table.  This key is not user-visible.

system table
       A  table  that contains  specific  data pertaining  to the
       characteristics  of  one  or   more  tablespaces  and  its
       contents  (meta-data), that  is used  by the  View Manager
       Facility  through  the  Data  Dictionary  Interface (DDI).
       (MTB-645)

table
       A  set of  rows defined  as being made  up of  one or more
       columns, where each  row in the table has  the same number
       and type of columns.  The term  may be applied to either a
       base table or a view.

Table Interface Module
       A  standard  interface that  is called  by vm_  to perform
       various  relational operations  on different  table types.
       There will  be, for example,  a MRDS TIM, a  Lister TIM, a
       TIM for native tables, etc.  (MTB-642)

table namespace


MTB-641                                              View Manager

       The set of names of tables (native and non-native), views,
       synonyms and links in the data dictionary.  All names must
       be unique within this set of names.

tablespace
       A Multics  storage system directory  which contains tables
       created by the View Manager.  Native tables must reside in
       a tablespace.

TIM
       See Table Interface Module.

transaction
       The basic unit  of work with a well  defined beginning and
       end.   At  the  end  of  the  transaction  all  relational
       operations  which  resulted  in tables  being  changed are
       either made permanent or removed.

trigger
       A  method for  specifying the  automatic execution  of SQL
       statements upon occurence of a specified action.

view
       A virtual  table generated from  one or more  tables by an
       SQL select statement.  As a  view is a specialized form of
       table, the term table may often refer to a view.

View Manager
       Used  to  refer  to  the  set  of  software  programs that
       includes:   vm_,  View Master,  the  SQL parser,  the Data
       Dictionary  Interface,  and  the  TIM  for  native tables.
       Could conceivably  be used to include  other TIMs supplied
       by Honeywell as standard offerings.

View Master
       The  end  user  facility  to  the  View  Manager Facility.
       (MTB-643)

vm_
       That portion of the View Manager Facility that accepts CQL
       as  input  and  invokes   TIMs  to  do  relational  tasks,
       performing  relational tasks  itself, where  necessary, on
       the data returned by the TIMS.  The vm_ subroutine entries
       will  be  called  primarily  by  the  View  Master  and by
       programs requiring a more  sophisticated interface than is
       provided by the API.  (MTB-642)

unique index
       An index defined on a table that allows only unique values
       to  be  stored  in  the column  or  columns  the  index is
       composed of.


View Manager                                              MTB-641

                           Appendix C.

               Permissions on View Manager Objects

Permissions         Tables    Views

select                X         X
insert                X         X
delete                X         X
update                X         X
link                  X         X
null                  X         X
index                 X
alter                 X

Permissions         Dictionaries

access                   X
administrator            X
create_synonym           X
create_tablespace        X
create_view              X
deregister_link          X
deregister_table         X
drop_synonym             X
drop_tablespace          X
drop_view                X
register_link            X
register_table           X
null                     X

Permissions         Tablespaces

alter_tablespace         X
create_table             X
drop_table               X
null                     X

Statement                     Permissions         Permission Type

alter table . . . . . . . . . alter               table
alter tablespace  . . . . . . alter_tablespace    tablespace
comment . . . . . . . . . . . alter               table
commit work . . . . . . . . .
create assertion  . . . . . . alter (1)           table
create index  . . . . . . . . index               table
create synonym  . . . . . . . create_synonym      dictionary
create table  . . . . . . . . create_table        tablespace
create tablespace . . . . . . create_tablespace   dictionary
create trigger  . . . . . . . alter (2)           table
create view . . . . . . . . . create_view,        dictionary,


MTB-641                                              View Manager

                              select              table
delete  . . . . . . . . . . . delete              table
deregister link . . . . . . . deregister_link,    dictionary,
                              link                table, view
deregister table  . . . . . . deregister_table    dictionary
drop assertion  . . . . . . . alter (1)           table
drop index  . . . . . . . . . index               table
drop synonym  . . . . . . . . drop_synonym        dictionary
drop table  . . . . . . . . . drop_table          tablespace
drop tablespace . . . . . . . drop_tablespace     dictionary
drop trigger  . . . . . . . . alter               table
drop view . . . . . . . . . . drop_view           dictionary
grant . . . . . . . . . . . . (3)                 dictionary,
                                                  tablespace,
                                                  table, view
insert  . . . . . . . . . . . insert              table, view
lock table  . . . . . . . . . (4)                 table
register link . . . . . . . . register_link       dictionary
                              link                table, view
register table  . . . . . . . register_table      dictionary
restore . . . . . . . . . . .
revoke  . . . . . . . . . . . (5)                 dictionary,
                                                  tablespace,
                                                  table, view
rollback work . . . . . . . .
save  . . . . . . . . . . . .
select  . . . . . . . . . . . select              table, view
update  . . . . . . . . . . . update              column,
                                                  table, view

(1)  Alter permission  on the  table or  tables the  assertion is
being  made  on,  and at  least  select on  the  other referenced
tables.

(2) Necessary access to the referenced tables or views to perform
the requested operations.

(3) Must have grant option for permission being granted.

(4) Must have at least select permission on the table.

(5) Can  only revoke permissions  that you have  granted, or have
administrator permission.