Don wrote:
Kagamin wrote:
Some people used to use unsigned integers as 'mere' non-negative
numbers, but they're actually not numbers from range, they're numbers
from entirely different algebra. And - yes - this causes subtle bugs.
Amen!
I actually think it's worse in D, because 'uint' is so easy to type,
it's far more seductive than 'unsigned int'.
Recent introduction of integer range promotions works hard in hiding
those bugs, so they manifest only in very rare corner cases.
It's not complete. It doesn't apply to arithmetic operations yet. Once
that's in place, D could become really harsh about mixing signed and
unsigned: if there's any chance the highest bit is set in the signed
type, mixing signed and unsigned should be illegal.
Ever wondered wheter a-=b and a+=-b equivalent? In most cases they
are, except for one. Do you remember? uint is not propagated to long.
I've hit it recently. It's exactly because unsigneds are not just
non-negative numbers. So it's pain to see their wide use as numbers
for which sign-sensitive arithmetic operations are meaningful.
I think it also depend on the representation of that number, for example
most numbers I would feel more natural as hex value such bit fields and
flags, address offsets, and whatnot are all unsigned, almost everything
else is signed.
I did get annoyed about uint not automatically propagated to long
however, when I was reading lo/hi uint offset pairs from a file and
making it a long offset (lo + (hi<<32)), turns out it can't be done
without explicit long casts.