1.2.1

• Fix incorrect index being used when decoding into anonymous struct fields. PR #42 by Andy Bursavich.

1.2.0

• The database decoder now does bound checking when decoding data from the database. This is to help ensure that the reader does not panic when given a corrupt database to decode. Closes #37.
• The reader will now return an error on a data structure with a depth greater than 512. This is done to prevent the possibility of a stack overflow on a cyclic data structure in a corrupt database. This matches the maximum depth allowed by libmaxminddb. All MaxMind databases currently have a depth of less than five.

1.1.0

• Added appengine build tag for Windows. When enabled, memory-mapping will be
  disabled in the Windows build as it is for the non-Windows build. Pull
  request #35 by Ingo Oeser.
• SetFinalizer is now used to unmap files if the user fails to close the
  reader. Using r.Close() is still recommended for most use cases.
• Previously, an unsafe conversion between []byte and string was used to
  avoid unnecessary allocations when decoding struct keys. The decoder now
  relies on a compiler optimization on string([]byte) map lookups to achieve
  this rather than using unsafe.

1.0.0

New release for those using tagged releases.