Nebula 1995 August/NEBULA.mdf/SourceCode/Classes/RCString/internal.string.rep

Text File | 1993-01-10 | 1.4 KB | 49 lines

It seems to me that the most simplistic internal string representation
for a reference counting string class is what I chose:

// internal string representation
struct srep {
    char *s;  // pointer to ASCIIZ data
    int   n;  // reference count
    int   l;  // string length
};

// basic object
@interface RCString : Object
{
    struct srep *p;
    BOOL yCaseSensitive;
}

 ...
@end

Possibly the interface should be:

@interface RCString : Object
{
    struct srep p[MAX_FRAGS];
    int  iCurrentFrags;
    BOOL yCaseSensitive;
}

 ...
@end

This would allow reference counting of substrings, it would eliminate some
of the bcopy()'s and free()'s necessary when inserting substrings or
extra characters.  It would also entail tracking how many fragments had been
created, coalescing fragments periodically and stringing them together in
a single ASCIIZ buffer on a -data message.  I don't want to go through testing
to decide what MAX_FRAGS should be: overall efficiency of this depends heavily
on choice of that number.

Another option would be to define struct srep as an object.  This would
allow programmers to plug in appropriate internal string representations
as long as these objects provided a few very basic methods.

Or you could have the instance variable(s) for the internal string rep
be a bit more elaborate, including function pointers to functions that
are knowledgable about the struct/internal string rep being used.  That
would avoid method call overhead.