15 Nov

Structure clashes and program decomposition

Structure clashes

The design method that has been presented in the preceding chapters works when there is a correspondence between input and output data structures.

What happens when such correspondences don't exist, as in the following example?

Structure clashes:Net Movement Report

A warehouse records a transaction for every item received into or issued out of the warehouse. At the end of the day, the transaction file is sorted by item number, and a "Daily Net Movement Summary" showing the net movement of each item into or out of the warehouse is produced. The format is shown below:

Daily Net Movement SummaryA12345 40A23456 -30..........Z13579 25End Summary

Net Movement ReportSystem Diagram

Input file

C-Input; P- Report

Daily Net Movement Report

Warehouse Net Movementinput – output structure diagrams

Transaction file

Transaction group

Transaction*

Report

Title line

Report body

End line

Total line

*

*

Input file Report file

Warehouse Net MovementProgram Structure

* Consume records; accumulate net movement

Consume daily transaction file; Produce report

Consume file body; Produce report body

Close file; Display

ending line

Open file; Display report title

Consume group; Produce report line

*

Warehouse Net Movement:Elaborated program structure

Consume group; Accumulate netqty

9

* Consume record; accumulate net movement

Consume daily transaction file; Produce report

Consume file body; Produce report body

Close file; Display

ending line

Open file; Display report title

Consume group; Produce report line

*

Initialize group

5,6

7,8

1,2, 8, 10 3,4,

10

Modified Net Movement Problem

Report

Title line Report body End line

Report line*

Report file

Daily transaction

file

Block*

Record count Block body

*

Input file

Transaction record

Suppose the input file is blocked, with each block containing a record count followed by a number of records.

Modified Net Movement Problem:The essence of the difficulty

The program must have an operation that is executed once per block and an operation that is executed once per group

So there must be both a block component and a group component

But we cannot have a single program structure with process block and process group components.

We have a boundary clash--the boundaries of blocks are not synchronized with the boundaries of groups.

Solution to Boundary Clash:Decomposition

Daily Net Movement Report

Input file PA

Inter- mediate

filePB

Input, intermediate and report files, together with their correspondences

Daily transaction

file

Block *

Record count

Block body

*

Daily transaction

file

Transaction record

Transaction record

Input file Intermediate Intermediate Report file

*

Daily transaction

file

*

*Transaction

record

Part group

Daily net movement

report

Title lineReport body End line

Total line *

Program PA Program PB

file file

JSP Solution

PA seq reset(infile); xread(infile, block); rewrite(outfile); PA-BLOCK iter <while not eofbit> j := 1; PA-DEBLOCK iter <while not (j>reccnt)> outrec.itemno := inrec.i temno[j]; outrec.transcode := inrec.transcode[j]; outrec.qty := inrec.qty[ j]; write(outfile, outrec); j := j + 1; PA-DEBLOCK end xread(infile,block); PA-BLOCK end close(infile); close(outfile); PA end

PB seq reset(infile); xread(infile, rec); writeln(' Daily Net Movement Summary '); writeln; PB-REPORTBODY iter <while not eofbit> groupid := rec.itemno; netqty := 0; PB-GROUP iter <while not eofbit and (groupid = rec.itemno)> netqty := netqty + rec.qty; xread(infile, rec); PB-GROUP end writeln(' ', groupid, ' ', n etqty); PB-REPORTBODY end close(infile); PB end

Comments on JSP Solution

• The programs we obtain are distinct. – A serial file forms a boundary between any pair of programs.

We don't have to think "dynamically". • For example, we don't need to ask, "What if a group extends over

several blocks?" or • "What if a group has no data records?“

– We know our programs are correct, because we can think in terms of static data structures

• Inefficient– By introducing an intermediate file, we have roughly doubled the

execution time (in comparison with a program that produced a report without an intermediate file).

– We will learn a little later how to optimize our design by a simple program transformation, program inversion.

Interleaving clash

Let us suppose that our input file is incompletely sorted by part number. Total lines for each part group on the report may be in any order.

Daily transaction

file

*Transaction

Part group

Transaction*

Solution to Interleaving ClashSince the input file is not sorted completely by part number, we cannot

show the group structure and the input file structure on one diagram. • Our input file is an interleaving of part groups. • To resolve this "interleaving" clash, we split the input file into part groups

as shown below:

Psplit Input file

PG1

PGn

PG1

PGn

Output PreportDaily Net

Movement Report

JSP Solution

Input file

Transaction record

*

Intermediate file(s)

Part Group°

Transaction record

*

Net Movement Report

Report title

Report body

Part group total line

*

<PG1...PGn>

End line

To resolve this "interleaving" clash, we split the input file into part groups as shown below:

Structure ClashesThree types of structural clash:1. Boundary clash (physical blocks on input

don’t coincide with logical groups on output)

Boundary ClashSolution to Bounday Clash

2. Ordering clash – for example, we wish a report sorted by a person’s last name (surname), but input is unsorted)

3. Interleaving clash (wish to produce report of user job times, inputs are interleaved)

Solution to Structural Clashes

• General view of problemStructure diagram

• Solution: program decompositionProgram decomposition

There is no structure clash between A and I, so the structure of PA can be constructed without difficulty; the same is true of I and X and PB

Processing of general view:batch processing (пакетная обработка данных)

1. Batch processing

2. Parallel processing

3. Quasi-parallel processing (program inversion)

program inversion

Program Inversion инверсия программ

• We wish to design PA and PB independently of each other so that the structure of each is based on the data structure representation of its own problem environment.

• Program inversion is a purely mechanical transformation of the independent programs, PA and PB, into a main program and subroutine

• The subroutine has a single ENTRY point and stores its return address within its state vector so that it is resumable.

state vector (вектор состояния)

A program consists of:– program text

and– variables– text pointer to next instructionj to be executed

state vector

Significance of program Inversion

• Significance of program inversion

• Uses of inversion