Is .size_hint().1 ever used?

scottmcm · 2018-08-13T05:54:14.797Z

I tried to make use of it in this PR, but was unsuccessful:

github.com/rust-lang/rust

Try to guess a smarter initial capacity in Vec::from_iter

by scottmcm on 11:05AM - 05 Aug 18

2 commits changed 1 files with 13 additions and 11 deletions.

Out of curiosity, I started looking through rust to see if it’s ever actually used. And I found lots of things like

github.com

rust-lang/rust/blob/ab93561b5fa54954159480ddc10bbb69f015e539/src/liballoc/collections/binary_heap.rs#L1180-L1182


let iterator = iter.into_iter();
let (lower, _) = iterator.size_hint();

github.com

rust-lang/rust/blob/ab93561b5fa54954159480ddc10bbb69f015e539/src/librustc_data_structures/small_vec.rs#L172-L174


let iter = iter.into_iter();
self.reserve(iter.size_hint().0);
match self.0 {

github.com

rust-lang/rust/blob/ab93561b5fa54954159480ddc10bbb69f015e539/src/liballoc/vec.rs#L2639-L2641


// There may be more elements. Use the lower bound as an estimate.
// FIXME: Is the upper bound a better guess? Or something else?
let (lower_bound, _upper_bound) = self.replace_with.size_hint();

But not a single case of something actually using the upper bound. (There were lots of passing it along or validity-checking it, but never doing something useful with it.)

Have you ever found a use for it?

kornel · 2018-08-14T12:09:02.911Z

That’s a shame it isn’t used.

Using something like upper_bound.min(lower_bound * 2) as the initial capacity for collect could save a reallocation.

withoutboats · 2018-08-14T13:09:25.646Z

The lower bound is used to guide allocations for collecting. I don’t know why this is, I’d guess to avoid massive overallocations when the upper bound is far too vague.

The higher bound is used in TrustedLen, but then, its the same value as the lower bound in that case.

scottmcm · 2018-08-14T22:06:04.046Z

kornel:

Using something like upper_bound.min(lower_bound * 2) as the initial capacity for collect could save a reallocation.

Yup, tried that

github.com

scottmcm/rust/blob/aa080f40413693e9bb8dd55e645c47984952a517/src/liballoc/vec.rs#L1864-L1868


let guess = if let Some(upper) = upper {
    lower.saturating_mul(2).min(upper)
} else {
    lower
};

The perf results weren’t a clear win, though.

withoutboats:

The higher bound is used in TrustedLen , but then, its the same value as the lower bound in that case.

Yeah, ExactSizeIterator has a similar check, which of course isn’t really much of a use:

github.com

rust-lang/rust/blob/c0955a34bcb17f0b31d7b86522a520ebe7fa93ac/src/libcore/iter/traits.rs#L691-L695


// Note: This assertion is overly defensive, but it checks the invariant
// guaranteed by the trait. If this trait were rust-internal,
// we could use debug_assert!; assert_eq! will check all Rust user
// implementations too.
assert_eq!(upper, Some(lower));

system · 2019-03-25T08:30:41.686Z

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.