Aims

This assignment aims to give you

·   an understanding of how data is treated inside a DBMS

  practice in adding a new base type to PostgreSQL

The goal is to implement a new data type for PostgreSQL, complete with    input/output functions, comparison operators, formatting functions, and the ability to build indexes on values of the type.

Summary

Deadline    Friday 15 March, 11:59pm

Pre-requisites:    before starting this assignment, it would be useful to

complete Prac Work P04

Late Penalty:    0.03 marks off the final mark for each hour late

for the first 5 days late; total mark of zero thereafter

Marks:   This assignment contributes 15 marks toward your total mark for this course.

Submission:   Webcms3 > Assignments > Ass1 Submission > Make Submission

or, on CSE machines, give cs9315 ass1 pname.c pname.source

Make sure that you read this assignment specification carefully and completely before starting work on the assignment.

Questions which indicate that you haven't done this will simply get the

response "Please read the spec".

We use the following names in the discussion below

 PG_CODE ... the directory where your PostgreSQL source code is located (on vxdb, /localstorage/$USER/postgresql-15.6/)

.  PG_HOME ... the directory where you have installed the PostgreSQL

binaries   (on vxdb, /localstorage/$USER/pgsql/bin/)

 PG_DATA ... the directory where you have placed PostgreSQL's data (on vxdb, /localstorage/$USER/pgsql/data/)

.  PG_LOG ... the file where you send PostgreSQL's log output   (on vxdb, /localstorage/$USER/pgsql/data/log)

Introduction

PostgreSQL has an extensibility model which, among other things, provides a  well-defined process for adding new data types into a PostgreSQL server. This capability has led to the development by PostgreSQL users of a number of

types (such as polygons) which have become part of the standard distribution. It also means that PostgreSQL is the database of choice in research projects   which aim to push the boundaries of what kind of data a DBMS can manage.

In this assignment, we will be adding a new data type for dealing with people's names. "Hmmm", you say, "but aren't they just text strings, typically

implemented as two attributes, one for family name and one for given names?". That may be true, but making names into a separate base data type allows us   to explore how we store and manipulate them.

One common way of writing names (e.g. used in UNSW student systems) is

Note: some of the examples above have a space after the comma; some don't.  We give a more precise description of what text strings are valid PersonNames below.

Adding Data Types in PostgreSQL

The process for adding new base data types in PostgreSQL is described in the following sections of the PostgreSQL documentation:

  38.13 User-defined Types

  38.10 C-Language Functions

  38.14 User-defined Operators

·   SQL: CREATE TYPE

·   SQL: CREATE OPERATOR

·   SQL: CREATE OPERATOR CLASS

Section 38. 13 uses an example of a complex number type, which you can use as a starting point for defining your PersonName data type (see below). There are other examples of new data types under the directories:

 PG_CODE/contrib/chkpass/ ... an auto-encrypted password datatype

·   PG_CODE/contrib/citext/ ... a case-insensitive character string datatype

·   PG_CODE/contrib/seg/ ... a confidence-interval datatype

These may or may not give you some useful ideas on how to implement the     PersonName data type. For example, many of these data types are fixed-size, while PersonNames are variable-sized. A potentially useful example of

implementing variable-sized types can be found in:

·   PG_CODE/s rc/tutorial/funcs.c ... implementation of several data types

Setting Up

You ought to start this assignment with a fresh copy of PostgreSQL, without    any changes that you might have made for the Prac exercises (unless these   changes are trivial). Note that you only need to configure, compile and install   your PostgreSQL server once for this assignment. All subsequent compilation takes place in the s rc/tutorial directory, and only requires modification of the files there.

Once you have re-installed your PostgreSQL server, you should run the

following commands:

Note the pname.* files will contain many references to complex; I do not   want to see any remaining occurences of the word complex in the files that you eventually submit. These files simply provide a template in which  you create your PersonName type.

Once you've made the pname.* files, you should also edit the Makefile in this directory and add the green text to the following lines:

The rest of the work for this assignment involves editing only the pname.c and pname.source files. In order for the Makefile to work properly, you must

use the identifier _OBJWD_ in the pname.source file to refer to the directory  holding the compiled library. You should never modify directly the pname.sql file produced by the Makefile. Place all of your C code in the pname.c file;  do not create any other *.c files.

Note that your submitted versions of pname.c and pname.source should not contain any references to the complex type. Make sure that the

documentation (comments in program) describes the code that you wrote.     Leaving the word complex anywhere in either pname.* file will result in a 1 mark penalty.

The Person Name Data Type

We wish to define a new base type PersonName to represent people's names,  in the format FamilyName,GivenNames. We also aim to define a useful set of operations on values of type PersonName and wish to be able to create

indexes on attributes of type PersonName. How you represent PersonName values internally, and how you implement the functions to manipulate them     internally, is up to you. However, they must satisfy the requirements below.

Once implemented correctly, you should be able to use your PostgreSQL

server to build the following kind of SQL applications:

Having defined a hash-based file structure, we would expect that the queries would make use of it. You can check this by adding the keyword EXPLAIN   before the query, e.g.

which should, once you have correctly implemented the data type and loaded sufficient data, show that an index-based scan of the data is being used. Note that this will only be evident if you use a large amount of data (e.g. one of the  larger test data samples to be provided).

Person Name values

Valid PersonNames will have the above format with the following qualifications:

  there may be a single space after the comma

  there will be no people with just one name (e.g. no Prince, Jesus,

Aristotle, etc.)

  there will be no numbers (e.g. noGates, William 3rd)

  there will be no titles (e.g. no Dr, Prof, Mr, Ms)

  there will be no initials (e.g. no Shepherd,John A)

In other words, you can ignore the possibility of certain types of names while implementing your input and output functions.

If titles occur, you can assume that they will occur after a comma after the given names, e.g. "Smith, John, Dr".

A more precise definition can be given using a BNF grammar:

You should not make any assumptions about the maximum length of a

PersonName.

Under this syntax, the following are valid names:

The following names are not valid in our system:

Think about why each of the above is invalid in terms of the syntax definition.

Important: for this assignment, we define an ordering on names as follows:

·   the ordering is determined initially by the ordering on the Family Name   if the Family Names are equal, then the ordering is determined by the

Given Names

  ordering of parts is determined lexically

There are examples of how this works in the section on Operations on

PersonNames below.

Representing Person Names

The first thing you need to do is to decide on an internal representation for your PersonName data type. You should do this, however, after you have looked at  the description of the operators below, since what they require may affect how   you decide to structure your internal PersonName values.

When you read strings representing PersonName values, they are converted into your internal form, stored in the database in this form, and operations on

PersonName values are carried out using this data structure. It is useful to

define a canonical form for names, which may be slightly different to the form in which they are read (e.g. "Smith, John" might be rendered as "Smith,John").

When you display PersonName values, you should show them in canonical form, regardless of how they were entered or how they are stored.

The first functions you need to write are ones to read and display values of type PersonName. You should write analogues of the functions complex_in(),

complex_out that are defined in the file complex.c. Call them, e.g.,

pname_in() and pname_out(). Make sure that you use the V1 style function interface (as is done in complex.c).

Note that the two input/output functions should be complementary, meaning    that any string displayed by the output function must be able to be read using  the input function. There is no requirement for you to retain the precise string  that was used for input (e.g. you could store the PersonName value internally in a different form such as splitting it into two strings: one for the family

name(s), and one for the given name(s)).

One thing that pname_in() must do is determine whether the name has the correct structure (according to the grammar above). Your pname_out()

should display each name in a format that can be read by pname_in().

Note that you are not required to define binary input/output functions, called

receive_function and send_function in the PostgreSQL

documentation, and called complex_send and complex_recv in the

complex.cfile.

As noted above, you cannot assume anything about the maximum length of names. If your solution uses two fixed-size buffers (one for family, one for

given) then your mark is limited to a maximum of 8/ 15, even if you pass all of the tests.

Operations on person names

You must implement all of the following operations for the PersonName type:

  PersonName1 = PersonName2 ... two names are equal

Two PersonNames are equivalent if, they have the same family name(s) and the same given name(s).

  PersonName1> PersonName2 ... the first PersonName is greater than the second

PersonName 1is greater than PersonName2if the Family part of

PersonName 1is lexically greater than the Family part of PersonName2 . If the Family parts are equal, then PersonName 1is greater than

PersonName2if the Given part of PersonName 1is lexically greater than the Given part of PersonName2 .

  Other operations:   <>,   >=,   <,   <=

You should also implement the above operations, whose semantics is

hopefully obvious from the descriptions above. The operators can typically be implemented quite simply in terms of the first two operators.

  family(PersonName) returns just the Family part of a name

  given(PersonName) returns just the Given part of a name

  show(PersonName) returns a displayable version of the name

It appends the entire Family name to the first Given name (everything before the first space, if any), separated by a single space.

Hint: test out as many of your C functions as you can outside PostgreSQL (e.g. write a simple test driver) before you try to install them in PostgreSQL. This will  make debugging much easier.

You should ensure that your definitions capture the full semantics of the

operators (e.g. specify commutativity if the operator is commutative). You

should also ensure that you provide sufficient definitions so that users of the PersonName type can create hash-based indexes on an attribute of type    PersonName.

Submission

You need to submit two files: pname.c containing the C functions that

implement the internals of the PersonName data type, and pname.source  containing the template SQL commands to install the PersonName data type  into a PostgreSQL server. Do not submit the pname.sql file, since it contains absolute file names which are not helpful in our test environment.

Have fun, jas