--- AnyEvent-DBI/Changes	2008/12/20 21:28:57	1.5
+++ AnyEvent-DBI/Changes	2020/02/01 21:43:28	1.20
@@ -1,5 +1,72 @@
 Revision history for AnyEvent::DBI
 
+TODO: why does make test pass even though Slave.pm has a version mismatch?
+3.04 Mon Apr 23 18:30:40 CEST 2018
+	- forgot to bump slave version (reported by Felix Ostmann).
+
+3.03 Thu Apr 19 06:24:14 CEST 2018
+	- the main process socket was not actually set to nonblocking mode,
+          causing lockups under load (reported by Felix Ostmann).
+
+3.02 Mon Sep  4 13:45:56 CEST 2017
+	- fixed module version mismatch (reported by Martin Barth).
+
+3.01 Sun Aug 27 11:53:26 CEST 2017
+	- add missing dependency to AnyEvent::Fork.
+
+3.0  Tue Aug 15 09:05:30 CEST 2017
+        - the exec_server constructor argument is now ignored, fork vs. exec
+          is now decided by availability of AnyEvent::Fork::Early or not.
+        - experimental support for fetching (but not setting) statement handle
+          attributes: ->stattr.
+        - use AnyEvent::Fork for process management.
+	- switch to CBOR::XS and Convert::Scalar for further speed and memory
+          improvements - a loop doing "select 0" is now almost twice as fast,
+          more impressive speedups are expected for larger results.
+          This also considerably simplified the code.
+	- t/02_sql_lite.t wrongly assumed sqlite doesn't implement instr(), but
+          nowadays, it does.
+        - added stability canary support.
+        - no longer load Scalar::Util, Fcntl, Socket and POSIX modules.
+        - move child code into it's own module, AnyEvent::DBI::Slave.
+        - no longer load DBI (except in slaves).
+
+2.3  Tue Apr  2 02:59:12 CEST 2013
+	- fix begin_work, commit and rollback, that were broken due
+          to a precedence problem (analyzed by Jerry Lundström).
+        - slightly improve tests.
+
+2.2  Thu May 17 04:13:27 CEST 2012
+	- timeouts didn't trigger (Adam Rosenstein).
+        - do not leave zombie processes (Adam Rosenstein).
+
+2.1  Sat Oct 30 22:22:05 CEST 2010
+	- no longer kill -9 the child, just kill -TERM and close the connection,
+          to avoid database corruption - this also works around a perl bug
+          (global destruction).
+	- convert to the faster AE API - major speedup.
+        - use common::sense.
+
+2.0  Mon Jun 29 10:18:58 CEST 2009
+	- INCOMPATIBLE CHANGE: callbacks now are passed the $dbh
+          as first argument, always.
+        - INCOMPATIBLE CHANGE: really pass rv as third argument to exec
+          callbacks, as documented but not implemented for 1.1.
+        - thanks to viacheslav.t for testing.
+	- major patch by Adam Rosenstein:
+        - add an on_connect callback (for reliable failure notification)
+        - add timeouts to connect and exec.
+        - reaps server processes if you undef your AE::DBI handle and keep
+          running.
+        - option to start the server via exec, otherwise try to emulate
+          close-on-exec behavior.
+        - add the begin_work/commit/rollback methods required to actually use
+          transactions.
+        - add accessors for database handle attibutes such as AutoCommit,
+          RaiseError, etc.
+        - add support for the func() method to interact directly with a driver.
+        - add more/real tests.
+
 1.1  Sat Dec 20 22:28:11 CET 2008
 	- INCOMPATIBLE CHANGE: pass result value in place of %extra
           to callbacks.