Steve Atkins <steve@blighty.com> writes:
> The cidr type, including it's external interface, is simply broken.
That is a large claim that I don't think you have demonstrated.
The only one of your examples that seems to me to contradict the
documentation is this one:
steve=# select '224.0.0.0'::cidr; cidr ------------- 224.0.0.0/4
which should be /32 according to what the docs say:
: If y is omitted, it is calculated using assumptions from the older
: classful network numbering system, except that it will be at least large
: enough to include all of the octets written in the input.
The bogus netmask is in turn responsible for this case:
steve=# select '224.10.0.0'::cidr; ERROR: invalid cidr value: "224.10.0.0" DETAIL: Value has bits set to right of
mask.
Looking at the source code, there seems to be a special case for "class D"
network numbers that causes the code not to extend y to cover the
supplied inputs:
/* If no CIDR spec was given, infer width from net class. */ if (bits == -1) { if (*odst >= 240) /*
ClassE */ bits = 32; else if (*odst >= 224) /* Class D */ bits = 4; else if (*odst
>=192) /* Class C */ bits = 24; else if (*odst >= 128) /* Class B */ bits = 16;
else /* Class A */ bits = 8; /* If imputed mask is narrower than specified octets,
widen.*/ if (bits >= 8 && bits < ((dst - odst) * 8)) ^^^^^^^^^ bits = (dst - odst) * 8; }
I think the test for "bits >= 8" should be removed. Does anyone know
why it's there?
regards, tom lane