Some bugs fixed --> new alpha GPC on agnes

Fri, 7 Feb 1997 03:40:22 +0100 (MET)

Hello, folks!

A snapshot of my current GPC source is now available via anonymous ftp
on agnes.dida.physik.uni-essen.de, directory gnu-pascal/alpha.  It is
a diff against gpc-2.0.

Changes:

  * Some cosmetical changes not worth to talk about.  ;-)

  * Patched the runtime library to allow conio operations and access
    to device files on the MSDOS platform.

  * Fixed wrong --pedantic warning for `rewrite'.

  * New command-line option `-dY' to dump the source to stderr.  This
    is useful when trying to figure out *where* gpc crashed ...

  * Fixed a really crazy bug concerning objects.  (See below.)

  * Handle assignment between sets of different sizes correctly.
    Beware:  When assigning a large set to a smaller one, it is truncated
    without warning or runtime check.  And there might still be errors in
    this part of the compiler.


The story about the "crazy" object-related bug follows.
It's hard to believe, but it's true - just check it with gpc-2.0!


    Program Test;


    Type
      TestObj = object
        FooBar: array [ 0..27 ] of Char;
      end (* TestObj *);


    Procedure Init ( x: Integer );

    begin (* Init *)
      x:= 7;
    end (* Init *);


    begin
      Init ( 3 );
    end.


This program compiles, but it crashes at runtime.  )-:

Change `FooBar' to be an `array [ 0..28 ]' of Char, and it works.
Note that the object is just a type declaration and everything which
is really executed has nothing to do with it.

I tried different upper bounds for `FooBar' from 0 to 100.  The crash
happens with 23..27.

The solution:  There was an error with variable initializers.

  :-)-:-(-:-)-:-(-:-)-:-(-:-)-:-(-:-)-:-(-:-)-:-(-:-)-:-(-:-)-:-(-:

To understand this, one must know about the internal structure of objects
in contrast to that of records.  While records have just their fields,
objects have an additional invisible "vmt" field which contains a pointer
to the "Virtual Method Table".  This table contains (a) the size of the 
object and (b) the addresses of all virtual methods.  `TestObj' above
does not have virtual methods, but it has a size - which is 32 bytes if
the upper bound of `FooBar' is 23..27 - thus it needs a VMT.  This VMT
is an initialized variable, and the initial value of 32 for the size of
the object triggered the error.  Voila!

It's fixed now.

As a side-effect, initializers for other structured variables are
more stable now, too:


    Var
      foo: record
        i: Integer;
        x: Real;
      end (* foo *) value ( 1; 3.1415 );

      bar: array [ 1..3 ] of Char value ( 'b', 'a', r' );


Okay, so let's go on!

Good night,  |-)

    Peter

  Dipl.-Phys. Peter Gerwinski, Essen, Germany, free physicist and programmer
peter.gerwinski@uni-essen.de - http://home.pages.de/~peter.gerwinski/ [970201]
       maintainer GNU Pascal - http://home.pages.de/~gnu-pascal/ [970125]

Peter Gerwinski (peter@agnes.dida.physik.uni-essen.de)