Reserve enough mem that you have enough left over for the next reserve
Disable memtrim, because it breaks the assumption that reserve will not
change the exsting vector contents.

Tests are still hanging on linux. I think our mutex impl may be broken
when under contention.

Make sure all tests and benches actually use ralloc as their global allocator

See merge request !63

Master

Closes #45

See merge request !62

Fixed occasional failing of 'partial_realloc'. This is because our bookkeeper's vec gets too cramped and should (theoretically) need resizing

Fix sbrk to actually return the *old* break.

 Use new allocator API, and otherwise update for new Rust

- New allocator API
- Remove the "allocator" feature, which should be unnecessary due to how
the new allocator API works
- NonZero no longer implements Deref (https://github.com/rust-lang/rust/pull/41064)
- NonZero::new() returns an Option; use NonZero::new_unchecked()
- Thread locals are no longer 'static (https://github.com/rust-lang/rust/pull/43746)
- Changes to feature flags
- Use unsafe to access extern static (https://github.com/rust-lang/rust/issues/36247)

MacOSX is not supported anymore

Grammar fix

it is missing some needed syscalls.

fixes #37

by switching from the syscall crate to the sc crate. The sc crate is a
fork of the syscall crate that adds supports for several other
platforms.

* fix the build

the `unborrow!` macro is need in the block and bookkeeper modules

* cfg(tls) away impl LocalAllocator

as the struct LocalAllocator is not available when the tls feature is
disabled

To avoid linking hell.

David Teller authored 8 years ago and Jeremy Soller committed 8 years ago

This feature is now in Rust stable, so adding !#[feature(question_mark)] is now useless and causes a warning.

We make the `Leak` trait unsafe and consituting an invariant. Secondly,
we implement `Leak` for all `Copy` types.

We equip all `unsafe` blocks with a timestamped confirmations of an
audit. Whenever a block is changed, it should be removed until a new
audit is made.

The format of these messages is:

```
LAST AUDIT: yyyy-mm-dd (Username)
```

Close #28.

Various issues (like premature checks) were present in the byte count
beforehand. We fix these.

- Fix invalid byte counts.
- Add a `LOCAL_MEMTRIM_STOP` constant to `shim`, which serves for the
  minimum memtrimmed allocator (i.e. it will stop when the byte count is
  below this value).
- Add more logs.
- Add assertions about libc interference with SBRK.

There is still some problem when running

This commit is a complete revision of `shim`. Anything which is
configurable of nature is now moved to `shim` allowing complete
customizability of `ralloc`.

But it doesn't end there!

The logs got quite messy before due to the classical TMI problem. We
solve this by adding multiple "log levels" each describing a different
class of importance.

- Add multi-level logging.
- Move each component of shim into distinct modules.
- Add allocator canonicalization to `shim`.
- Add adjustable constant to `shim`.
- Add default OOM handler to `shim`.
- Add logging to everything.
- Remove the `libc_write` module and add it to `log` instead.
- Add colourful OOM message.
- Tweak the allocator canonicalization strategy.
- Rename the `log!` macro to `bk_log!`, and use the old name for the
  generalized logging macro.
- Remove the `unsafe_no_brk_lock` feature, due to being unsafe in
  multithreaded environment (and in singlethreaded ones,
  `unsafe_no_mutex_lock` is sufficient).
- Remove the `sys` module in favor of direct `shim` imports.

This commit introduces OS/BRK memtrimming, in which the memory from the
global allocator is freed to the OS when a certain limit is reached.
This avoids long-running processes from acquiring too much memory from
the system and never releasing it again.

- Implement safe BRK releasing.