Add documentation of compile-time width function alternative.

[dcsommer]

While the PrimInt trait itself doesn't have compile time width functions, there are other ways to find the width of the underlying type. It is useful to present the alternative.

[dcsommer]

Yeah, I can qualify counting bytes vs bits in the text. Typically the size of integers is specified by bytes, not bits, however. Do you think all instances of the pattern T::zero().count_zeros() should be rewritten in terms of core::mem::size_of::<T>() instead? I could modify the text to that effect if you like.

On Sat, Jan 4, 2020 at 9:51 PM Josh Stone [email protected] wrote:

@cuviper commented on this pull request.

In src/int.rs https://github.com/rust-num/num-traits/pull/147#discussion_r363071190:

@@ -22,7 +22,7 @@ use {Num, NumCast}; /// /// All PrimInt types are expected to be fixed-width binary integers. The width can be queried /// via T::zero().count_zeros(). The trait currently lacks a way to query the width at -/// compile-time. +/// compile-time, but core::mem::size_of::<T>() can be used in the meantime.

Should we say 8 * size_of() so it counts bits?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/rust-num/num-traits/pull/147?email_source=notifications&email_token=AAIIKIFWDUD6NOK6B3HXQZDQ4FYNRA5CNFSM4KB5PIX2YY3PNVWWK3TUL52HS4DFWFIHK3DMKJSXC5LFON2FEZLWNFSXPKTDN5WW2ZLOORPWSZGOCQVUZOY#pullrequestreview-338382011, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAIIKIBZLXCYL44UOP5YLMDQ4FYNRANCNFSM4KB5PIXQ .

[cuviper]

Typically the size of integers is specified by bytes, not bits, however.

You think so? The bit width is right in the name: i32, u8, etc., and C's stdint.h is similar.

Do you think all instances of the pattern T::zero().count_zeros() should be rewritten in terms of core::mem::size_of::<T>() instead?

AFAICS "all" is just this one instance, but sure, I'd be OK with preferring size_of since it's const.

[dcsommer]

I just remembered that in the case of dyn T we can't use core::mem::size_of<T>, so there is definitely a use case for both runtime and compile-time width calculation. I'll do it in terms of bits too.

On Tue, Jan 7, 2020 at 12:02 PM Josh Stone [email protected] wrote:

Typically the size of integers is specified by bytes, not bits, however.

You think so? The bit width is right in the name: i32, u8, etc., and C's stdint.h is similar.

Do you think all instances of the pattern T::zero().count_zeros() should be rewritten in terms of core::mem::size_of::<T>() instead?

AFAICS "all" is just this one instance, but sure, I'd be OK with preferring size_of since it's const.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/rust-num/num-traits/pull/147?email_source=notifications&email_token=AAIIKIGSIKUGNGXG2F5GNV3Q4TNVFA5CNFSM4KB5PIX2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEIKC5EQ#issuecomment-571747986, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAIIKIE4ZMMSJQS2A4GOCHDQ4TNVFANCNFSM4KB5PIXQ .

[cuviper]

How does dyn T apply in this context though? We already require PrimInt: Sized.

[dcsommer]

I mean when dealing with a PrimInt trait object, we can't know the size of the type at compile time. For instance, in code dealing with a Box<dyn PrimInt>, there isn't a way to use the const fn core::mem::size_of<T>() to get the concrete type's width at compile time. We need to dynamic dispatch on the trait object to calculate it.

On Tue, Jan 7, 2020 at 12:19 PM Josh Stone [email protected] wrote:

How does dyn T apply in this context though? We already require PrimInt: Sized.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/rust-num/num-traits/pull/147?email_source=notifications&email_token=AAIIKIDT6L5UK4DUHAEHJVDQ4TPVNA5CNFSM4KB5PIX2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEIKEMEQ#issuecomment-571754002, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAIIKIGXTOPD7J6UUMZJL53Q4TPVNANCNFSM4KB5PIXQ .

[cuviper]

PrimInt is not trait object safe -- you can't form any sort of dyn PrimInt at all.