Reasons for the fork

Here we expose the reasons for the fork on haskell-lmdb.

New exported functions

As we need to be able to resize the environment in order to not pre-allocate too much space and to be able to grow the database afterwards, we make use of the mdb_env_set_mapsize function so we needed to expose it. Please refer to the documentation of that function to check its behavior.

We also need to close and reopen the database handlers so we needed to expose the mdb_dbi_close' function. Once again, check the documentation for more information.

Writer mutex changes

A bug in lmbd-simple can cause a write transaction to be aborted twice when a put fails. This can lead to a segmentation fault on the next write transaction, because LMDB internally reuses a single write transaction, yet the transaction is freed if it is aborted/committed twice. However, haskell-lmdb will deadlock if the transaction is aborted or committed twice, since it uses an MVar mutex to prevent concurrent writes.

To make things a little bit safer, we now check that the thread holds the MVar mutex before committing or aborting a write transaction. Thus, an exception is thrown before a double commit/abort can happen.