fix: reset pointer to the end of chunk not start

[my4ng]

Fix #314.

In the case where (try_)alloc_try_with has an Err result and it was the last allocation with a different chunk than the rewind one, the current_ptr was erroneously set to the start of the chunk rather than the end, which implies that the chunk has no more capacity left for future allocation, since we are bumping upwards. This causes the allocation doubling since it keeps getting new chunks.

I suspect this might be from when the bump direction was changed.

[fitzgen]

Thanks! Can you add a regression test to tests/all/try_alloc_try_with.rs?

[fitzgen]

Thanks! One last thing below

[fitzgen]

Can this test be simplified into the following?

Suggested change

	let bump = Bump::new();
	bump.alloc_layout(Layout::from_size_align(139040, 128).unwrap());
	bump.alloc_layout(Layout::from_size_align(139040, 128).unwrap());
	bump.alloc_layout(Layout::from_size_align(139040, 128).unwrap());
	for _ in 0..36 {
	bump.alloc_try_with(|| Err::<[u8; 128], _>(())).unwrap_err();
	}
	// only ONE new chunk should have been allocated
	assert_eq!(bump.allocated_bytes(), 974656);
	let bump = Bump::new();
	bump.alloc_layout(Layout::from_size_align(139040, 128).unwrap());
	bump.alloc_layout(Layout::from_size_align(139040, 128).unwrap());
	bump.alloc_layout(Layout::from_size_align(139040, 128).unwrap());
	for _ in 0..36 {
	bump.try_alloc_try_with(|| Err::<[u8; 128], _>(()))
	.unwrap_err();
	}
	// only ONE new chunk should have been allocated
	assert_eq!(bump.allocated_bytes(), 974656);
	let mut bump = Bump::with_capacity(100);
	assert_eq!(bump.iter_allocated_chunks().count(), 1);
	type R = Result<[u8; 99], ()>;
	assert_eq!(mem::size_of::<R>(), 100);
	for _ in 0..10 {
	bump.alloc_try_with(|| R::Err(())).unwrap_err();
	}
	assert_eq!(bump.iter_allocated_chunks().count(), 1);
	bump.alloc_try_with(|| R::Ok(Default::default())).unwrap();
	assert_eq!(bump.iter_allocated_chunks().count(), 1);

[my4ng]

Unfortunately the suggested change doesn't fail when the fix is reverted, however I have minimised the test case and taken some of the ideas here.