Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Introduce try_reserve friends #33

Open
wants to merge 3 commits into
base: develop
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion src/drop.rs
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,7 @@ impl<T> Drop for MiniVec<T> {
} = core::ptr::read(self.buf.as_ptr().cast::<Header>());

core::ptr::drop_in_place(core::ptr::slice_from_raw_parts_mut(self.data(), len));
alloc::alloc::dealloc(self.buf.as_ptr(), make_layout::<T>(cap, alignment));
alloc::alloc::dealloc(self.buf.as_ptr(), make_layout(cap, alignment, core::mem::size_of::<T>()));
};
}
}
19 changes: 11 additions & 8 deletions src/impl/helpers.rs
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,10 @@ pub const fn next_capacity<T>(capacity: usize) -> usize {
};
}

2 * capacity
match capacity.checked_mul(2) {
Some(cap) => cap,
None => capacity,
}
Comment on lines +25 to +28
Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is a good catch

}

pub fn max_align<T>() -> usize {
Expand All @@ -31,13 +34,13 @@ pub fn max_align<T>() -> usize {
core::cmp::max(align_t, header_align)
}

pub fn make_layout<T>(capacity: usize, alignment: usize) -> alloc::alloc::Layout {
pub fn make_layout(capacity: usize, alignment: usize, type_size: usize) -> alloc::alloc::Layout {
Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It's not clear to me what we gain by removing the generic parameter here.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It removes number of instantiations for function, reducing compile times.
If you're curious std library does the same https://github.com/rust-lang/rust/blob/207c80f105282245d93024c95ac408c622f70114/library/alloc/src/raw_vec.rs#L442-L445

By extracting as much code out of generics, you improve compile times automatically equivalent to number of possible T in your code

Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hmm.

Alright, we can explore this later.

This PR should focus solely on adding the two new associated functions and the tests.

If we want to work on reducing code bloat, we should first get some metrics set up and then we'll know if our refactors are actually doing anything or not.

let header_size = core::mem::size_of::<Header>();
let num_bytes = if capacity == 0 {
next_aligned(header_size, alignment)
} else {
next_aligned(header_size, alignment)
+ next_aligned(capacity * core::mem::size_of::<T>(), alignment)
+ next_aligned(capacity * type_size, alignment)
};

alloc::alloc::Layout::from_size_align(num_bytes, alignment).unwrap()
Expand Down Expand Up @@ -83,14 +86,14 @@ mod tests {
fn make_layout_test() {
// empty
//
let layout = make_layout::<i32>(0, max_align::<i32>());
let layout = make_layout(0, max_align::<i32>(), core::mem::size_of::<i32>());

assert_eq!(layout.align(), core::mem::align_of::<Header>());
assert_eq!(layout.size(), core::mem::size_of::<Header>());

// non-empty, less than
//
let layout = make_layout::<i32>(512, max_align::<i32>());
let layout = make_layout(512, max_align::<i32>(), core::mem::size_of::<i32>());
assert!(core::mem::align_of::<i32>() < core::mem::align_of::<Header>());
assert_eq!(layout.align(), core::mem::align_of::<Header>());
assert_eq!(
Expand All @@ -100,7 +103,7 @@ mod tests {

// non-empty, equal
//
let layout = make_layout::<i64>(512, max_align::<i64>());
let layout = make_layout(512, max_align::<i64>(), core::mem::size_of::<i64>());
assert_eq!(
core::mem::align_of::<i64>(),
core::mem::align_of::<Header>()
Expand All @@ -112,7 +115,7 @@ mod tests {
);

// non-empty, greater
let layout = make_layout::<OverAligned>(512, max_align::<OverAligned>());
let layout = make_layout(512, max_align::<OverAligned>(), core::mem::size_of::<OverAligned>());
assert!(core::mem::align_of::<OverAligned>() > core::mem::align_of::<Header>());
assert_eq!(layout.align(), core::mem::align_of::<OverAligned>());
assert_eq!(
Expand All @@ -124,7 +127,7 @@ mod tests {
);

// non-empty, over-aligned
let layout = make_layout::<i32>(512, 32);
let layout = make_layout(512, 32, core::mem::size_of::<i32>());
assert_eq!(layout.align(), 32);
assert_eq!(
layout.size(),
Expand Down
192 changes: 162 additions & 30 deletions src/lib.rs
Original file line number Diff line number Diff line change
Expand Up @@ -77,6 +77,64 @@ use crate::r#impl::splice::make_splice_iterator;

pub use crate::r#impl::{Drain, DrainFilter, IntoIter, Splice};

#[inline]
#[allow(clippy::cast_ptr_alignment)]
fn is_default(buf: core::ptr::NonNull<u8>) -> bool {
core::ptr::eq(buf.as_ptr(), &DEFAULT_U8)
}
Comment on lines +80 to +84
Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't understand the purpose of this refactor either.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

To use outside of generic context

Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

To try and reduce code bloat again?


#[derive(Debug, Clone, Eq, PartialEq)]
//TODO: Switch to https://doc.rust-lang.org/alloc/collections/enum.TryReserveErrorKind.html once stable
/// The error type for `try_reserve` methods.
pub struct TryReserveError {
}

struct Grow {
buf: core::ptr::NonNull<u8>,
old_capacity: usize,
new_capacity: usize,
alignment: usize,
len: usize,
type_size: usize,
}
Comment on lines +92 to +99
Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't get why we have a Grow struct that also contains a type_size.

I'm not sure I understand this refactoring in general.

We should basically just update fn grow() to return a Option<()> instead of what we're doing now which is invoking handle_alloc_error.

To this end, the current usages just need to match on grow()'s new return type and invoke handle_alloc_error there and then the try_reserve_* associated functions can match on the None and return the Result type.

This should be largely good enough because any other allocation failures (for example, our layout is invalid) would be caused by bugs in the library.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This Grow struct here is to just to make sure not to make mistake when passing arguments to grow function as I moved it outside of Vec as part of refactoring.

This is ofc your choice if you do not need refactoring to remove unneeded generic code.
As you said it would be simply done by refactoring original grow function

Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

To me it seems silly to create a struct just to give it an impl with a single associated function.

This pattern, imo, fits better for DropGuard-like types.


impl Grow {
#[inline]
fn grow(self) -> Result<Option<core::ptr::NonNull<u8>>, alloc::alloc::Layout> {
if self.new_capacity == self.old_capacity {
return Ok(None);
}

let new_layout = make_layout(self.new_capacity, self.alignment, self.type_size);

let new_buf = if is_default(self.buf) {
unsafe { alloc::alloc::alloc(new_layout) }
} else {
let old_layout = make_layout(self.old_capacity, self.alignment, self.type_size);

unsafe { alloc::alloc::realloc(self.buf.as_ptr(), old_layout, new_layout.size()) }
};

match core::ptr::NonNull::new(new_buf) {
Some(new_buf) => {
let header = Header {
len: self.len,
cap: self.new_capacity,
alignment: self.alignment,
};

#[allow(clippy::cast_ptr_alignment)]
unsafe {
core::ptr::write(new_buf.cast::<Header>().as_ptr(), header);
}

Ok(Some(new_buf))
},
None => Err(new_layout),
}
}
}

/// `MiniVec` is a space-optimized implementation of `alloc::vec::Vec` that is only the size of a single pointer and
/// also extends portions of its API, including support for over-aligned allocations. `MiniVec` also aims to bring as
/// many Nightly features from `Vec` to stable toolchains as is possible. In many cases, it is a drop-in replacement
Expand Down Expand Up @@ -130,9 +188,8 @@ fn header_clone() {
static DEFAULT_U8: u8 = 137;

impl<T> MiniVec<T> {
#[allow(clippy::cast_ptr_alignment)]
fn is_default(&self) -> bool {
core::ptr::eq(self.buf.as_ptr(), &DEFAULT_U8)
is_default(self.buf)
}

fn header(&self) -> &Header {
Expand Down Expand Up @@ -167,41 +224,44 @@ impl<T> MiniVec<T> {
fn grow(&mut self, capacity: usize, alignment: usize) {
debug_assert!(capacity >= self.len());

let old_capacity = self.capacity();
let new_capacity = capacity;

if new_capacity == old_capacity {
return;
}

let new_layout = make_layout::<T>(new_capacity, alignment);

let len = self.len();

let new_buf = if self.is_default() {
unsafe { alloc::alloc::alloc(new_layout) }
} else {
let old_layout = make_layout::<T>(old_capacity, alignment);

unsafe { alloc::alloc::realloc(self.buf.as_ptr(), old_layout, new_layout.size()) }
let grower = Grow {
buf: self.buf,
old_capacity: self.capacity(),
new_capacity: capacity,
alignment,
len: self.len(),
type_size: core::mem::size_of::<T>(),
};

if new_buf.is_null() {
alloc::alloc::handle_alloc_error(new_layout);
match grower.grow() {
Ok(Some(new_buf)) => {
self.buf = new_buf;
},
Ok(None) => (),
Err(new_layout) => alloc::alloc::handle_alloc_error(new_layout),
}
}

let header = Header {
len,
cap: new_capacity,
alignment,
fn try_grow(&mut self, capacity: usize, alignment: usize) -> Result<(), TryReserveError> {
debug_assert!(capacity >= self.len());

let grower = Grow {
buf: self.buf,
old_capacity: self.capacity(),
new_capacity: capacity,
alignment,
len: self.len(),
type_size: core::mem::size_of::<T>(),
};

#[allow(clippy::cast_ptr_alignment)]
unsafe {
core::ptr::write(new_buf.cast::<Header>(), header);
match grower.grow() {
Ok(Some(new_buf)) => {
self.buf = new_buf;
Ok(())
},
Ok(None) => Ok(()),
Err(_) => Err(TryReserveError {}),
}

self.buf = unsafe { core::ptr::NonNull::<u8>::new_unchecked(new_buf) };
}

/// `append` moves every element from `other` to the back of `self`. `other.is_empty()` is `true` once this operation
Expand Down Expand Up @@ -1056,6 +1116,78 @@ impl<T> MiniVec<T> {
self.grow(total_required, self.alignment());
}

/// `try_reserve` ensures there is sufficient capacity for `additional` extra elements to be either
/// inserted or appended to the end of the vector. Will reallocate if needed otherwise this
/// function is a no-op.
///
/// Guarantees that the new capacity is greater than or equal to `len() + additional`.
///
/// # Errors
///
/// In case of failure, old capacity is retained.
///
/// # Example
///
/// ```
/// let mut vec = minivec::MiniVec::<i32>::new();
///
/// assert_eq!(vec.capacity(), 0);
///
/// vec.try_reserve(128).expect("Successfully allocate extra memory");
///
/// assert!(vec.capacity() >= 128);
/// ```
///
pub fn try_reserve(&mut self, additional: usize) -> Result<(), TryReserveError> {
let capacity = self.capacity();
let total_required = match self.len().checked_add(additional) {
Some(total_required) => total_required,
None => return Err(TryReserveError {}),
};

if total_required <= capacity {
return Ok(());
}

let mut new_capacity = next_capacity::<T>(capacity);
while new_capacity < total_required {
new_capacity = next_capacity::<T>(new_capacity);
}

self.try_grow(new_capacity, self.alignment())
}

/// `try_reserve_exact` ensures that the capacity of the vector is exactly equal to
/// `len() + additional` unless the capacity is already sufficient in which case no operation is
/// performed.
///
/// # Errors
///
/// In case of failure, old capacity is retained.
///
/// # Example
///
/// ```
/// let mut vec = minivec::MiniVec::<i32>::new();
/// vec.try_reserve_exact(57).expect("Oi, no capacity");
///
/// assert_eq!(vec.capacity(), 57);
/// ```
///
pub fn try_reserve_exact(&mut self, additional: usize) -> Result<(), TryReserveError> {
let capacity = self.capacity();

let total_required = match self.len().checked_add(additional) {
Some(total_required) => total_required,
None => return Err(TryReserveError {}),
};
if capacity >= total_required {
return Ok(());
}

self.try_grow(total_required, self.alignment())
}

/// `resize` will clone the supplied `value` as many times as required until `len()` becomes
/// `new_len`. If the current [`len()`](MiniVec::len) is greater than `new_len` then the vector
/// is truncated in a way that's identical to calling `vec.truncate(new_len)`. If the `len()`
Expand Down
49 changes: 49 additions & 0 deletions tests/mine.rs
Original file line number Diff line number Diff line change
Expand Up @@ -1262,3 +1262,52 @@ fn minivec_assume_minivec_init() {
assert_eq!(bytes[0], 137);
assert_eq!(bytes[511], 137);
}

#[test]
fn minivec_should_fail_try_reserve_impossible() {
let mut buf = minivec::mini_vec![0; 512];
buf.try_reserve(usize::max_value()).expect_err("FAIL");
}
Comment on lines +1266 to +1270
Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This test is incomplete.

We've tested failure which is good but we need to test the success arm as well.

I also don't see try_reserve_exact() here either.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is good but we also need to comment back in the tests in vec.rs as well. Seems like there's a good bit of tests the stdlib has there.


#[test]
fn minivec_should_fail_try_reserve_exact_impossible() {
let mut buf = minivec::mini_vec![0; 512];
buf.try_reserve_exact(usize::max_value()).expect_err("FAIL");
}

#[test]
fn minivec_test_try_reserve() {
let mut v = MiniVec::new();
assert_eq!(v.capacity(), 0);

v.try_reserve(2).expect("to try 2");
assert!(v.capacity() >= 2);

for i in 0..16 {
v.push(i);
}

assert!(v.capacity() >= 16);
v.try_reserve(16).expect("to try 16");
assert!(v.capacity() >= 32);

v.push(16);

v.try_reserve(16).expect("to try 16 again");
assert!(v.capacity() >= 33)
}

#[test]
fn minivec_overaligned_allocations_with_try_reserve_exact() {
#[repr(align(256))]
struct Foo(usize);
let mut v = mini_vec![Foo(273)];
for i in 0..0x1000 {
v.try_reserve_exact(i).expect("try reserve exact");
assert!(v[0].0 == 273);
assert!(v.as_ptr() as usize & 0xff == 0);
v.shrink_to_fit();
assert!(v[0].0 == 273);
assert!(v.as_ptr() as usize & 0xff == 0);
}
}