Fix string/bytes issues under python 3. 1) cgi/client.py: override cgi.FieldStorage's make_file so that file is always created in binary/byte mode. This means that json (and xml) are bytes not strings. 2) rest.py: try harder to find dicttoxml in roundup directory or on sys.path. This just worked under python 2 but python 3 only searches sys.path by default and does not search relative like python 2. 3) rest.py: replace headers.getheader call removed from python 3 with equivalent code. 4) rest.py: make value returned from dispatch into bytes not string. 5) test/caseinsensitivedict.py, test/test_CaseInsensitiveDict.py: get code from stackoverflow that implements a case insensitive key dict. So dict['foo'], dict['Foo'] are the same entry. Used for looking up headers in mocked http rewuset header array. 6) test/rest_common.py: rework tests for etags and rest to properly supply bytes to the called routines. Calls to s2b and b2s and use of BytesIO and overriding make_file in cgi.FieldStorage to try to make sure it works under python 3.

A number of tests uses the infrastructure of
	db_test_base.py

grep "from db_test_base" -l *.py
benchmark.py
session_common.py
test_anydbm.py
test_indexer.py
test_memorydb.py
test_mysql.py
test_postgresql.py
test_security.py
test_sqlite.py
test_userauditor.py

grep "import db_test_base" -l *.py
test_cgi.py
test_jinja2.py
test_mailgw.py
test_xmlrpc.py

grep "import memory\|from memory" -l *.py 
test_mailgw.py
test_memorydb.py


The remaining lines are an 2001 description from Richard,
which probably is outdated:

Structure of the tests:

   1   Test date classes
   1.1 Date
   1.2 Interval
   2   Set up schema
   3   Open with specific backend
   3.1 anydbm
   4   Create database base set (stati, priority, etc)
   5   Perform some actions
   6   Perform mail import
   6.1 text/plain
   6.2 multipart/mixed (with one text/plain)
   6.3 text/html
   6.4 multipart/alternative (with one text/plain)
   6.5 multipart/alternative (with no text/plain)