The Datafile PD-CD 3/PDCD_3.iso/languages/pascal/PTC/_PtC1/_PTCDoc/MS

Text File | 1991-03-02 | 34.6 KB | 1,504 lines

.\"            @(#)doc.ms    1.6 Date 87/06/29
.if t .rm CM
.ds LH Holistic Technology AB
.ds CH PTC
.ds RH HD870410-1 Rev: 1.6
.ds LF
.ds CF
.ds RF
.nr LL 6.5i
.nr PO 1i
.nr HM 0.75i
.nr FM 1.5i
.ie t .pl 842p
.el   .pl 72v
.ch BT -\n(FMu/2u
.\" 3 tabs/inch at 12 cpi corresponds to 4 chars/tab
.nr t 1i/3u
.am DS
.ta \ntu +\ntu +\ntu +\ntu +\ntu +\ntu +\ntu +\ntu +\ntu +\ntu +\ntu +\ntu +\ntu +\ntu +\ntu +\ntu +\ntu +\ntu
..
.LP
.rs
.sp |8c
.nf
.ce 10
.if t .ps +6
.B "PTC implementation note"
.if t .ps -6

by

Per Bergsten

Holistic Technology AB
Grona Gatan 59
414 54 Gothenburg
Sweden
.ce 0
.sp 12
.LP
This note describes the implementation of ptc, a Pascal to C translator.
The program was developed by Per Bergsten of Holistic Technology AB,
Gothenburg, Sweden.
The paper is intended to provide a guide for those who need to transport
ptc to a new environment,
it describes how Pascal constructs are mapped onto C constructs.
.bp
.NH S 1
Background
.LP
.ds CF - % -
.nr % 1
The aim was originally to provide a simple tool for transporting finished
applications to systems lacking Pascal compilers.
The first versions,
constructed in about 1984,
were capable of translating simple Pascal programs.
It was never intended to become a released product,
however,
as time went by, programs and ambitions grew to the point where nearly
the full Pascal language was supported.
Thus the program as it stands today has a long ancestry
but it has never been redesigned (which it should have been).
.NH 1
Pascal vs C
.LP
To anyone familiar with the two languages it is obvious that
they are very similar in structure.
The major features may be summarised as follows:
.TS
c c
l l .
Pascal    C
.sp
Block-structured    Block-structured
- multiple declaration levels    - single declaration level
Statically scoped    Statically scoped
Strongly typed    Weakly typed
    - allows unbounded pointers
Call by value    Mostly call by value
- and call by reference
Highly independent    Highly integrated
- of host system    - with system
Self contained    Allows external definitions.
.TE
.LP
On the syntactic level the languages differ only in minor ways,
mostly in the order in which keywords and other symbols occur,
and of course in that the languages uses different symbols for the same
purposes.
The only complication is that C prohibits nested subroutine declarations.
.LP
On the semantic level the situation is worse.
C has the peculiarity that array variables are treated differently from other
variables,
this forces us to adopt some general way to handle array variables.
Furthermore, since Pascal offers nested subroutine declarations
it becomes necessary to simulate part of the activation record mechanism
in the translated code,
in one case it is extremely difficult to completely do this.
It is also clear that the C typedef mechanism has some shortcomings that
preclude an easy translation of Pascal types.
.bp
.NH 1
Mapping Pascal to C
.LP
In this part of the paper we briefly illustrate how to translate
Pascal code into equivalent C code.
.NH 2
Programs
.LP
A minimal Pascal program:
.DS
program p;
begin
end.
.DE
translates to the C equivalent:
.DS
extern void exit(\^);
main(\^)
{
        exit(0);
}
.DE
.LP
It should be noted here that
the translator does not actually require a complete Pascal program,
the implementation is such that any consistent set of declarations can
be translated.
.NH 2
Lexical issues
.LP
The C language uses ASCII as a carrier,
almost all of the availible characters are used.
The Pascal definition uses a smaller set of characters.
Since few features of the languages depend on the underlying character set
this does not introduce much difficulties.
.LP
One serious problem does occur.
Both language definitions states that comments have no meaning and no
clear place in the language syntax.
Furthermore, the Pascal definition states that a comment is equivalent
to a blank character.
This implies that if comments are handled accurately
the translator should also be able to collect and classify each
blank character in a Pascal program and to generate a C program with the
same number of blank characters in the corresponding positions.
This implication conflicts with the fact that the languages have different
syntax rules, it is not obvious what the "corresponding positions" would be.
.LP
Since comments have no defined meaning a user is free to interpret them
in any way and, in particular, to associate them with the surrounding code
in any way s/he chooses.
Although humans usually are able to deduce what bearing a comment has on the
surrounding program code there are no formal rules for how to do this.
Therefore there is no
.I "a priori"
correct way to translate comments
and the translator described here ignores comments altogether.
If/when a reasonable
.I "ad hoc"
solution is found that feature may be incorporated.
.NH 2
Declarations
.LP
The program may introduce local declarations which are handled as follows.
.bp
.NH 3
Labels
.LP
.DS
program p;

label    9;

begin
9:
        goto 9
end.
.DE
which we simply translate into:
.DS
extern void exit(\^);
main(\^)
{
L9:
        goto L9;
        exit(0);
}
.DE
'ne 12
.LP
If the label is reached from an inner procedure:
.DS
program p;

label    100;

        procedure q;

        begin
                goto 100
        end;

begin
100:
end.
.DE
a more complicated translation must be used:
.DS
# define Line __LINE__
void Caseerror(\^);
# include <setjmp.h>
static struct Jb { jmp_buf jb; } J[1];

 void
q(\^)
{
        longjmp(J[0].jb, 100);
}

main(\^)
{
        if (setjmp(J[0].jb))
                goto L100;
L100:
        exit(0);
}
.DE
.LP
We assume the existence of the standard
.I setjmp(\^)
and
.I longjmp(\^)
library functions.
Jumpbuffers are statically allocated as needed depending on the number
of declarationlevels in the program.
.NH 3
Constants
.LP
Constant declarations are treated in two different ways.
Numbers aliases etc are simply
.I "# define" 'd
but string constants are converted to static character arrays in order
to avoid unnecessary duplication of string-constants in the object code,
thus:
.DS
const
    p    = 3.14;
    pie    = '3.14';
.DE
translates to:
.DS
# define pi 3.14
static char pie[\^] = "3.14";
.DE
.NH 3
Types and variables
.LP
Types and variables are mostly easy to translate:
.DS
program p;

const length    = 15;

type
    struct        = 0 .. length;
    vect        = array [ struct ] of struct;
    color        = (red, blue, ada, yellow);
    pointer    = ^ recd;
    recd        = record
                r    : pointer;
                case b : boolean of
                  false:    (c : color);
                  true:    (v : vect);
              end;

var    SP        : pointer;

begin
    new(SP, true);
end.
.DE
becomes
.DS
typedef char    boolean;
# define false (boolean)0
# define true (boolean)1
extern char *Bools[\^];
# define Unionoffs(p, m) (((long)(&(p)->m))-((long)(p)))
extern char *malloc(\^);
# define length 15
typedef unsigned char C47_struct;
typedef struct { C47_struct A[length + 1]; } vect;
typedef enum { red, blue, ada, yellow } color;
typedef struct S57 *pointer;
typedef struct S57 {
    pointer    r;
    boolean    b;
    union {
        struct {
            color    c;
        } V1;
        struct  {
            vect    v;
        } V2;
    } U;
}    recd;
pointer    sp;

main(\^)
{
    sp = (struct S57 *)malloc((unsigned)
        (Unionoffs(sp, U.V2.v) + sizeof(sp->U.V2)));
    exit(0);
}
.DE
The rationale is as follows:
.LP
Identifiers in the Pascal program which conflicts with reserved words in C are
renamed by adding a unique prefix Cnnn where nnn is a number.
.LP
We also note here that uppercase letters in identifiers and keywords
are translated to lowercase.
Pascal specifies that upper/lower case is insignificant whereas C
(for the present) separates the two.
This fact is used to advantage by the translator as all
subroutines and macros defined by the translator have an initial uppercase
letter which prevents confusion.
.IP \-
The type
.I boolean
is a predefined Pascal type,
when it is used the translator emits code which defines boolean to
be the smallest convenient type:
.I char .
The constants
.I false
and
.I true
are defined and the vector
.I Bools
will contain textstrings for output if needed.
.IP \-
The predefined types
.I integer
and
.I real
are likewise mapped directly onto the standard C types
.I int
and
.I double
through a typedef declaration.
.sp
Integer subranges are mapped onto standard C arithmetic types according to
a short table in the translator.
The table is scanned from top to bottom until an enclosing range is found
and the corresponding type-name is emitted.
.IP \-
C-arrays have peculiar semantix.
To unify the treatment of arrays and other datatypes we always encapsulate
an array in a struct,
thus an array always becomes a
.I struct
with a single member named A.
.IP \-
Records and their variants are translated into C
.I struct
and
.I union
definitions.
Since C requires all union members to have a name we must supply artificial
names for all record variants.
A record with variants will therefore always contain one field with the name U
which have sub-fields named Vnnn where nnn is a positive number.
.sp
'ne 12
When allocating dynamic storage for a record with variants naming
the desired variant
.DS
new(sp, true)
.DE
we face the problem of determining the amount of memory needed.
.QP
.B
C does not provide a safe way
to compute the size of a particular struct variant.
.IP
The strategy adopted to cope with this problem is to attempt to compute the
offset of a fieldmember in the variant matching the constant and then
to add the size of the variant.
The offset computation is expressed as a macro,
.I Unionoffs ,
which uses rather foul typecasting to achive the result.
The only availible alternative would be to use the same size of all variants.
This method,
while being safe,
wastes memory when many small and few large
records of the same type are created dynamically.
.IP \-
Pascal enumeration types are converted directly to C
.I enum
types.
.IP \-
Pascal pointer types are translated into C pointer types.
Pascal allows the programmer to construct recursive types as pointer
types need not be fully defined until the end of the declaration-section
where the pointer type is used.
In practice this is only used to introduce record types which
contain pointers to themselves.
This problem is partially solved by introducing a name for the record type.
.ne 10
Hence
.DS
type
    ptr    = ^ node;
    node    = record
            next    : ptr;
            info    : integer
          end;
.DE
becomes
.DS
typedef struct S56 * ptr;
typedef struct S56 {
    ptr        next;
    integer        info;
} node;
.DE
we note in passing that the problem cannot be completely solved since
.DS
type    pureptr    = ^ pureptr;
.DE
which is valid Pascal, cannot be expressed in C.
.ne 10v
.IP \-
A pascal set-type does not have any direct counterpart in C.
The C language does however have a adequate set of operators for
bit manipulation.
We use these to implement a Pascal set as an array of
.I setword .
So:
.DS
type
    s    = set of 1 .. 100;

var
    ss    : s;
.DE
is translated into:
.DS
typedef unsigned short setword;
typedef struct { setword S[8]; } s;

s    ss;
.DE
The situation is slightly complicated by the fact that Pascal has
a set constructor which permits the construction of arbitrary large sets,
for example:
.DS
s := [ 50 .. maxint ] * [ 1 .. 80 ]
.DE
for that reason the first member in the array of words gives
size of the set (in words).
In the example above s.S[0] would have the value 7,
and s.S[1] through s.S[7] would hold the bits.
The number 7 is computed on the assumption that the type
.I "unsigned short"
on the target host is sufficiently large to holds 16 bits.
The set operators of Pascal are implemented as C functions returning
pointers to arrays of setwords,
the intermediary results are held in a static area of fixed size.
.IP \-
Pascal files are implemented using the standard i/o package availible
in most C implementations.
Since Pascal has the requirement that the next element of a file is visible
in the filebuffer before it is read,
and the requirement that linemarkers in textfiles are given special treatement
we are forced to extend the
.I FILE
type provided in
.I <stdio.h> .
.ne 20
Hence:
.DS
var    f    : text;
.DE
becomes
.DS
typedef struct {
    FILE    *fp;
    unsigned short
            eoln:1,
            eof:1,
            init:1,
            out:1,
            :12;
    char    buf;
}    text;
text    f;
.DE
where
.I buf
is our filebuffer and
.I eoln ,
.I eof
and
.I init
are flags giving the state of the file.
All Pascal i/o operations are implemented using macros that maintain the flags
and the buffer variable.
The actual reading and writing of data is deferred to the standard i/o package.
.NH 3
Procedures and functions
.LP
Pascal procedures and functions are somewhat difficult to translate to C.
The main problems lie in nested declarations and procedural parameters.
Nested declarations are handled in the following manner:
.RS
.IP \-
If procedure B is declared in procedure A,
then procedure B is emitted before A and A is forward declared before B.
.IP \-
Any constants and types declared in A is moved to the global scope,
this may force renaming of those objects.
.IP \-
Any variable declared in A
.I
and used in B
.R
is converted to a pointer and moved to the global scope.
The pointer value is saved and re-initialized upon each entry of A
and restored upon exit from A.
.RE
.br
'ne 20
.LP
Hence:
.DS
procedure A;

const    limit    = 20;

type    loctyp    = 0 .. limit;

var    i, j    : loctyp;

    procedure B(k : loctyp);

    begin
        j := k + 2
    end;

begin
    B(i)
end;
.DE
becomes
.DS
typedef unsigned char    loctyp;
loctyp    *G56_j;

void a(\^);

 void
b(k)
    loctyp  k;
{
    (*G56_j) = k + 2;
}

 void
a(\^)
{
# define limit 20
    loctyp  i, j;
    loctyp  *F57;

    F57 = G56_j;
    G56_j = &j;
    b(i);
    G56_j = F57;
}
.DE
we see that references to
.I j
inside procedure
.I b
are replaced by the pointer
.I G56_j
which points to the local variable
.I j
in procedure
.I a.
In order to preserve the proper semantix in the face of recursion
the value of the pointer need also be saved in the local variable
.I F57
during the invocation of
.I a .
.IP \-
Procedure parameters offer little problems.
We translate Pascal value-parameters into ordinary C parameters
and Pascal var-parameters are treated as pointers.
.br
'ne 20
.IP \-
Procedural parameters appear at first to be easy to handle.
The ordinary program:
.DS
program p;

procedure pp(procedure q(i : integer));

begin
    q(2)
end;

procedure qq(i : integer);
begin
end;

begin
    pp(qq)
end.
.DE
becomes
.DS
extern void    exit(\^);
typedef int    integer;

 void
pp(q)
    void    (*q)(\^);
{
    (*q)(2);
}

 void
qq(i)
    integer i;
{
}

main(\^)
{
    pp(qq);
    exit(0);
}
.DE
which looks simple enough.
.br
However,
Pascal requires that the scope of a procedure
.I "remains unchanged"
throughout its lifetime.
.ne 35
Consider:
.DS
program demo(output);

var    i    : integer;

    procedure p(procedure q);

    var    j    : integer;

        procedure qq;

        begin
            writeln(j)
        end;

    begin
        j := i;
        q;
        if i < 1 then
          begin
            i := i + 1;
            p(qq)
          end
    end;

    procedure dummy;
    begin
    end;

begin
    i := 0;
    p(dummy)
end.
.DE
When
.I p
is first invoked it assigns the local variable
.I j
the value 0.
This variable is accessible from
.I qq
which is passed as a parameter in
the recursive call of
.I p .
The second invocation of
.I p
then sets
.I its
variable
.I j
to 1,
and calls
.I q
which is bound to the first instance of
.I qq ,
and should therefore print the number
.I 0 .
.B
Sadly,
the code produced by the translator fails to do this.
.R
It is obvious that the program above calls for a complete simulation
of the activation record mechanism of Pascal to work correctly.
.RS
.LP
A workable but unpractical solution would be:
.IP 1)
When qq is used as parameter we call a function q1 which saves the environment
for qq (i.e. the address of j) in a well known place
and returns a pointer to q2.
.IP 2)
When qq is later called (under the name q) the actual target will be q2
which sets up the proper environment calls qq.
.RE
.IP
The problem is that this requires a save-area for
.I each
procedural parameter which can hold the intresting
parts of its environment for
.I each
of its invocations.
In the example above we need one are which holds the address of i
for each instance of qq
(say N instances, where N depends on the run-time behaviour of p).
It also requires a set of different procedures to perform the work of q2
(N different procedures which sets up the environment for the N instances).
.B
This requires much to much work to implement so the problem is left unsolved,
.R
this is hardly a problem in practice since humans rarely write such code
but
.B
it could introduce problems
.R
in machine generated Pascal code.
.IP
It should be noted that
the translator accepts the keyword
.B external
in place of the Pascal
.B forward
directive and assumes that the so declared procedure is defined elsewhere.
.bp
.NH 2
Statements.
.LP
Pascal statements are comparatively easy to translate to C.
The only parts that require special care are
non-local
.I goto ,
.I with
and
.I for
statements.
The code
.DS
program p(output);

type
    msgtyp    = packed array [ 1 .. 12 ] of char;

var
    a    : array [ 1 .. 10 ] of
            record
                r    : real
            end;
    i    : integer;
    ok    : boolean;

    procedure fatal(m : msgtyp);

    begin
        writeln('Fatal error: ', m)
    end;

begin
    while true do
      repeat
        ok := false;
        i := 1;
        for i := i + 1 to i + 10 do
            if i > 10 then
                fatal(' 10 exceeded')
            else
              with a[i] do
                if r > 9.99e+38 then
                    ok := true
                else
                    writeln(r)
      until ok
end.
.DE
'ne 20
becomes
.DS
typedef char boolean;
# define false (boolean)0
# define true (boolean)1
typedef int integer;
typedef double real;

typedef struct { char A[12 - 1 + 1]; } msgtyp;
typedef struct { struct    S57 {
    real    r;
}    A[10 - 1 + 1]; }  T56;
T56        a;
integer    i;
boolean    done;

 void
fatal(m)
    msgtyp    m;
{
    (void)fprintf(output.fp, "Fatal error: %.12s", m.A),
                    Putchr('\en', output);
}

main(\^)
{
    while (true)
      do {
        done = false;
        i = 1;
        {
          integer        B1 = i + 1, B2 = i + 10;

          if (B1 <= B2)
            for (i = B1; ; i++) {
                if (i > 10)
                    fatal(*((msgtyp *)" 10 exceeded"));
                else {
                    register struct    S57 *W3 = &a.A[i - 1];

                    if (W3->r > 9.99e+38)
                        done = true;
                    else
                        (void)fprintf(output.fp, "% 20e", W3->r),
                            Putchr('\en', output);
                }
                if (i == B2) break;
            }
        }
      } while (!(done));
    exit(0);
}
.DE
for the following reasons:
.NH 3
Begin
.LP
The compound statements
are very similar in the two languages and need no further explanation.
.NH 3
If
.LP
Pascal if-statements
have the same structure and semantix as C if-statments.
.NH 3
Case
.LP
Pascal case-statements
have the same structure and semantix as C switch-statements provided that a
.I break
is always added to each entry.
.LP
The translator supports a common Pascal extension
in that it recognizes the keyword
.B otherwise
to signify a default entry in a case-statement.
.NH 3
Labels
.LP
Pascal labeled statements and labels have the same structure and semantix as
C labeled statements except that Pascal labels are numbers where C labels
are identifiers,
this difference is solved by simply prefixing the labels with the letter
.I L .
.NH 3
Goto
.LP
Pascal goto-statements
have the same structure as C goto statements but the semantix differ in
that Pascal allows a goto to lead out of the current procedure.
This is implemented using the
.I setjmp/longjmp
library functions of C as described earlier.
.NH 3
With
.LP
The with-statement
of Pascal has no counterpart in C.
It is translated into nested compund statements where pointervariables,
referencing the corresponding records,
are declared and initialized.
Within the scope of the with-statement,
the accessible record fields are renamed to include the pointer.
The effect of this is that the record address is evaluated once as the
Pascal standard requires.
.NH 3
For
.LP
The for-statement
of Pascal has a structure that is similar to the C for-statement
but the semantix differ completely.
Pascal requires that a loop be exited when the upper bound
has been reached,
Pascal also requires that the loop-boundaries be evaluated exactly once.
The standard C for-loop is exited when the loop-condition becomes false.
This implies that it is not always true that
.DS
for (i = 0; i <= e; i++) ;
.DE
behaves in the same manner as
.DS
for i := 0 to e do ;
.DE
since (in most implementations)
the C version becomes an infinite loop if
.I e
equals
.I maxint
or if
.I e
is the expression
.I i .
For that reason Pascal for-statments are translated into compound statements
where the upper/lower bounds of the for-loop are held in local variables.
It is also necessary to add a conditional break-statement at
the end of the loop.
It is possible to obtain the more relaxed translation by configuring the
translator appropriately (see "Tuning" below).
.NH 3
While
.LP
The while-statement behaves exactly the same in both languages.
.NH 3
Repeat
.LP
The repeat-statement of Pascal matches the do-while statement of C except
for the trivial difference that Pascal permits a statement-list
where C permits a single statment in the loop.
.NH 3
Empty
.LP
The empty statement has (of course) the same structure and semantix
in both languages.
.NH 2
Expressions and assignments
.LP
In most cases Pascal expressions can be literally translated into equivalent
C expressions.
.IP identifiers 15
Except where identifiers clash with reserved words or with other
identifiers declared in the same scope,
they may simply be printed.
In some cases the translator is forced to rename identifiers or to
invent new identifiers.
.IP operators
The operators +, -, *
.I div
and
.I mod
when applied to real or integer operands
have exact counterparts in C and are therefore easy to handle.
The operator for real division, /,
corresponds to the same C operator except that the operands may be integers.
In that case a cast is necessary.
When the operands are sets the expression is converted into a function call.
.sp
The operators <, <=, >, >=, = and <> all have exact counterparts in C
for integer and real operands.
Most C implementations disallows
.I enum
operands,
the translator therefore casts such operands to
.I int .
Comparisons on structures (i.e. strings or sets)
are converted into function calls.
.IP assignments
Assignments are straightforward to handle since arrays are encapsulated
in structures.
Therefore:
.DS
a := b
.DE
becomes
.DS
a = b
.DE
.I unless
b is a string or a set,
in which case the assignment is converted into a function call.
.IP indexing
Given the translation for array declarations (described above) we are forced
to translate
.DS
a[i]
.DE
into
.DS
a.A[i - c]
.DE
where
.I c
is the lower bound for the index type.
.IP selection
Given the translation for records with variants (described above) the
translation of
.DS
a.b
.DE
becomes
.DS
a.b
.DE
.I or ,
if b is declared in a variant,
.DS
a.Vxxx.b
.DE
where Vxxx is a name manufactured by the translator for the particular variant.
.IP dereferencing
Pointer references and
.B var -parameters
are handled by prefixing the expression with an asterisk,
but the special case dereferencing followed by selection is also recognized,
so:
.DS
p^ := q^.f
.DE
becomes
.DS
*p = q->f
.DE
.IP miscellanea
The boolean operators
.B and ,
.B or
and
.B not
are simply translated into their C counterparts.
The set contructors
.B "[\^]" ,
and
.B ".."
and the operator
.B in
are converted to function calls.
.bp
.NH 1
Implementation
.LP
The general strategy is to convert the Pascal source program
into a parsetree.
The tree is then traversed by a set of procedures that perform some
necessary transformations.
The tree is finally traversed by a set of procedures that print the
corresponding C constructs.
The translator consists of three major procedures that perform
these actions.
They are augmented by a set of procedures that maintain a symbol table
that holds information about identifiers found in the source,
and by a procedure that initializes all internal datastructures.
.LP
There are three major datastructures that interact in complicated ways:
.IP 1)
a store for identifiers and strings
.IP 2)
a multilevel symbol table
.IP 3)
a parse-tree.
.LP
The identifier and string store,
.B strstor
is in principle an array of characters that grow
in increments of one string block.
Exactly one copy of each identifier is stored in that array.
Whenever an identifier is found in the input stream the array is
scanned to see if that identifier has been seen before.
In order to speed up the search all identifiers are represented by nodes
which are chained together such that all nodes in a particular chain
have the same hashvalue as determined by the function
.B hash .
Each
.B idnode
holds an index to strstor where the corresponding identifier text is stored.
The start of the hashchains are found in the global variable
.B idtab .
.de Ds
.nr x \\w'\\$2'u
.ie t \{
.nr x \\nx/2
.ds \\$1 \\\\h'-\\nxu'\\$2\\\\h'-\\nxu'
.\}
.el .ds \\$1 \\$2\\\\h'-\\nxu'
..
.ds l \-
.ie t .ds a \z\*l\a
.el   .ds a \a\z\*l
.nr x \ntu/2
.ds g \z\*l\\l'\nxu\&\*l'
.ds h \\h'\nxu'
.Ds c +
.Ds d \(da
.Ds u \(ua
.Ds r \(br
.ds s \\*r\z\*l
.ds > \*l\z\*l\(->
.ds < \(<-\\h'-1n'\*l
.ds k \\kx
.ds b \\h'|\\nxu'
.ds t \t
.DS L
.lc \*l
.fc #
idtab
\*c\*a\*c
\*r\*t\*r\*tchain of idnodes with same hashvalue
\*c\*a\*c\*t\*c\*a\*a\*c\*t\*c\*a\*a\*c
\*k\*r\*t\*r\*b\*h\*a#\*> #\*r\*t\*k\*t\*r\*b\*h\*a#\*> #\*r\*t\*t\*r idnode representing
\*c\*a\*c\*t\*r\*t\*t\*r\*t\*k index=2\*b\*r\*t\*t\*r identifier "demo"
\*r\*t\*r\*t\*c\*a\*a\*c\*t\*c\*a\*a\*c

\*tstrstor
\*c\*a\*a\*a\ \*l \*a\*a\*c
\*r\*t\*r\*t\*r\*t\*t\*r\*t\*r
\*k\*c\*a\*a\*a\ \*l \*a\*a\*c\*b\*h\*r
\*h\*r
\*h\*d
\*c\*a\*a\*a\*a\*a\*a\*a\*a
\*r\*t\*r\*t\*r# d #\*r# e #\*r# m #\*r# o #\*r# / #\*r
\*c\*a\*a\*a\*a\*a\*a\*a\*a\*tfirst idblock
\*t\*t# \*u #
\*t\*tindex=2
.DE
.LP
So:
the global representation of the identifier "demo" is a particlular
idnode;
whenever the lexical analysis routine
recognizes the identifier "demo" it returns a pointer to that idnode.
.LP
In order to represent different identifiers with the same name we need to
be able to distinguish between identifiers declared in different scopes.
This is accomplished by the
.B symnode
structures.
When an identifier is first encountered in a given scope it is "declared",
meaning that a new symnode is created that references the identifier.
Occurrences of the same identifier in that scope are then represented in
the parse-tree by treenodes referencing the same symnode.
.bp
The program:
.DS
program p;

var    demo    : integer;

begin
    demo := 3
end.
.DE
yilds the following structure:
.DS L
.lc \*l
.fc #
\*t# top #
\*t# \*d #
\*t\*c\*a\*a\*a\*a\*c treenode representing
\*t\*k npgm\*b\*r\*t\*t\*t\*t\*r the program
\*t\*c\*a\*s\*a\*s\*a\*s\*a\*c
\*t\*t\*r\*t\*r\*h\*u\*t\*r\*h\*u
\*t\*t\*r\*t\*r\*h\*r\*t\*r\*h\*c\*a\*a\*a\*a\*a\*a\*a\*g\*c
\*t\*t\*r\*t\*r\*h\*r\*t\*c\*a\*a\*a\*a\*a\*a\*a\*c\*h\*r
\*t\*t\*r\*t\*d\*h\*r\*t\*t\*t\*t\*t\*t\*t\*t\*r\*h\*r
\*t\*t\*r\*h\*c\*a\*g\*s\*a\*a\*c\*k treenode representing\*b\*t\*t\*t\*t\*t\*t\*r\*h\*r
\*t\*t\*r\*h\*k nvar\*b\*r\*t\*t\*t\*\*kr the var-declaration\*b\*t\*t\*t\*t\*t\*t\*d\*h\*r
\*t\*t\*r\*h\*c\*g\*s\*a\*s\*a\*c\*t\*t\*t\*c\*a\*a\*a\*g\*s\*a\*c treenode repr.
\*t\*t\*r\*t\*r\*h\*u\*t\*r\*t\*t\*t\*t\*k nassign\*b\*r\*t\*t\*t\*t\*r assignment
\*t\*t\*r\*t\*r\*h\*r\*t\*c\*a\*k\*> to type\*b\*t\*t\*t\*c\*a\*s\*a\*a\*s\*a\*c
\*ksymtab\*b\*t\*t\*r\*t\*d\*h\*r\*t\*t\*t\*t\*t\*t\*d\*h\*u\*t\*t\*d\*h\*u
\*c\*a\*c\*t\*r\*h\*c\*a\*g\*s\*a\*c\*k treenode repr.\*b\*t\*t\*t\*t\*c\*a\*g\*s\*a\*c\*h\*c\*a\*g\*s\*a\*a\*c
\*r\*t\*r \*<\*a\*c\*h\*k nid\*b\*r\*t\*t\*r\*k occurrence of\*b\*t\*t\*t\*t\*k nid\*b\*r\*t\*t\*r\*h\*k ninteger\*b\*r\*t\*t\*t\*r
\*t\*t\*h\*c\*a\*s\*a\*c\*k id. "demo"\*b\*t\*t\*t\*t\*c\*a\*s\*a\*c\*h\*c\*a\*a\*s\*a\*c
\*r\*t\*r\*t\*t\*r\*h\*u\*t\*t\*t\*t\*t\*t\*r\*t\*t\*t\*r\*h\*u
\*c\*a\*c\*t\*t\*r\*h\*r\*t\*c\*a\*a\*a\*a\*a\*c\*t\*t\*t\*r\*h\*r
\*r\*t\*r\*t\*t\*d\*h\*r\*t\*d\*t\*t\*t\*t\*t\*t\*t\*t\*d\*h\*r
\*c\*a\*c\*t\*c\*a\*g\*s\*a\*a\*c\*k symnode representing\*b\*t\*t\*t\*t\*t\*t\*c\*a\*g\*s\*a\*g\*c
\*r\*k\*t\*r\*b\*h\*a\*>\*t\*k lidentifier\*b\*r\*t\*t\*t\*r\*k identifier "demo"\*b\*t\*t\*t\*t\*t\*t\*k linteger\*b\*r\*t\*t\*h\*r
\*c\*a\*c\*t\*c\*a\*s\*a\*a\*c\*k in the current scope\*b\*t\*t\*t\*t\*t\*t\*k linum=3\*b\*r\*t\*t\*h\*r
\*t\*t\*t\*r\*t\*t\*t\*t\*t\*t\*t\*t\*c\*a\*a\*g\*c
\*kidtab\*b\*t\*t\*t\*c\*a\*a\*a\*c
\*c\*a\*c\*t\*t\*t\*t\*t\*r
\*r\*t\*r\*t\*t\*t\*t\*t\*d
\*c\*a\*c\*t\*c\*a\*a\*c\*t\*c\*a\*a\*c
\*k\*r\*t\*r\*b\*h\*a#\*> #\*r\*t\*k\*t\*r\*b\*h\*a#\*> #\*r\*t\*t\*r idnode representing
\*c\*a\*c\*t\*r\*t\*t\*r\*t\*k index=2\*b\*r\*t\*t\*r identifier "demo"
\*r\*t\*r\*t\*c\*a\*a\*c\*t\*c\*a\*a\*c

\*tstrstor
\*c\*a\*a\*a\ \*l \*a\*a\*c
\*r\*t\*r\*t\*r\*t\*t\*r\*t\*r
\*c\*g\*s\*a\*a\*a\ \*l \*a\*a\*c
\*h\*r
\*h\*d
\*c\*a\*a\*a\*a\*a\*a\*a\*a
\*r\*t\*r\*t\*r# d #\*r# e #\*r# m #\*r# o #\*r# / #\*r
\*c\*a\*a\*a\*a\*a\*a\*a\*a\*tfirst idblock
\*t\*t# \*u #
\*t\*tindex=2
.fc
.lc
.DE
We see that the two occurrences of the identifier "demo" are represented by
two
.B treenodes
of variant
.B nid
that both have pointers to the same
.B symnode
representing the declaration of "demo".
All symnodes at a given declarationlevel are chained together (in the
same manner as the idnodes) and the chains are attached to the global
variable
.B symtab .
In order to find out if an identifer is declared in the current scope we
scan the chain of symnodes starting from symtab, and check if any of them
references the idnode we are looking for.
A symnode also have a pointer to the treenode which defines the symbol.
The
.B symtabs
themselves are also chained,
the chain implements a stack of declarationlevels.
The symtab which is created when the scope of a procedure is entered
is also attached to that procedure.
When a procedure is entered we create a new symtab, push it onto the stack,
parse the procedure and pop the stack.
The symbols then visible are those that still can be reached from the stack.
.LP
The parse-tree consists of
.B treenodes
representing each declaration, statement, expression etc.
Each node except the nprogram node
has a pointer to its immediate father (giving the defining point),
to its children (if it has any) and to its right brother (if it is
a member of a list).
The top of the tree is found in the global variable
.B top .
.LP
In order to find the defining point for the identifier in the assignment,
we follow pointers from the nassign-treenode
to the nid-treenode, to the lidentifier-symnode,
and then up to the nid-treenode in the declaration.
If we want to know the type of the identifier
we proceed up to the nvar-treenode
and then down to the node giving the type in the declaration
(in our example that node would also be a nid-treenode referencing a
linteger-symnode that represents the identifier "integer").
There is a function
.B typeof
that performs exactly this operation.
It will return a pointer to a npredef-, nsubrange-, nscalar-, nptr-
narray-, nrecord-, nfileof- or nsetof-treenode.
In those cases where the translator pays attention to types
it uses a pointer (obtained from typeof) as representation of a type.
.LP
Given the parse-tree and the layered symbol table
it is not hard to see how the translations described earlier are performed.
The one place where the code becomes more than usually complex is when a
.B write
statement with format specifications is to be translated into a call to
.B fprintf .
.bp
.NH 1
Tuning
.LP
The behaviour of the translator may be modified for some cases simply
by changing constants.
These constants are all located at the top of the program text.
.IP output 12
The translator will copy the source during input if the constant
.B echo
is true.
The copy is preceeded by the line
.DS
# ifdef PASCAL
.DE
and ended by the line
.DS
# else
.DE
and then follows the translated code
and finally
.DS
# endif
.DE
This may be used to hold both Pascal and C source in the same file.
If the Pascal part is changed the C part may be updated through:
.DS
    cpp -P -DPASCAL < orig > tmp
    ptc < tmp > new
    move new orig
.DE
assuming that
.B echo
is true and that
.B cpp
is the standard C preprocessor.
.DS
Default value:

    echo        = false;
.DE
.IP comments
The translator recognizes both (* and { as comment delimiters.
They are treated as different,
allowing 1 level nested comments,
if the constant
.B diffcom
is true.
.DS
Default value:

    diffcomm    = false;
.DE
.IP symbols
The translator accepts default entries in case-statements provided
that the keyword defined through
.B othersym
is used in place of the constant list.
.DS
Default value:

    othersym    = 'otherwise ';
.DE
substitute for
.DS
    othersym    = 'otherwise%';
.DE
if that feature is undesired.
.IP
The translator accepts externally declared procedures and functions
provided that the directive defined through
.B externsym
is used in place of the keyword
.B forward .
.DS
Default value:

    externsym    = 'external  ';
.DE
.IP sets
Sets are implemented as arrays of
.B wordtype .
The type is assumed to hold
.B "setbits + 1"
bits numbered from 0.
.DS
Default value:

    wordtype    = 'unsigned short';
    setbits    = 15;
.DE
.ne 10
.IP files
The implementation of files uses a flag-word which has the type given as
.B filebits
which is assumed to hold
.B "filefill + 4"
bits.
.DS
Default value:

    filebits    = 'unsigned short';
    filefill    = 12;
.DE
.ne 20
.IP stmts
If the Pascal source is known to be "normal" in the sense that for-loops
always have an upper bound that is less than the maximum value of the
type of the loop-variable, and in the sense that the upper bound doesn't
change by computations inside the loop, then the translator may generate
a more natural code.
I.e:
.DS
for i := 1 to 10 do ;
.DE
becomes
.DS
for (i = 1; i <= 10; i++) ;
.DE
Since the requirements cannot be determined by the translator itself
this kind of code is generated when the constant
.B lazyfor
is true.
.DS
Default value:

    lazyfor    = false;
.DE
.ne 10
.IP new
The second and following parameters to the procedure
.B new
will be ignored if the constant
.B unionnew
is false.
.DS
Default value:

    unionnew    = true;
.DE
.ne 10
.IP strings
All identifiers and strings are stored in the translator with the special
character having the ordinal value
.B null
as endmarker.
Hence, that character can not occur in strings in the Pascal source.
.DS
Default value:

    null        = 0;
.DE
.ne 20
.IP types
The names of predefined types are given by the constants:
.B inttyp ,
.B chartyp ,
.B floattyp ,
and
.B doubletyp .
.DS
Default value:

    inttyp        = 'int';
    chartyp    = 'char';
    floattyp    = 'float';
    doubletyp    = 'double';
.DE
The typename for real variables and functions defined by the user
is given by the constant
.B realtyp .
.DS
Default value:

    realtyp    = doubletyp;
.DE
The typename for procedures is given by the constant
.B voidtyp .
.DS
Default value:

    voidtyp    = 'void';
.DE
.ne 10
.IP i/o
The default fieldwidths for integer and real values written on textfiles
are given by the constants
.B intlen
and
.B fixlen .
.DS
Default value:

    intlen    = 10;
    fixlen    = 20;
.DE
.IP types
A table in the translator gives the mapping from Pascal
integer subrange types to C arithmetic types.
The table is initialized by code located at the end of procedure
.I initialize
giving the following default configuration:
.TS
l l l
r r l .
Low bound    High bound    Selected type
.sp
0    255    unsigned char
-128    127    char
0    65535    unsigned short
-32768    32767    short
-2147483647    2147483647    long
.TE