v0.2.0

Changelog

• Exposes a high-level virtual WAL interface
• Allow general purpose allocation functions to be customized for the process (via calicodb::configure() API)
• Correct the cursor history during upward rebalancing passes, then skip straight to the correct record afterward
• Fixes some concurrency issues and defers creation of the WAL until the first write transaction
• Go back to using memcpy and memcmp for the WAL index headers
  • Was copying in chunks of atomic uint64_t, but this takes too long
  • Causes readers to fail to read the header too frequently, slowing down the process of connecting
• Fixes issues relating to 32-bit platforms, and files larger than 4 GiB
• Adds CALICODB_DEBUG_DELAY macro (sometimes adds a random-length delay in debug builds) for provoking synchronization problems
• Got the file format fuzzer to work

v0.1.1

Fixes WAL File::sync() behavior

• Fixes a problem in wal.c that would have likely resulted in data loss/corruption
• Added some tests to test_crashes.cpp that make sure File::sync() is being called at the correct locations, and that any unsync'd data can be lost without corrupting the database
• Added a fuzzer, db_format_fuzzer.cpp that tries to run a transaction on a database file made up of fuzzer input data
  • This will need a lot more work: the fuzzer produces crashes quickly (when provided a corpus of valid database files to start with)
  • Fixed a few easy corruption problems revealed by the new fuzzer
• Added to the design docs

v0.1.0

First pre-release 🐱