Re: Support for multiple sized pointers in a program

[Kevin Buettner <kevinb at redhat dot com>]

On Apr 1,  3:37pm, CUTHBERTSON,REVA (HP-Cupertino,ex3) wrote:

> I am porting gdb to a platform which allows both 32-bit and 64-bit pointers
> to coexist in a program.  I am running into a problem creating pointer types
> when 2 different sized pointers point to the same target type in the same
> program.  
> 
> The problem seems to be in make_pointer_type() in gdbtypes.c.  The function
> creates a pointer type for the target type if one does not exist, fills in
> the length of the pointer, fills in the flags and returns the pointer type.
> What it also does is save the new pointer type into the "pointer_type" field
> of the target type.  The next time I call this function for the same target
> type, it will return the pointer type I just made even if the next pointer
> to the target type is a different size.  This will be a problem when you try
> to fetch the value of a pointer, etc.
> 
> The problem appears to be the 1-1 mapping between the pointer type and the
> target type.  I made a change in make_pointer_type()  which disables the
> code that saves the pointer type in the "pointer_type" field of the target
> type.  I simply return the pointer type which ultimately gets attached to a
> symbol.  With this change, I'm able to print out pointers of different sizes
> even when they point to the same thing.  What I don't understand is why
> there is the 1-1 mapping in the first place between a type and a pointer
> type.  I see code all over gdb which calls lookup_pointer_type() on a type
> -- such as builtin_type_void and builtin_type_char.   What is the purpose
> for looking up the pointer type of a type that is not associated with a
> symbol?  I'm concerned that my change will have consequences on code that
> calls lookup_pointer_type() in this way.  Is my fix correct or should I be
> doing something a different way?
> 
> Any help on this would be greatly appreciated!!!!

Take a look at the "address class" support that I added last year.  I
needed it to handle this precise problem.  (Grep the gdb sources for
ADDRESS_CLASS.)

Kevin