Chapter 11 - Pointers and Dynamic Allocation
Prerequisites for this material
In order to understand this chapter, you should have a good grasp of the entirety of Part I and a clear understanding of chapter 10 on Records.
For certain types of programs, pointers and dynamic allocation can be a tremendous advantage, but most programs do not need such a high degree of data structure. For that reason, it would probably be to your advantage to lightly skim over these topics and come back to them later when you have a substantial base of Modula-2 programming experience. It would be good to at least skim over this material rather than completely neglecting it, so you will have an idea of how pointers and dynamic allocation work and that they are available for your use when needed.
A complete understanding of this material will require deep concentration as it is very complex and not at all intuitive. Nevertheless, if you pay close attention, you will have a good grasp of pointers and dynamic allocation in a short time.
What are pointers, and what good are they?
If you examine POINTERS.MOD, you will see a very trivial example of pointers and how they are used. In the VAR declaration, you will see that the two variables have the two reserved words POINTER TO in front of their respective types. These are not actually variables, instead, they point to dynamically allocated variables that have not been defined yet, and they are called pointers. We will see, when we get to chapter 14, that a pointer can be used to point to any variable, even a statically defined one, but that will have to wait awhile.
The pointer "MyName" is a pointer to a 20 character string and is therefore not a variable into which a value can be stored. This is a very special TYPE, and it cannot be assigned a character string, only a pointer value or address. The pointer actually is a variable that contains the memory address of (and therefore "points to") data stored at that memory address.
Back to our example program, POINTERS.MOD. When we actually begin executing the program, we still have not defined the variables we wish to use to store data in. The first executable statement in line 14 generates a variable for us with no name and stores it on the heap. Since it has no name, we cannot do anything with it, except for the fact that we do have a pointer "MyAge" that is pointing to it. By using the pointer, we can store any INTEGER in it, because that is its type, and later go back and retrieve it.
What is dynamic allocation?
The variable we have just described is a dynamically allocated variable because it was not defined in a VAR declaration, but with an ALLOCATE procedure. The ALLOCATE procedure creates a variable of the type defined by the pointer, and assigns the address of the variable to the pointer itself. Thus "MyAge" contains the address of the variable generated. The variable itself is referenced by using the pointer to it followed by a ^, and is read, "the variable to which the pointer points".
The ALLOCATE procedure requires 2 arguments, the first of which is a pointer which will be used to point to the desired new block of dynamically allocated menory, and the second which gives the size of the block in bytes. The supplied function TSIZE will return the size of the block of data required by the TYPE supplied to it as an argument. Be sure to use the TYPE of the data and not the TYPE of the pointer to the data for the argument. Another procedure is available named SIZE which returns the size of any variable in bytes.
The next statement assigns memory space to an ARRAY type variable and puts its address in "MyName". Following the ALLOCATE statements we have two assignment statements in which the two variables pointed at are assigned values compatible with their respective types, and they are both written out to the video display. Notice that both of these operations use the ^ which is the dereference operator. By adding the dereference operator to the pointer name, you can use the entire name just like any other variable name.
The last two statements are illustrations of the way the dynamically allocated variables are removed from use. When they are no longer needed, they are disposed of with the DEALLOCATE procedure, and the space in memory is freed up for use by other dynamically allocated variables.
In such a simple program, pointers cannot be appreciated, but it is necessary for a simple illustration. In a large, very active program, it is possible to define many variables, dispose of some of them, define more, and dispose of more, etc. Each time some variables are deallocated, their space is then made available for additional variables defined with the ALLOCATE procedure.
It may be interesting to know that the ALLOCATE procedure uses memory of that section of memory that is called "the heap". The heap is a part of memory that is free for programs to use dynamicaly. The order in which memory space is ALLOCATED and DEALLOCATED to pointers, is totally under the control of the programmer so there is no guarantee that the first variable to be destroyed is the last that was created. Hence you can have memory fragmentation. The actual allocation of memory is performed by the operating system.
The heap can be made up of any assortment of variables, they do not have to all be the same. One thing must be remembered. Anytime a variable is defined, it will have a pointer pointing to it. The pointer is the only means by which the variable can be accessed. If the pointer to the variable is lost or changed, the data itself is lost for all practical purposes.
What about the "NEW" and "DISPOSE" procedures?
The NEW and DISPOSE procedures are standard Modula-2 procedures that are simply translated internally into calls to ALLOCATE and DEALLOCATE which must be imported in order to use NEW and DISPOSE.
Dynamically storing records;
The next example program, DYNREC.MOD, is a repeat of one we studied in an earlier chapter. For your own edification, review the example program BIGREC.MOD before going ahead in this chapter.
Assuming that you are back in DYNREC.MOD, you will notice that this program looks very similar to the earlier one, and in fact they do exactly the same thing. The only difference in the TYPE declaration is the addition of a pointer "PersonID", and in the VAR declaration, the first four variables are defined as pointers here, and were defined as record variables in the last program.
We just broke the great rule of Modula-2
Notice in the TYPE declaration that we used the identifier "Person" (in line 21) before we defined it (in line 22), which is illegal in Modula-2. Foreseeing the need to define a pointer prior to the record, the designer of Modula-2 allows us to break the rule in this one place. The pointer could have been defined after the record in this case, but it was more convenient to put it before, and in the next example program, it will be required to put it before the record. We will get there soon.
Examining the VAR declaration, we see that "Friend" is really 50 pointers, so we have now defined 53 different pointers to records, but no variables other than "Temp" and "Index". We dynamically allocate a record with "Self" pointing to it, and use the pointer to fill the new record. Compare the statements that fill the record with the corresponding statements in "BIGREC" and you will see that they are identical except for the addition of the ^ to each use of the pointer to designate the data pointed to.
This is a trick, be careful
Now go down to the place where "Mother" is assigned a record and is then pointing to the record. It seems an easy thing to do then to simply assign all of the values of "Self" to all the values of "Mother" as shown in the next statement, but it doesn't work. All the statement does, is make the pointer "Mother" point to the same place where "Self" is pointing. The data space that was allocated to the pointer "Mother" is now somewhere on the heap, but since we have lost the original pointer to it, we cannot find it, use it, or deallocate it. This is an example of losing data on the heap. The proper way is given in the next two statements where all fields of "Father" are defined by all fields of "Mother" which is pointing at the original "Self" record. Note that since "Mother" and "Self" are both pointing at the same record, changing the data with either pointer results in the data appearing to be changed in both because there is, in fact, only one data field.
A note for Pascal programmers
In order to WRITE from or READ into a dynamically assigned record it is necessary to use a temporary record since dynamically assigned records are not allowed to be used in I/O statements in Pascal. This is not true in Modula-2, and you can write directly out of a dynamically allocated record in Modula-2. This is illustrated in the section of the program that writes some data to the monitor.
Finally, the dynamically allocated variables are deallocated prior to ending the program. For a simple program such as this, it is not necessary to deallocate them because all dynamic variables are deallocated automatically when the program is terminated, but the DEALLOCATE steps are included for illustration. Notice that if the "DEALLOCATE(Mother)" statement was included in the program, the data could not be found due to the lost pointer, and the program would be unpredictable, probably leading to a system crash.
So what good is this anyway?
Remember when you were initially studying BIGREC? I suggested that you see how big you could make the constant "NumberOfFriends" before you ran out of memory. At that time you probably found that it could be made slightly greater than 1000 before you got the memory overflow message at compilation. If your compiler uses a large memory model, you may have been able to go much larger. Try the same thing with DYNREC to see how many records it can handle, remembering that the records are created dynamically, so you will have to run the program to actually run out of memory. The final result will depend on how much memory you have installed, and how many programs you are using simultaneously.
Now you should have a good idea of why Dynamic Allocation can be used to greatly increase the usefulness of your programs. There is, however, one more important topic we must cover on dynamic allocation. That is the linked list.
What is a linked list?
Understanding and using a linked list is by far the most baffling topic you will confront in Modula-2. Many people simply throw up their hands and never try to use a linked list. I will try to help you understand it by use of an example and lots of explanation. Examine the program LINKLIST.MOD for an example of a linked list. I tried to keep it short so you could see the entire operation and yet do something meaningful.
To begin with, notice that there are two TYPEs defined, a pointer to the record and the record itself. The record, however, has one thing about it that is new to us, the last entry, "Next" is a pointer to another record of the same type "FullName". This record therefore has the ability to point to another record of the same type which is extremely useful in some cases. In fact, this is the way a linked list is used. I must point out, that the pointer to another record, in this case called "Next", does not have to be last in the list, it can be anywhere it is convenient for you.
A couple of pages ago, we discussed the fact that we had to break the great rule of Modula-2 and use an identifier before it was defined. This is the reason the exception to the rule was allowed. Since the pointer points to the record, and the record contains a reference to the pointer, one has to be defined after being used, and by rules of Modula-2, the pointer can be defined first. That is a mouthful but if you just use the syntax shown in the example, you will not get into trouble with it.
Still no variables?
It may seem strange, but we still will have no variables defined, except for our old friend "Index". In fact for this example, we will only define 3 pointers. In the last example we defined 54 pointers, and had lots of storage room. Before we are finished, we will have at least a dozen pointers but they will be stored in our records, so they too will be dynamically allocated.
Lets look at the program itself now. First, we create a dynamically allocated record and define it by the pointer "PlaceInList". It is composed of the three data fields, and another pointer. We define "StartOfList" to point to the first record created, and we will leave it unchanged throughout the program. The pointer "StartOfList" will always point to the first record in the linked list which we are building up.
We define the three variables in the record to be any name we desire for illustrative purposes, and set the pointer in the record to NIL. NIL is a reserved word that doesn't put an address in the pointer but defines it as empty. A pointer that is currently NIL cannot be used to write a value to the display as it has no value, but it can be tested in a logical statement to see if it is NIL. It is therefore a dummy assignment. With all of that, the first record is completely defined.
Defining the second record
When you were young you may have played a searching game in which you were given a clue telling you where the next clue was at. The next clue had a clue to the location of the third clue. You kept going from clue to clue until you found the prize. You simply exercised a linked list. We will now build up the same kind of a list in which each record will tell us where the next record is at.
We will now define the second record. Our goal will be to store a pointer to the second record in the pointer field of the first record. In order to keep track of the last record, the one in which we need to update the pointer, we will keep a pointer to it in "TempPlace". Now we can create another new record and use "PlaceInList" to point to it. Since "TempPlace" is still pointing at the first record, we can use it to store the value of the pointer to the new record in the old record. The 3 data fields of the new record are assigned nonsense data for our illustration, and the pointer field of the new record is assigned NIL.
Lets review our progress to this point. We now have the first record with a name, composed of 3 parts, and a pointer to the second record. We also have a second record storing a different name and a pointer assigned NIL. We also have three pointers, one pointing to the first record, one pointing to the last record, and one we used just to get here since it is only a temporary pointer. If you understand what is happening so far, lets go on to add some additional records to the list. If you are confused, go back over this material again.
Ten more records
The next section of code is contained within a FOR loop so the statements are simply repeated ten times. If you observe carefully, you will notice that the statements are identical to the second group of statements in the program (except of course for the name assigned). They operate in exactly the same manner, and we end up with ten more names added to the list. You will now see why the temporary pointer was necessary, but pointers are cheap, so feel free to use them at will. A pointer only uses 4 bytes of memory.
We now have generated a linked list of twelve entries. We have a pointer pointing at the first entry, and another pointer pointing at the last. The only data stored within the program itself are three pointers, and one integer, all of the dynamically allocated data is on the heap. This is one advantage to a linked list, it uses very little internal memory, but it is costly in terms of programming. You should never use a linked list simply to save memory, but only because a certain program lends itself well to it. Some sorting routines are extremely fast because of using a linked list, and it could be advantageous to use in a database.
A graphic picture of the data should aid in your understanding of what we have done so far.
How do we get to the data now?
Since the data is in a list, how can we get a copy of the fourth entry for example? The only way is to start at the beginning of the list and successively examine pointers until you reach the desired one. Suppose you are at the fourth and then wish to examine the third. You cannot back up, because you didn't define the list that way, you can only start at the beginning and count to the third. You could have defined the record with two pointers, one pointing forward, and one pointing backward. This would be a doubly-linked list and you could then go directly from entry four to entry three.
Now that the list is defined, we will read the data from the list and display it on the video monitor. We begin by defining the pointer, "PlaceInList", as the start of the list. Now you see why it was important to keep a copy of where the list started. In the same manner as filling the list, we go from record to record until we find the record with NIL as a pointer.
Finally, it is necessary to DEALLOCATE the list, being careful to check for the ending NIL before you deallocate it.
There are entire books on how to use linked lists, and many Modula-2 programmers will seldom, if ever, use them. For this reason, additional detail is considered unnecessary, but to be a fully informed Modula-2 programmer, some insight into linked lists is necessary.