C Changes

This is the contents of the 'C Changes' help file, which is believed to refer to Typesetter C.

Type:

  >help c_old

to find out about older changes to C

****************************************************

 Changes to C preprocessor:

1. Defines may be continued from line to line with
   backslash-newline

2. You can undefine things with # undef xxx

3. You can undefine standard things from the
   command line with
 	cc -Uxxx
   (like cc -Dxxx, but reversed).

4. The <> brackets on include now have a distinct
   meaning again from ""; they do not search the
   local directory, but do search -I and standard places.

5. There is a new preprocessor command
 	# if expression
   where expression should be an integer expression
   that can be evaluated at compile time.  Defined
   things may be used.  Strings that are not integers
   are either errors, or if they are undefined names
   are taken as zero.  Strings defined as themselves
   are 1 (i.e. # if unix works).  Strings defined as
   anything else are expanded and retried.  There are
   no assignments, floating point, pointers, etc.
   but the full range of integer operators work.

6. There is another new command # else.

*************************************

Much new stuff in C.

(1)	The notion of a `union' type is introduced.
The type is spcified by the construction

 	union�����_____ { ... }
or
 	union�����_____ name { ... }

which is isomorphic to a structure declaration.  The declaration

 	union { int i; double d;} u;

makes u a cell which may hold either an integer i or a double d.
To name the double, use `u.d'; to name the integer,
use `u.i'.  It is undefined to store into u.d and then to access
u.i, and vice versa.

The actual amount of storage allocated is the maximum required by the 
members.  This facility is intended to solve some
problems encountered in dealing with data structures
such as trees, wherea node may contain pointers
either to other nodes or to leaves.
You will not go far wrong if you think of a union
declaration as a structure declaration
in which all the fields have offset 0 from the
start of the structure.

(2)	Generalized conversion operators have been introduced.
The syntax is

 	(type)expression

where `type' is the name of a type to which the expression is
converted.  The semantics of the conversion are the same
as if the expression were assigned to a variable of the
named type.

The type is specified by giving a type keyword (like `int'
etc.) possibly followed by a declarator (in the sense
of the C manual) except that the name is left out.
In other words, the whole thing looks like a declaration of
a single variable with the variable deleted.
For example:

 	(int)3.14159	convert to integer
 	(int *) p		convert to integer pointer
 	(int ())		convert to function (meaningless, but
 				that's how it would be said)
 	(int (*)())p		convert to pointer to functiion

The (type) construction behaves just like an ordinary unary
operator in precedence and is likewise right-associative.

(3)	The (type) construction can also be used in the form

 	sizeof(type)

which is a (compile-time) number equal to the number of bytes of
storage occupied by an object of the named type.

(4)	It is now possible, and in fact encouraged, to write
initializations using an '=', thus:

int	x = 10;
double	y[] = {1,2,3};

There is no change in meaning from what was previously available.

(5)	The assignment-type operators may now be written with the
binary suboperator first and the '=' second:

 	a += b;
 	a <<= b;

Doing things this way removes such lexical unpleasantness
as the expression 'a=-b' which is really ambiguous.
It would be nice to remove the older forms altogether,
but this won't happen instantly.

(6)	The treatment of the keyword `extern' at the top level
is more in line with GCOS and IBM compilers;
the declaration

 	extern int x;

does not reserve storage, and in fact no reference
to x is generated unless it is used.  (It will not
be loaded from a library just because of such a declaration).

(7)	This one may affect a number of programs; we'll see how it goes.
Using an undefined name in an initializer gets a warning.
It is a rather suspicious
sort of thing to do anyway (the compiler is forced to
impute a type to the name without sufficient information)
and it led to a bug in initialization inside functions
not fixable without adding a wart on the wart.

(8)	The following bugs in C have been fixed.

In the declaration
  f() register x; {...}
no diagnostic was produced, except an internal error
when x was later used.

There were two separate optimizer (c2) bugs which caused
c2 to loop (on rather strange input).

Most but not all #define names the same in their first eight characters
but different later were taken to be different.  To avoid
false confidence, only the first eight are now significant,
uniformly.

A function name with a subscript was not diagnosed.

Register declarations in inner blocks were sometimes unjustly
treated as plain auto declarations.

Arrays declared with typedef sometimes had the internal
notion of the array size wrong.

In certain initializations, a missing ".even" caused
an assembler "odd address" diagnostic.

Certain differences of rather strange pointer
expressions got an internal diagnostic.
(reported in Minisystems Newsletter).

unsigned =>> ...  compiled wrong code.

long = !long compiled wrong code.

Use of ?: in initializers led to an unjustified syntax complaint.

*************************************

1. The <> brackets in "include" statements are
   no longer needed.  Any include file is looked
   for first relative to the directory where the
   source file is; then relative to any directories
   named in -I arguments, in order; and then in
   /usr/include.

   In a few days we will start giving warning
   messages for # include <>.

2. You can now use -Dxx=yy on the command line
   to define 'xx' with a value.

*************************************

Several significant, but, it is hoped, upward compatible changes
in C have been installed.

1. Type `unsigned'.

A new fundamental data type with keyword `unsigned,' is
available.  It may be used alone:

        unsigned u;

or as an adjective with `int'

        unsigned int u;

with the same meaning.  There are not yet (or possibly ever)
unsigned longs or chars.  The meaning of an unsigned variable is
that of an integer modulo 2^n, where n is 16 on the PDP-11.  All
operators whose operands are unsigned produce results consistent
with this interpretation except division and remainder where the
divisor is larger than 32767; then the result is incorrect.  The
dividend in an unsigned division may however have any value (i.e.
up to 65535) with correct results.  Right shifts of unsigned
quantities are guaranteed to be logical shifts.

When an ordinary integer and an unsigned integer are combined
then the ordinary integer is mapped into an integer mod 2^16 and
the result is unsigned.  Thus, for example `u = -1' results in
assigning 65535 to u.  This is mathematically reasonable, and
also happens to involve no run-time overhead.

When an unsigned integer is assigned to a plain integer, an
(undiagnosed) overflow occurs when the unsigned integer exceeds
2^15-1.

It is intended that unsigned integers be used in contexts where
previously character pointers were used (artificially and
nonportably) to represent unsigned integers.

2.  Block structure.

A sequence of declarations may now appear at the beginning of any
compound statement in {}.  The variables declared thereby are
local to the compound statement.  Any declarations of the same
name existing before the block was entered are pushed down for
the duration of the block.  Just as in functions, as before, auto
variables disappear and lose their values when the block is left;
static variables retain their values.  Also according to the same
rules as for the declarations previously allowed at the start of
functions, if no storage class is mentioned in a declaration the
default is automatic.

Implementation of inner-block declarations is such that there is
no run-time cost associated with using them.

3.  Initialization

Declarations, whether external, at the head of functions, or
in inner blocks may have initializations whose syntax is the same
as previous external declarations with initializations.  The only
restrictions are that automatic structures and arrays may not be
initialized (they can't be assigned either); nor, for the moment
at least, may external variables when declared inside a function.

The declarations and initializations should be thought of as
occurring in lexical order so that forward references in
initializations are unlikely to work.  E.g.,

        { int a a;
          int b c;
          int c 5;
          ...
        }

Here a is initialized by itself (and its value is thus
undefined); b is initialized with the old value of c (which is
either undefined or any c declared in an outer block).