Summary

There have been a few cases already of cases where the BSON code used by the driver is a little too happy-path oriented. Invalid BSON or documents seem to have a high chance of triggering panics or otherwise erroneous behavior.

Me and divjot.arora at some point investigated some issues in the BSON code for handling invalid documents that were sent by another user's fuzz testing of a library using the driver. And craig.wilson@mongodb.com has found a case where a negative length causes a panic in validation code.

In general, it seems like what we need is a fuzzer to more systematically find all of these edge cases and increase the robustness of the BSON package.

Implementing a fuzzer for BSON should (who knows, really) be easy to write for BSON, since it is just a lump of bytes.

Documentation

[Scope Document|]
[Technical Design Document|]