  3                       The 4GL Problem and Solution.   4                     3GL Solution by Bart Z. Lederman/                          System Resources Corp.    B     I volunteered to do a 3GL solution to the problem to provide aA  comparison between traditional programming languages and the 4GL ?  solutions.  Although some people requested a COBOL solution as B  being the 'typical' traditional business solution language (which;  point could be disputed), and while I might have preferred =  FORTRAN for this, it happens that where I work now only C is @  available (except for Macro-32, which is not a good choice) and'  so that is the language I plan to use.   B     If I had a forms package such as FMS or TDMS available I would8  probably have used it, because the screen painting, and:  especially the requirement to scroll through data in both=  directions, takes a lot of programming effort: but again, it ?  happens that none are available on the system I use, and I did B  not want to have to develop an application remotely.  I therefore?  used SMG routines (which are supplied with VMS) for the screen @  management functions.  This greatly simplified the presentation@  and selection from menus, the bi-directional scrolling regions,B  and integration of help messages: but it may place some limits on6  browsing in the way the original problem was defined.  ?     I do have a choice between using an Rdb relational database ?  for the data, or using RMS files.  I ended up using Rdb, which ?  made some of the retrievals easier, and would ultimately allow B  database triggers to do update the content of some MASTER recordsA  from the DETAIL records.  However, nothing I did so far would be 3  difficult to migrate to RMS or other file systems.   A     An advantage to using SMG and a 3GL is that once the  program ?  is compiled and linked, it can be run on any VMS system. Since ?  permission to use Run Time Only Rdb is now included with a VMS =  license, using it requires that a media kit be purchased and A  installed but no extra licenses should have to be purchased: and B  this is where the greatest cost is.  Conversion to RMS would even@  eliminate this.  Not having to buy a run-time license for a 4GL>  can be a significant cost savings, and would simplify program?  distribution, if the application is intended to run on a large B  number of systems.  If it has to be distributed to a large number@  of users the savings in license fees can pay for the additional  development time.  <     Another advantage to a 3GL is that there is nearly total@  freedom to design the application the way one wants.  With 4GLs;  one is often forced into a particular style of application $  development or operation or "feel".  A     The HELP which was built into the application uses a standard A  VMS HELP library.  This means that HELP can be accessed with DCL <  and from other applications as well as from this particular?  application, only a normal text editor is needed to create and A  maintain the HELP text source, and that HELP can be changed (for >  example, translated into another language) without making anyA  changes to the application source or executable image.  At least @  some 4GLs require that HELP be built into the application using<  only their special format, the HELP cannot be accessed from?  outside of the 4GL application or from other applications, and =  changing HELP requires changing the application with the 4GL   development tools.   @     There is also the advantage that the project could have beenB  split up into modules so that more than one programmer could have@  worked on it at a time, and there could have been an additional@  person working on the HELP file.  At least some 4GLs allow onlyA  one person at a time to edit an application, and the application ?  cannot be divided into 'pieces' which can be worked on by more   than one person.   9     Most of the Browse phase was implemented.  All of the >  selection types are allowed, and the no record and one recordA  selected cases are correct.  However, I mis-read the description ?  and the display of multiple records is not the way the problem >  originally specified: only the MASTER part of the information>  (with the number of donations, and total dollar amount of allB  donations) is displayed for browsing.  I discovered this part way@  through the problem and decided to try to finish as much of theA  problem as I could rather than try to go back and fix that.  The @  multiple records are displayed in the specified sort order, and4  the user can browse forwards and backwards on them.  ?     Once a MASTER record is selected it is displayed along with A  all DETAIL records: the DETAIL records are in a scrolling region ?  so that if there are more than will fit on the screen the user @  can browse forwards and backwards, and they are in correct date  order.   =     I did not have time to get to the processing phase or the @  report phase.  I estimate the report would take between one and  two hours.   >     1. Data was loaded into an Rdb database using Datatrieve. B  Datatrieve was also used later to add the count of detail records:  and total of donations to the master record.  Very little?  database design or optimization was done other than allocating #  storage areas for various indices.   A     2. The count of DETAIL records and the total of donations for A  each donor were made real fields in the MASTER record, because I >  thought they would be needed this way.  They do simplify some?  processing, and can act as a validation check on the database.   >     3. The date of donations was stored as a 64-bit binary VMS  date data type.  ;     4. The fund code was retained in the DETAIL records.  A A  database view was defined which linked the FUNDCODE table to the B  DETAIL table to make it easy to display the fund abbreviation and>  full name during the browse phase, and I intended to use this0  table to simplify storing or modifying records.  8     5. Adding and modifying MASTER records was not done.  @     6. The MASTER table was indexed on ID_NUMBER, LAST_NAME, and?  LAST_NAME_SOUNDEX, because records will be retrieved on all of A  these fields as specified for the problem.  The DETAIL table was @  indexed on ID_NUMBER (because records are matched to the MASTER:  record this way) and FUND_CODE (which may not actually be>  needed).  The FUNDCODE table was indexed on FUND_CODE because<  they are retrieved often when matching with DETAIL records,A  though the table is so small that indexing might not actually be   necessary.    0     7. Modifying or adding DETAILS was not done.             0                   Session Reporting Information:  1     1. The experience of the developer with 3GLs:   =     I have over 15 years experience programming in 3GLs, with @  about 3 years of programming in C and about 1 year of accessingB  Rdb from within programs.  I had only about 1 month of experienceA  using SMG and ran into a few new problems during the 4GL problem 
  solution.    >     2. What background knowledge of the user has been assumed:  >     The intention was to provide an application for a user who;  would already know the intent of the application (donation B  management), but who would have little knowledge of computers and?  would receive an hour or so of training with the application.  B  This does not include training the user to log into a VAX system.  >     3. What training is, or would be required, for the user to=  operate the browse and processing phases of the application.   ?      The browse phase should require less than one half hour of    training.   <     4. What software is required to run the application (all  licensed products):  A     To run the product currently requires only VAX/VMS and an Rdb <  Run-time kit.  The license for Rdb-RTO is bundled with VMS.  >     5. What part or parts of the application is done in a 3GL:  >     All of the application is done in C except for the SoundexA  algorithm, of which I already had a working routine available in 
  Macro-32.  A     6. What part or parts of the application could not be done in   their 4GL:   (      Everything is possible using a 3GL.  ?     7. How long it took to develop each part of the application   B     Starting with a working database, it took about 9 hours to get?  to the point where I could select MASTER records by all of the @  specified methods.  This included linking some sample HELP intoA  the first two menus of the application, and some documentation.  A  I also had to fix a few SMG problems I ran into during this time @  which I probably would have known about if I had been using SMG  longer.  <     Adding retrieval of matching DETAIL records took about 4?  hours: at this point most SMG problems were solved and most of <  the work was in testing for error conditions (not trying to4  retrieve details when there are none, for example).  >     A major problem I encountered is that I had to work on the@  problems when bits of time were available: mostly in 1/4 or 1/2<  hour segments.  The only time I could work on it for longer?  stretches is when I dialed into our system from home, and that @  meant that I had to work at 2400 Baud, and with no manuals.  It=  also required working on a Zenith / Heath H19 terminal which A  doesn't do VT100 emulation quite right, and I had to modify some >  of the SMG screen layouts compensate for this.  If I had beenB  able to work on it as a dedicated project I believe it would have  taken less time.   B     I would estimate that adding processing for the DETAIL records@  (add or delete) would take about 1 hour, now that I have all of>  the problems of working with SMG and Rdb solved.  Modifying aB  record would take a bit longer, as I would have to work out a wayB  to select the fields to be modified (possibly with another menu).A  Deleting a MASTER record and it's DETAIL records would be fairly    simple: probably 1/2 to 1 hour.  @     8. How long does it take to find and display a single recordA  which has 36 detail records for a system with a specific CPU and   disk:  ?     With a MV 3500 (equivalent) and RA81 disks, after the first =  retrieval has been performed, retrieval of a specific MASTER >  record with 36 DETAIL records (which include full fund names)B  take less than 1 second.  The very first retrieval of any kind is<  slower as it is at this time that the initial attach to the  database is done.  