Re: Pascal Calling Convention

[Phil Clayton <phil.clayton@lineone.net>]

On Mar 29, 1:23 pm, stefan-lu...@see-the.signature wrote:
> The original C++ from Hyman Rosen was
>
> template <unsigned N>
> class set
> {
>     unsigned n;    // number of members in set
>     unsigned d[N]; // members are in d[0..n-1]
>     unsigned s[N]; // d[s[k]] == k if k is a member
>     // unspecified elements of d and s are arbitrary
>
> public:
>     set() : n(0) { }
>     bool has(unsigned k)
>         { return k < N && s[k] < n && d[s[k]] == k; }
>     void add(unsigned k)
>         { if (k < N && !has(k)) { d[n] = k; s[k] = n++; }
>     void del(unsigned k)
>         { if (has(k)) { unsigned i = s[k]; d[i] = d[--n]; s[d[i]] = i; }
>     void clr() { n = 0; }
>
> };

Interesting.  It's worth noting that if N is equal to the number of
keys representable by unsigned and every key is added to the set, n++
in the final add will 'overflow' because n (of type unsigned) needs to
be able to have the range 0 .. N, i.e. N + 1 distinct values.  I doubt
this would cause a problem in practice because
1. unsigned is likely to be at least 32 bits these days(?) so N will
be much smaller;
2. these are designed for sparse arrays, there is a low probability of
filling the entire array.

Clearly I'm not the first person to notice this though...

On Mar 29, 1:23 pm, stefan-lu...@see-the.signature wrote:
>    type Elements is mod <>;
>    ...
>    subtype Counter_Type is Integer range 0 .. Integer(Elements'Last)+1;

Phil