* Check all DB API calls
    Make sure a) we check return codes for for all calls into DB and
    that b) when there is an error we output it only when configured
    to be verbose and c) then return the proper DBSQL error indicating
    what went wrong at a higher level.

* Review all reused DB code
    Make sure that we're up to date with the latest in DB.  Things such
    as hashing, etc.

* Find printf/fprintf calls which print error messages
   The library should never print messages unless configured to be
   verbose.

* splint
   Clean up all items pointed out by splint.

* makedepend

* valgrind
   Fix all memory related bugs.

* review dbsql/dbsql.c
   General clean-up including changing calls to malloc, free, etc to
   __os_malloc, __os_free, etc.

* Review SQLite's latest lemon.c and lempar.c code

* Review code for trailing spaces, tab characters, lines > 79 col

* api.c/__os_sleep()
   This is now static in DB's code, so to hack the compile I commented
   out the use of __os_sleep() in api.c

* test/scr026
   This test has not been updated to check for DBSQL API calls rather
   than DB API calls.  Review and fix it.

* test/scr030
   Update this test to match the configure script.  Augment it so that
   as the configure script changes the script reports options not
   tested.

* Use LIST/QUEUE macros for linked lists, queues, etc.
   Find/replace all custom code implementing these simple data
   structures.

* Review memory allocation code in general

* A few scr??? < 050 examine source but do not include the TCL testing code

* Consider using MPOOL to manage some of DBSQL state across processes
   Function pools and other aspects of the DBSQL runtime need not be
   re-created in-memory across multiple processes.  When doing this
   revisit data structures using the LIST/QUEUE macros and shift them
   to the SH_ equivalents.

* Investigate the TCL Extension Architecture (http://www.tcl.tk/doc/tea/)