On Wed, 8 Nov 2023 at 21:45, Christoph Hellwig <hch@xxxxxxxxxxxxx> wrote:
>
> For my use case it'd treat it exactly like __bitwise except for also
> allowing arithmetics on it. Bonus for always allowing 0 withou explicit
> __force cast just like __bitwise.
It's too long for me to really remember, but I think that was the
*intention* of the "nocast" attribute originally.
The whole "nocast" thing goes back to pretty early in sparse (first
few weeks of it, in fact), and is almost entirely undocumented.
The commit that starts parsing it doesn't actually even mention it -
it mentions the *other* attributes it also starts parsing, but not
'nocast'.
Iirc, what happened was that one of the opriginal goals was that I
really wanted to have that "warn about any implicit integer casts",
and added "nocast" quite early, before the sparse type system was even
very strong. But it ended up being almost just a placeholder.
And then later on, Al came in, and he did the much stronger 'bitwise'
thing, which was immediately useful for all the byte order handling,
and 'nocast' ended up falling by the wayside.
Anyway, I think what you ask for is what 'nocast' was always supposed
to be, but just plain isn't.
Linus
