- check whether simple index fields actually correspond to methods or
  elements. if index field names are misspelled, for example, indexing
  ends up not adding records, and it's hard to figure out why.

- die'ing in an index collection code section seems to make the
  "undo_collection" fail -- figure out why. also, the die is silently
  ignored. is this the right thing?

- improve and flexibilify "escape" functionality. perhaps: there
  should be a short-cut element-level def tag that allows escape
  behavior to be specified. the tag content should e a two-argument
  list: first argument denoting escape and second argument denoting
  unescape method. arguments will be turned into code refs,
  somehow. defaults will remain XML_basic_escape and
  XML_basic_unescape, BUT default behavior will change. If escaping is
  specified in the Def, then escape=>1 becomes the default for set()
  and unescape=>1 becomes the default for get(). 

- file_permissions stuff should be part of the "location" arguments,
  not a stand-alone element. there should be a file_group argument, to
  go with file_permissions.

- DON'T LET RESERVED METHOD NAMES BE USED AS ELEMENT NAMES

- misspelled methods called on classes should throw more understandable errors:
  ex 'XML::Comma::Doc->retrrieve( ... )'

- document "distinct" iterator arg

- add the "test existence of element without creating" construction to
  the FAQ (same as the "how many instances" construction.

- check for multiply same-named stores (and indexes) in Defs

- read-only errors should give a tiny bit of context, so they're not
  so hard to track down.

- Index.pm line 286. A die'ing index hook indicates to the system that
  a Doc should not be indexed at all. The entirely reasonable action
  of index_remove()ing the Doc is therefore taken. But that breaks on
  any read_only Doc (as with rebuild()). What to do?

- an $index->pre_store_check_for_duplicates($doc) method. requires
  parsing sql_type and sql_index methods for UNIQUEosity, and figuring
  out how to check whether fields (or combinations thereof) are
  already present in the database. this method should be automatically
  called in the pre_store_hook for any index_on_store'ing store.

- quote() method to expose $dbh->quote() to DBH_User children

- replace references in the source to xymbollab.com/tools/comma with
  xml-comma.org

- document the "aggregate" stuff in Iterator/Index

- framework to handle translation between two related (but differently
  Def'ed) document collections

- sherrard found a bug in rebuild such that flags are set for all
  tables that hve a given index_name (not just for all tables of a
  certain doctype with a given index_name. It looks like the following
  three methods need to have more specific where clauses:
    sql_set_all_table_comma_flags_politely
    sql_get_all_tables_with_comma_flags_set
    sql_unset_all_table_comma_flags

- sherrard notes that we should be tricky enough to spool blobs
  somewhere for documents that haven't been stored yet, and then save
  the blobs on eventual store.

- retrieve does not do a permissions check on retrieve -- that's no
  good. retrieve should fail, rather than store failing. and probably
  store needs some better code to fail earlier and more gracefully
  than it does?

- does a <default>0</default> work as it should?

- make SQL syntax be an ISA thing rather than a crufty begin-block thing

- think about whether plural elements should be "order-able" by a def
  (the def should maintain plural element order based on a code block
  passed to sort, or something?)

- work on permissions for files (Sherrard says doc->erase() doesn't
  erase and doesn't warn if you don't have write permission to file --
  but that index entries are removed.)

- maybe Indexing::Iterator->_make_sort_spec shouldn't die just because
  it doesn't find a sort table

- figure out what to do about postgres 31-char name limit

- sql defaults for index fields should maybe be something other than NULL?

- rebuild() should read() rather than retrieve()?

- rebuild() needs to NOT erase docs at the end that were added during
  rebuild. wasn't this the origingal idea?

- should errors have an "object context" field, passing a reference to
  the object that is "most involved" in the error. Ie, for
  validate_content errors, the element that failed to validate?

- add error-handling and a better error message around the req_el
  fetch in the validate_hook that <required> installs. As it is, the
  error message is confusing if the <required> list gets out of sync
  with the declared elements.

- AbstractElement should define an is_empty() test that is overridden
  by Element and NestedElement. This will make required's
  validate_hook simpler and eliminate the bug wherein an nested
  element with no sub-elements is okay. 

- range macro needs argument checking

- macros: length (min=> and max=>) ;  email

- validate_content() should return the empty string

- auto-CDATA on element set?

- validate defs on load, to make use of 'required' and any other
available internal functions, and avoid special-purpose validation
code in Def.pm

- function arg-checking (like Class::NamedParms, or something)

- command line tools to examine/prune Doc locks

- throw an error if a process is trying to lock a file that it
  (itself) already has locked. (that's an easy "deadlock" to catch)

- re-add permissions checking to index add and delete

- add error-throwing to Index->rebuild() and Index->clean()

- make clean use the PID as its comma flag, and if that PID is no
  longer running, then clean anyway?

- make SQL stuff really robust -- for example, recover from multiple
  fields with the same name, needing to reconstruct a broken data
  table, etc.

- clean up the errors thrown by iterators a little more

- throw an error if you try to keep_open a new-then-store()ed Doc?

- add_element hook

- improve line-number reporting from parser

- re-write PurePerl parser to use index() rather than getc()