lmdb-zero is a near-zero-cost wrapper around LMDB designed to allow using the full range of features offered by LMDB while keeping it reasonably easy to write safe programs.
lmdb-zero is as much as possible a 1:1 mapping of the raw API, mainly
providing RAII constructs and integration into Rust's borrow checker to ensure
Zero-copy API. Reads return references into the memory-mapped file. Using
MDB_RESERVEto allocate space in the file and directly write to it is supported.
Cursors directly map to the same operations provided by LMDB, but in a typesafe manner.
Full integration with the borrow checker. Read references are checked to not outlive their transaction or overlap with a write in the same transaction.
Cursors and read transactions can be reset and reused.
The API is complete and reasonably stable and is believed to be sound insofar as Rust's unsafety rules are actually defined.
This crate has not been thoroughly tested on architectures with strong alignment constraints, though the tests pass on ARM7. While the conversion API checks for correct alignment by default, issues such as #27060 could come up, and it is of course possible there are bugs in handling alignment here.
Database methods to enable interoperating
with native C/C++ code.
Database::dbi() renamed to
as_raw(). The old name
is still available but deprecated.
0.4.3: Fix panic on
Cursor::get_multiple() if the current key has exactly
LmdbResultExt is now reexported from the crate root for better
0.4.2: Fix being unable to open databases in read-only environments. Fix
future-incompatibility warning arising from
0.4.1: Tests updated to work on Rust 1.20.
bitflags dependency updated.
Neither of these are expected to have any impact on external code.
0.4.0: Minor breaking changes.
WriteAccessor can now
be dropped and re-obtained. Most types now support additional
ownership/borrowing modes, which allows for dynamic lifetime management and
other possibilities. Upgrade to
liblmdb-sys 0.2.2 and
Fixes to documentation.
0.3.1: Metadata updates to reflect change of crate ownership. No software changes were made in this version.
0.3.0: Breaking Changes to the API, see section below. Migration is
expected to be easy for most use-cases. Slight performance improvement due to
ResetTransaction is now actually public, making that part of the
API more accessible. Add documentation for lifetimes.
0.2.1: Fix use-after-free when passing database name to
mdb_txn_abort after transaction commit fails.
0.2.0: Switch from
lmdb-sys to newer
0.1.0: Initial release.
Breaking Changes in 0.4.0
A number of functions which formerly took an
&SomeType parameter now take an
Into<Supercow<SomeType>>. For the vast majority of existing code, this has no
effect, but it could cause issues if older code was relying on an implicit
Deref call (for example, via
lazy_static!) to produce the correct reference
type. If this causes issues, explicitly writing the dereferencing is required.
For example, if your code originally had
lmdb::Database::open(&ENV, ...) where
ENV was declared via
lazy_static!, it would need to be changed to
WriteAccessor must now be strictly outlived by their
transactions. No practical cases where this would be an issue are apparent, but
if it comes up, code must be rearranged to ensure the accessor is dropped
before the transaction. Note that now one can drop the accessor and later
Breaking Changes in 0.3.0
lmdb::Error has been completely reworked. It is now an enum with the
lmdb-zero errors cleanly separated from native LMDB errors.
includes an error message.
FromLmdbBytes.from_lmdb_bytes() now returns a
Result<&Self, String> instead
Option. This is mainly to make alignment issues less subtle and point
people directly to advice on how to fix the problem, but should be able to make
other things clearer as well.
The mostly untested and somewhat questionable
lax_alignment feature has been
LmdbRaw now always enforces alignment requirements. Client code
which wishes to operate on misaligned values which cannot use the
#[repr(packed)] solutions will need to provide its own
The primitive types which have alignment requirements (eg,
u64) are no
LmdbRaw, as this made it too easy to write code depending on
happenstance to align the values correctly. Client code now must wrap them in
Unaligned to read them directly, or else provide its own unit structs if it
has other needs. Note that these types and their arrays are still
Unfortunately, as a side-effect of the above,
are no longer
LmdbOrdKey, but instead only
LmdbOrdKeyIfUnaligned. Wrapping these in
will work in most cases without overhead.
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.