Simplify CharIndices.next

[jridgewell]

Char.len_utf8 is stable since #49698, making this a little easier to follow.

[rust-highfive]

Thanks for the pull request, and welcome! The Rust team is excited to review your changes, and you should hear from @rkruppe (or someone else) soon.

If any changes to this PR are deemed necessary, please add them as extra commits. This ensures that the reviewer can see what has changed since they last reviewed the code. Due to the way GitHub handles out-of-date commits, this should also make it reasonably obvious what issues have or haven't been addressed. Large or tricky changes may require several passes of review and changes.

Please see the contribution instructions for more information.

[Centril]

@bors rollup

[Centril]

cc @SimonSapin

[hanna-kruppe]

Thanks, LGTM!

@bors r+

[bors]

📌 Commit 7af83dc has been approved by rkruppe

[SimonSapin]

char::len_utf8 is implemented with three branches, which seems potentially more costly than the substraction that it replaces. Have you checked that this optimizes well? It’s possible that the optimizer realizes that this computation is redundant with next_code_point, but it’s not obvious.

[hanna-kruppe]

@bors r- (let's wait with merging until this performance question is cleared up)

[SimonSapin]

I don’t read assembly fluently enough to conclude anything from this:

• Before: https://rust.godbolt.org/z/_3EOcq
• After: https://rust.godbolt.org/z/ILP_Tc

[jridgewell]

I created a sample benchmark to compare the two implementations:

benchmark

#![feature(test)]externcrate test;use std::str::Chars;pubstructCharIndices<'a>{front_offset:usize,iter:Chars<'a>,}pubfnbefore(self_:&mutCharIndices) -> Option<(usize,char)>{let pre_len = self_.iter.as_str().len();match self_.iter.next(){None => None,Some(ch) => {let index = self_.front_offset;let len = self_.iter.as_str().len();
self_.front_offset += pre_len - len;Some((index, ch))}}}pubfnafter(self_:&mutCharIndices) -> Option<(usize,char)>{let ch = self_.iter.next()?;let index = self_.front_offset;
self_.front_offset += ch.len_utf8();Some((index, ch))}#[cfg(test)]mod tests {usesuper::*;use test::Bencher;#[bench]fnbefore(b:&mutBencher){let s = "ศไทย中华Việt Nam; Mary had a little lamb, Little lamb";let len = s.chars().count();
b.iter(|| {letmut chars = CharIndices{front_offset:0,iter: s.chars()};letmut i = 0;whileletSome(_) = super::before(&mut chars){
i += 1;}assert_eq!(i, len);});}#[bench]fnafter(b:&mutBencher){let s = "ศไทย中华Việt Nam; Mary had a little lamb, Little lamb";let len = s.chars().count();
b.iter(|| {letmut chars = CharIndices{front_offset:0,iter: s.chars()};letmut i = 0;whileletSome(_) = super::after(&mut chars){
i += 1;}assert_eq!(i, len);});}}

 Finished release [optimized] target(s) in 0.00s
Running target/release/deps/tmp-188dab753e97d367
running 2 tests
test tests::after ... bench: 44 ns/iter (+/- 3)
test tests::before ... bench: 44 ns/iter (+/- 3)
test result: ok. 0 passed; 0 failed; 0 ignored; 2 measured; 0 filtered out

It appears they're roughly the same speed? It might be that I'm just doing something wrong, the assembly definitely looks more complex the the after output above.

Before: https://rust.godbolt.org/z/sIJTXu
After: https://rust.godbolt.org/z/7bNVDu

[jridgewell]

Nevermind, when I updated the benchmark to actually use the index, it's slower:

benchmark using index

#![feature(test)]externcrate test;use std::str::Chars;pubstructCharIndices<'a>{front_offset:usize,iter:Chars<'a>,}pubfnbefore(self_:&mutCharIndices) -> Option<(usize,char)>{let pre_len = self_.iter.as_str().len();match self_.iter.next(){None => None,Some(ch) => {let index = self_.front_offset;let len = self_.iter.as_str().len();
self_.front_offset += pre_len - len;Some((index, ch))}}}pubfnafter(self_:&mutCharIndices) -> Option<(usize,char)>{let ch = self_.iter.next()?;let index = self_.front_offset;
self_.front_offset += ch.len_utf8();Some((index, ch))}#[cfg(test)]mod tests {usesuper::*;use test::Bencher;#[bench]fnbefore(b:&mutBencher){let s = "ศไทย中华Việt Nam; Mary had a little lamb, Little lamb";let len = s.len();
b.iter(|| {letmut chars = CharIndices{front_offset:0,iter: s.chars()};letmut i = 0;whileletSome((index, _)) = super::before(&mut chars){
i = index;}assert_eq!(i + 1, len);});}#[bench]fnafter(b:&mutBencher){let s = "ศไทย中华Việt Nam; Mary had a little lamb, Little lamb";let len = s.len();
b.iter(|| {letmut chars = CharIndices{front_offset:0,iter: s.chars()};letmut i = 0;whileletSome((index, _)) = super::after(&mut chars){
i = index;}assert_eq!(i + 1, len);});}}

 Compiling tmp v0.1.0 (/Users/jridgewell/tmp)
Finished release [optimized] target(s) in 0.45s
Running target/release/deps/tmp-188dab753e97d367
running 2 tests
test tests::after ... bench: 67 ns/iter (+/- 8)
test tests::before ... bench: 55 ns/iter (+/- 2)
test result: ok. 0 passed; 0 failed; 0 ignored; 2 measured; 0 filtered out