This extended prac is designed to take you the best part of two weeks. Hand in your solutions before lunch time on Monday 31 October, correctly packaged in a transparent folder with your cover sheet and individual assessment sheets. Since the practical will have been done on a group basis, please hand in one copy of the cover sheet for each member of the group. These will be returned to you in due course, signed by the marker. Please make it clear whose folder you have used for the electronic submission, for example g03A1234. Please resist the temptation to carve up the practical, with each group member only doing one or two tasks. The group experience is best when you work on tasks together.
The reason for requiring all submissions by 31 October is to free you up during swot week to prepare for the final examinations. I shall try to get the marking done as soon as possible after that.
In this practical you are to
This prac sheet is at http://www.cs.ru.ac.za/courses/CSc301/Translators/trans.htm.
When you have completed this practical you should understand
Hopefully after doing these exercises (and studying the attributed grammar and the various other support modules carefully) you will find you have learned a lot more about compilers and programming languages than you ever did before (and, I suspect, a lot more than undergraduates at any other university in this country). I also hope that you will have begun to appreciate how useful it is to be able to base a really large and successful project on a clear formalism - namely the use of attributed context-free grammars - and will have learned to appreciate the use of sophisticated tools like Coco/R.
By the hand-in date you are required to hand in, besides the cover sheets (one per group member):
I do NOT require listings of any Java or C# code produced by Coco/R.
Keep the prac sheet and your solutions until the end of the semester. Check carefully that your mark has been entered into the Departmental Records.
You are referred to the rules for practical submission which are clearly stated in our Departmental Handbook. However, for this course pracs must be posted in the "hand-in" box outside the laboratory and not given to demonstrators.
A rule not stated there, but which should be obvious, is that you are not allowed to hand in another group's or student's work as your own. Attempts to do this will result in (at best) a mark of zero and (at worst) severe disciplinary action and the loss of your DP. You are allowed - even encouraged - to work and study with other students, but if you do this you are asked to acknowledge that you have done so. You are expected to be familiar with the University Policy on Plagiarism, which you can consult at:
http://www.scifac.ru.ac.za/plagiarism_policy.pdf
The tasks are presented below in an order which, if followed, should make the practical an enjoyable and enriching experience. Please do not try to leave everything to the last few hours, or you will come horribly short. You must work consistently, and with a view to getting an overview of the entire project, as the various components and tasks all interact in ways that will probably not at first be apparent. Please take the opportunity of coming to consult with me at any stage if you are in doubt as how best to continue. By all means experiment in other ways and with other extensions if you feel so inclined.
Please resist the temptation simply to copy code from model answers issued in previous years.
This version of Parva has been restricted so as not to include Parva functions. This means that there will be no practical work set on chapter 14 of the text. Because of the timing of our courses this is unavoidable, if highly regrettable. You should be warned that some of the material of that chapter may be examinable.
The operator precedences in Parva as supplied use a precedence structure based on that in C++ or Java, rather than the "Pascal-like" ones in the book. Study these carefully and note how the compiler provides "short-circuit" semantics correctly (see page 167) and deals with type compatibility issues (see section 12.6.8).
You are advised that it is in your best interests to take this opportunity of really studying the code in the Parva grammar and its support files. The exercises have been designed to try to force you to do that, but it is always tempting just to guess and to hack. With a program of this size that often leads to wasting more time than it saves. Finally, remember the advice given in an earlier lecture:
Keep it as simple as you can, but no simpler.
Throughout this project you will need to ensure that the features you explore are correctly implemented. This means that you will have to get a feel for understanding code sequences produced by the compiler. The best way to do this is usually to write some very minimal programs, that do not necessarily do anything useful, but simply have one, or maybe two or three statements, the object code for which is easily predicted and understood.
When the Parva compiler has finished compiling a program, say SILLY.PAV, you will find that it creates a file SILLY.COD in which the stack machine assembler code appears. Studying this is often very enlightening.
Useful test programs are small ones like the following. There are some specimen test programs in the kit, but these are deliberately incomplete, wrong in places, too large in some cases and so on. Get a feel for developing some of your own.
$D+ // Turn on debugging mode
void main (void) {
int i;
int[] List = new int[10];
while (true) { // infinite loop, can generate an index error
read(i);
List[i] = 100;
}
}
The debugging pragma
It is useful when writing a compiler to be able to produce debugging output - but sometimes this just clutters up a production quality compiler. The PARVA.ATG grammar makes use of the PRAGMAS option of Coco/R (see text, page 128) to allow pragmas like those shown to have the desired effect.
$D+ /* Turn debugging mode on */
$D- /* Turn debugging mode off */
There are several files that you need, zipped up in the file PRAC25.ZIP (Java) or PRAC25C.ZIP (C#).
J:
md prac25
cd prac25
copy i:\csc301\trans\prac25.zip
unzip prac25.zip
This will create several other directories "below" the prac25 directory:
J:\prac25
J:\prac25\library
J:\prac25\Parva
containing the Java classes for the I/O library, and for the code generator and symbol table handler.
You will also find the executable version of Coco/R and batch files for running it, frame files, and various sample data, code and the Parva grammar, contained in files with extensions like
*.ATG, *.PAV
We have already commented on the $D+ pragma. At present it is only used to request the printout of a symbol table. How would you change the system so that one would have to use a similar pragma or command line option if one wanted to obtain the assembler code file - so that the ".COD" file with the assembler code listing would only be produced if it were really needed?
$C+ /* Request that the .COD file be produced */
$C- /* Request that the .COD file not be produced */
Another useful (run-time) debugging aid is the undocumented stackdump statement. Compilation of this is also controlled by the $D pragma (in other words, the stack dumping code is only generated when in debug mode - much of the time you are testing your compiler you will probably be working in "debugging" mode, I expect).
Hint: This addition is almost trivially easy. You will also need to look at (and probably modify) the Parva.frame file, which is used as the basis for constructing the compiler proper (see page 140).
If, like me, you first learned programming in languages other than Java, you may persist in making silly mistakes when writing Parva programs. For example, programmers familiar with Pascal and Modula-2 easily confuse the roles played by ==, != and = with the similar operators in Pascal denoted by =, <> and :=, and are prone to introduce words like then into IfStatements, giving rise to code like
if (a = b) then c := d;
if (x <> y) then p := q;
instead of
if (a == b) c = d;
if (x != y) p = q;
Can you think of (and implement) ways to handle these errors sympathetically - that is to say, to report them, but then "recover" from them without further ado?
(Confusing = with == in Boolean expressions is also something C, C++ and Java programmers can do. It is particularly nasty in C/C++, where a perfectly legal statement like
if (a = b) x = y;
does not compare a and b, but assigns b to a, as you probably know).
Although not strictly illegal, the appearance of a semicolon in a program immediately following the condition in an IfStatement or WhileStatement, or immediately preceding a closing brace, may be symptomatic of omitted code. The use of a so-called EmptyStatement means that the example below almost certainly will not behave as its author intended:
read(i);
while (i > 10);
{
write(i);
i = i - 1;
}
It should be possible to warn the user when this sort of code is parsed; do so. Here is another example that might warrant a warning
while (i > 10) {}
Warnings are all very well, but they can become irritating. Use a $W- pragma or a -w command line option to allow advanced users to suppress warning messages.
Extend the WriteStatement to allow a variation introduced by a new key word writeLine that automatically appends a line feed to the output after the last WriteElement has been processed.
Extend the ReadStatement to allow a variation introduced by a new key word readLine that automatically ignores the remainder of the line after the last ReadElement has been processed.
Add the DoWhile loop to Parva, as exemplified by
do { a = b; c = c + 10; } while (c < 100);
Add an else option to the IfStatement. Oh, yes, it is trivial to add it to the grammar. But be careful. Some IfStatements will have else parts, others may not, and the code generator has to be able to produce the correct code for whatever form is actually to be compiled. The following silly examples are all valid.
if (a == 1) { c = d; }
if (a == 1) {}
if (a == 1) {} else {}
if (a == 1) {} else { b = 1; }
In many languages an IfStatement also allows optional elsif clauses - the grammar for this simply being on the lines of
IfStatement = "if" "(" Condition ")" Statement
{ "elsif" "(" Condition ")" Statement }
[ "else" Statement ] .
Implement this extension (make sure all the branches are correctly set up). By now you should know that this will lead to LL(1) warnings, but if you get the system correct these will not really matter.
The BreakStatement is syntactically simple, but takes a bit of thought. Give it some! Be careful - breaks can currently only appear within while loops, but there might be several break statements inside a single loop, and loops can be nested inside one another.
Parva will get closer to C/C++/Java with each successive long hour spent in the Hamilton Labs. Seems a pity to stop now, so go right on and extend the system to allow for a character type as well as the integer and Boolean ones, enabling you to develop classic programs like the following:
void main () {
// Read a sentence and write it backwards
char[] sentence = new char[1000];
int i = 0;
char ch;
read(ch);
while (ch != '.') { // input loop
sentence[i] = ch;
i = i + 1;
read(ch);
}
while (i > 0) { // output loop
i = i - 1;
write(sentence[i]);
}
}
Hint: A major part of this exercise is concerned with the changes needed to apply various constraints on operands of the char type. In some ways it ranks as an arithmetic type, so that Expressions of the form
character + character
character > character
character + integer
character > integer
are all allowable. However, assignment compatibility is more restricted. Assignments like
integer = integer;
integer = character;
character = character;
are all allowed, but
character = integer;
is not allowed. Following Java and C#, introduce a casting mechanism to handle the situations where it is necessary explicitly to convert integer values to characters, so that
character = (char) integer
would be allowed and, for completeness, so would
integer = (int) character
integer = (char) character
character = (char) character
But be careful. Parva uses an ASCII character set, so that executing code generated from statements like
int i = -90;
char ch = (char) 1000;
char ch2 = (char) 2 * i;
should lead to run-time errors.
At last! Let's really make Parva useful and turn it into Parva++ by adding the increment and decrement statement forms (don't try to extend expressions), exemplified by
int parva;
int [] list = new int[10];
char ch = 'A';
...
parva++;
--ch;
list[10]--;
Suggestions for doing this - specifically by introducing new operations into the PVM - are made in section 13.5.1 of the text. Be careful - only integer and character variables (and array elements) can be handled in this way.
In week 20 (in the prac test) we explored adding a MIN opcode to the PVM. This week, modify Parva and the PVM to allow for expressions in which might appear min() and max() functions, which can take one or more arguments, for example
WriteLine(max(90), min(3, 5, 2, 12), max(min(2, 4), min(21, 4)));
You should start by checking that you know what output might be expected from the execution of that statement!
Things are getting more interesting by this stage, and more challenging.
Many languages provide for a ForStatement in one or other form. Although most people are familiar with these, their semantics and implementation can actually be quite tricky.
Suppose we were to add a simple Pascal-style ForStatement to Parva, to allow statements whose concrete syntax is defined by
ForStatement = "for" Ident "=" Expression ("to" | "downto") Expression Statement .
for example
for i = 1 to 10 write(i); // forwards 1 2 3 4 5 6 7 8 9 10
for i = 10 downto 1 write(i); // backwards 10 9 8 7 6 5 4 3 2 1
for i = 10 to 5 write(i); // no effect
for i = i - 1 to i + 6 write(i); // what does this do?
for i = 1 to 5 read(i); // should we allow this?
for i = 1 to 5 i = i + 1; // should we allow this?
The dynamic semantics at first sight look easy. The ForStatement is often explained to beginners as being a shorthand form of writing a while loop - indeed in the C family of languages it is deemed to be just that. So, for example, the statements
for i = 1 to 10 write(i);
for i = 12 downto 4 write(i);
seem to be equivalent to
i = 1; i = 12;
while (i <= 10) { while (i >= 4) {
write(i); write(i);
i = i + 1; i = i - 1;
} }
However, it is not quite as simple as that. Consider an example like
for i = i + 4 to i + 10 write(i);
Here the obvious equivalent code leads to an potentially infinite loop
i = i + 4;
while (i <= i + 10) {
write(i);
i = i + 1;
}
In principle, the condition i <= i + 10 would now always be true, although the program would probably misbehave when the value assigned to i overflowed the capacity of an integer.
Pascal was a much "safer" language than C, and the semantics of the Pascal-style ForStatement are better described as follows. The statements
for Control = Expression1 to Expression2 Statement
for Control = Expression1 downto Expression2 Statement
should be regarded as more closely equivalent to
Temp1 := Expression1 Temp1 := Expression1
Temp2 := Expression2 Temp2 := Expression2
IF Temp1 > Temp2 THEN GOTO EXIT IF Temp1 < Temp2 THEN GOTO EXIT
Control := Temp1; Control := Temp1;
BODY: Statement BODY: Statement
IF Control = Temp2 THEN GOTO EXIT IF Control = Temp2 THEN GOTO EXIT
Control := Control + 1 Control := Control - 1
GOTO BODY GOTO BODY
EXIT: EXIT:
respectively, where Temp1 and Temp2 are temporary variables. This code will not assign a value to the control variable at all if the loop body is not executed, and will leave the control variable with the "obvious" final value if the loop body is executed. Code generation for these semantics may appear to be a little awkward for an incremental compiler, since there are now multiple apparent references to extra variables and to the control variable.
Hint: The simplest way of handling all these issues in the PVM system is to note that the obvious calls to the parsing routines to handle the sequence
Control
Expression1
Expression2
will ensure that code to push the address of the control variable and the values of the temporary variables will have been generated by the time the Statement forming the loop body is encountered. At this point code must be generated for the initial test, and if we avail ourselves of the ability to define extensions to the opcode set of the PVM we can generate a special opcode at this point that will perform the test, but leave these three values on the stack so that they can be used again later. Similarly, after generating the code for the Statement we can generate a second special opcode that can be responsible for the final test and possible increment of the control variable. One last complication is that once the for loop completes its execution we have to discard these three elements from the stack, which suggests a variation on the use of the DSP opcode.
Develop these ideas in detail. Ensure that your loop also supports the break statement. Secondly, insist on type compatibility between the control variable and the expressions defining its initial and final values. And, lastly, see if you can find a way to prevent "threatening" (tampering with) the control variable within the body of the loop, thus forbidding this sort of nonsense:
for i = 4 to 10 {
writeLine(i);
i = 5;
}
Something else to think about. In Pascal the word do is also required, as illustrated below. Would it be a good idea to insist on it in Parva as well?
ForStatement = "for" Ident "=" Expression ("to" | "downto") Expression "do" Statement .
When you learned Python you were probably hugely impressed by a Python feature which allows one to write multiple assignments into a single statement, as exemplified by
A, B = X + Y, 3 * List[6];
A, B = B, A; // exchange A and B
A, B = X + Y, 3, 5; // incorrect
which can obviously be described by the context-free production
Designator { "," Designator } "=" Expression { "," Expression } ";"
Describing the syntax is one thing. Getting the semantics and code generation right should keep you out of mischief for a while.
Way back in Practical 20 we added some specialized opcodes like LDC_1, LDA_2 and so on to the PVM. They are still there in the version supplied with the kit. Seems a shame not to use them, so modify the code generator to achieve this for you.
Making use of LDL N and STL N in place of LDA, LDV and STO where possible is a bit more challenging, but for some bonus marks, feel free to experiment if you wish.
Have fun, and good luck.