Valgrind thinks the slice's memory is leaked since no real pointer remains to it

[federicomenaquintero]

In librsvg we got this issue about a leak reported by Valgrind:

==6237== 40 bytes in 1 blocks are definitely lost in loss record 1,053 of 1,774
==6237==    at 0x483A7F3: malloc (in /usr/lib/x86_64-linux-gnu/valgrind/vgpreload_memcheck-amd64-linux.so)
==6237==    by 0x49F2BB6: alloc (alloc.rs:80)
==6237==    by 0x49F2BB6: alloc (alloc.rs:174)
==6237==    by 0x49F2BB6: allocate_in<selectors::parser::Component<rsvg_internals::css::Selector>,alloc::alloc::Global> (raw_vec.rs:152)
==6237==    by 0x49F2BB6: with_capacity_in<selectors::parser::Component<rsvg_internals::css::Selector>,alloc::alloc::Global> (raw_vec.rs:135)
==6237==    by 0x49F2BB6: with_capacity<selectors::parser::Component<rsvg_internals::css::Selector>> (raw_vec.rs:92)
==6237==    by 0x49F2BB6: with_capacity<selectors::parser::Component<rsvg_internals::css::Selector>> (vec.rs:358)
==6237==    by 0x49F2BB6: <alloc::vec::Vec<T> as alloc::vec::SpecExtend<T,I>>::from_iter (vec.rs:2073)
==6237==    by 0x4ADC6DF: from_iter<selectors::parser::Component<rsvg_internals::css::Selector>,smallvec::IntoIter<[selectors::parser::Component<rsvg_internals::css::Selector>; 2]>> (vec.rs:1981)
==6237==    by 0x4ADC6DF: collect<smallvec::IntoIter<[selectors::parser::Component<rsvg_internals::css::Selector>; 2]>,alloc::vec::Vec<selectors::parser::Component<rsvg_internals::css::Selector>>> (iterator.rs:1660)
==6237==    by 0x4ADC6DF: into_vec<[selectors::parser::Component<rsvg_internals::css::Selector>; 2]> (lib.rs:1073)
==6237==    by 0x4ADC6DF: parse_negation<rsvg_internals::css::RuleParser,rsvg_internals::css::Selector> (parser.rs:1987)
==6237==    by 0x4ADC6DF: selectors::parser::parse_functional_pseudo_class (parser.rs:2088)

Here, selectors is the selectors crate from Servo; it uses thin-slice for a few things.

From servo/components/selectors/parser.rs:

fn parse_negation<'i, 't, P, Impl>(
    parser: &P,
    input: &mut CssParser<'i, 't>,
) -> Result<Component<Impl>, ParseError<'i, P::Error>>
{
    ...
    Ok(Component::Negation(
        sequence.into_vec().into_boxed_slice().into(),
    ))
}

Component::Negation is a ThinBoxedSlice:

pub enum Component<Impl: SelectorImpl> {
    // ...
    Negation(ThinBoxedSlice<Component<Impl>>),
    // ...
}

I think Valgrind sees the malloc from the call to sequence.into_vec(), then it gets turned to a boxed slice, then to a ThinBoxedSlice. But since that erases the pointer from malloc, Valgrind thinks it's leaked.

So... this is not a real bug in thin-slice, just a bad interaction with Valgrind. I don't know if it's possible to have a suppression file for it.