Abstract Data Types

Hi,

possibly repeating myself, I want to emphasise the fact that information
hiding and encapsulation has been applied relatively well in the Aldor
compiler. Unfortunately, there are also many violations.

First, I would like to show how data abstraction is achieved within the
compiler source. As a case study, I will explain how it does so with the
Buffer and String datatypes.

The Buffer type is used as a serialisation target. Most notably, it is
used in byte code storage. Aldor Object files (.ao) contain platform
independent byte code. This independence is achieved by using the bufXXX
functions. A Buffer is represented by a record containing its data
pointer, its current position and its current capacity.

The String type is used for text. The file 'strops.h' defines several
operations on this type. It also defines a function that can create a
String object from a C string literal or a 'char*'. A String is
represented by its data pointer.

When operating on Buffer objects, the code mostly uses dedicated
functions or macros. In some places, however, it does not do so. Instead,
it operates directly on the pointee and its internal representation. In
almost all cases, unnecessarily. The size of a Buffer is defined as being
its current capacity. Buffer's member variable 'argc' represents this
attribute. The 'bufSize' macro abstracts this knowledge away. I changed
all such direct accesses to macro calls, so now there are no more direct
struct member accesses.

I want to do this with some more structures, but I ran into trouble while
doing that before. Suddenly, things started breaking. I am doing it step
by step, very slowly, to improve the compiler code without breaking it.

The String object used in the aldor compiler uses a lot of knowledge
about its actual representation. There is pointer arithmetic, assignment
of C string literals to String objects, index operations, all of which
are not possible without knowing that String is actually char*. By
abstracting this knowledge away, one opens a lot of room for self
checking, flow graphing, diagnostics and even optimisation of the
compiler itself.

My question now is, should I make String abstract and extend information
hiding to this type of objects? What would be disadvantages of such a
major change?

You may also want to read:
  http://xinutec.org/~pippijn/en/programming_data-hiding.xhtml
  http://xinutec.org/~pippijn/en/programming_c_strings.xhtml

Pippijn


_______________________________________________
Aldor-l mailing list
Aldor-l@...
http://aldor.org/mailman/listinfo/aldor-l_aldor.org