False positive in cast_ptr_alignment

[djugei]

if i cast a highly-aligned pointer to u8 to a mid-aligned pointer (in my case its 4096 -> 1 -> 8) clippy errors. this is wrong. there are many cases where casting from u8 is acceptable alignment-wise, including when you have manually checked that the alignment works out, statically or dynamically.

[flip1995]

Do you have a code example for this?

[djugei]

sure thing: https://github.com/djugei/chunk-list

[flip1995]

So a minimal example would like this:

let _ = (ptr as *const u8) as *const u64;

This may be ok for align_of(ptr) >= 8byte, which could be fixed for the example above. But then again, why would you cast to u8 first?

including when you have manually checked that the alignment works out, statically or dynamically.

In (special) cases where you disagree with a lint, or have manually checked that a lint is not applicable, you should use allow(clippy::lint) to signal that the following code is really intended.

IMO casts from u8 pointer is not a FP in general, but a special case where a cast to a more strictly aligned pointer is fine ➡️ allow.

[djugei]

why would you cast to u8 first

because i want to add an offset in bytes.

i honestly think this lint should be completely removed, clippy is not equiped to deal with this situation.
you can't really guess at the validity of pointer casts from a simple syntax analysis. i feel like this job is better suited for miri.

im reasonably sure that this lint will also error on the standard library, for example in slice

[flip1995]

you can't really guess at the validity of pointer casts from a simple syntax analysis

That's a fair point. I don't think we should remove it completely, but rather move it to pedantic

[djugei]

i am not sure what further discussion is needed, i will open a pr.

[flip1995]

Thanks! I marked it as need-discussion, to have it over for some more opinions. But this was 10 days ago, so I think moving this is fine.