v2 data format is incompatible with other LMDB libraries such as py-lmdb

[kentbull]

I get the following malloc error when trying to open an LMDB database created with py-lmdb:

node(42074,0x1f6bde0c0) malloc: *** error for object 0x3c7e805c00e0: pointer being freed was not allocated
node(42074,0x1f6bde0c0) malloc: *** set a breakpoint in malloc_error_break to debug

When I downgrade to v1 database format then it works just fine.

What is the reason for using the non-backwards compatible v2 data format?

[kriszyp]

Because LMDB is prone to very poor performance when large blocks of data are used and free space becomes fragmented, as LMDB can become stuck in long searches for contiguous blocks of free space. The updated data format used by lmdb-js can track blocks of memory more efficiently, encoded with block lengths, and has a significantly optimized free space tracking that performs vastly better for these situations.

[kentbull]

This sounds like a very good upgrade for LMDB. Is this something you need help contributing back to LMDB core? It seems highly likely to be accepted in one form or another.

Nice work.