DB.lua

----------------------------------------------------------------------------------
--- Mudlet DB
----------------------------------------------------------------------------------


-- TODO will be already loaded in LuaGlobal
-----------------------------------------------------------------------------
-- General-purpose useful tools that were needed during development:
-----------------------------------------------------------------------------
if package.loaded["rex_pcre"] then
  rex = require "rex_pcre"
end

if not display then
  require "DebugTools"
end
if not table.contains then
  require "TableUtils"
end
if not string.trim then
  require "StringUtils"
end

-- TODO those functions are already definde elsewhere
function string.starts(String, Start)
  return string.sub(String, 1, string.len(Start)) == Start
end

function string.ends(String, End)
  return End == '' or string.sub(String, -string.len(End)) == End
end




-- TODO move to StringUtils?
-----------------------------------------------------------------------------
-- Some Date / Time parsing functions.
-----------------------------------------------------------------------------
datetime = {
  _directives = {
    ["%b"] = "(?P<abbrev_month_name>jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)",
    ["%B"] = "(?P<month_name>january|febuary|march|april|may|june|july|august|september|october|november|december)",
    ["%d"] = "(?P<day_of_month>\\d{2})",
    ["%H"] = "(?P<hour_24>\\d{2})",
    ["%I"] = "(?P<hour_12>\\d{2})",
    ["%m"] = "(?P<month>\\d{2})",
    ["%M"] = "(?P<minute>\\d{2})",
    ["%p"] = "(?P<ampm>am|pm)",
    ["%S"] = "(?P<second>\\d{2})",
    ["%y"] = "(?P<year_half>\\d{2})",
    ["%Y"] = "(?P<year_full>\\d{4})"
  },
  _pattern_cache = {},
  _month_names = {
    ["january"] = 1,
    ["febuary"] = 2,
    ["march"] = 3,
    ["april"] = 4,
    ["may"] = 5,
    ["june"] = 6,
    ["july"] = 7,
    ["august"] = 8,
    ["september"] = 9,
    ["october"] = 10,
    ["november"] = 11,
    ["december"] = 12
  },
  _abbrev_month_names = {
    ["jan"] = 1,
    ["feb"] = 2,
    ["mar"] = 3,
    ["apr"] = 4,
    ["may"] = 5,
    ["jun"] = 6,
    ["jul"] = 7,
    ["aug"] = 8,
    ["sep"] = 9,
    ["oct"] = 10,
    ["nov"] = 11,
    ["dec"] = 12
  }
}

-- the timestamp is stored in UTC time, so work out the difference in seconds
-- from local to UTC time. Credit: https://github.com/stevedonovan/Penlight/blob/master/lua/pl/Date.lua#L85
function datetime:calculate_UTCdiff(ts)
  local date, time = os.date, os.time
  local utc = date('!*t', ts)
  local lcl = date('*t', ts)
  lcl.isdst = false
  return os.difftime(time(lcl), time(utc))
end

-- NOT LUADOC
-- The rex.match function does not return named patterns even if you use named capture
-- groups, but the r:tfind does -- but this only operates on compiled patterns. So,
-- we are caching the conversion of 'simple format' date patterns into a regex, and
-- then compiling them.
function datetime:_get_pattern(format)
  if not datetime._pattern_cache[format] then
    local fmt = rex.gsub(format, "(%[A-Za-z])",
    function(m)
      return datetime._directives[m] or m
    end
    )

    datetime._pattern_cache[format] = rex.new(fmt, rex.flags().CASELESS)
  end

  return datetime._pattern_cache[format]
end



--- Parses the specified source string, according to the format if given, to return a representation of
--- the date/time. The default format if not specified is: "^%Y-%m-%d %H:%M:%S$" <br/><br/>
---
--- If as_epoch is provided and true, the return value will be a Unix epoch -- the number
--- of seconds since 1970. This is a useful format for exchanging date/times with other systems. If as_epoch
--- is false, then a Lua time table will be returned. Details of the time tables are provided
--- in the http://www.lua.org/pil/22.1.html. <br/><br/>
---
--- Supported Format Codes
---   </pre>
---   %b   Abbreviated Month Name
---   %B   Full Month Name
---   %d   Day of Month
---   %H   Hour (24-hour format)
---   %I   Hour (12-hour format, requires %p as well)
---   %p   AM or PM
---   %m   2-digit month (01-12)
---   %M   2-digit minutes (00-59)
---   %S   2-digit seconds (00-59)
---   %y   2-digit year (00-99), will automatically prepend 20 so 10 becomes 2010 and not 1910.
---   %Y   4-digit year.
---   </pre>
function datetime:parse(source, format, as_epoch)
  if not format then
    format = "^%Y-%m-%d %H:%M:%S$"
  end

  local fmt = datetime:_get_pattern(format)
  local m = { fmt:tfind(source) }

  if m and m[3] then
    m = m[3]
    dt = {}

    if m.year_half then
      dt.year = tonumber("20" .. m.year_half)
    elseif m.year_full then
      dt.year = tonumber(m.year_full)
    end

    if m.month then
      dt.month = tonumber(m.month)
    elseif m.month_name then
      dt.month = datetime._month_names[m.month_name:lower()]
    elseif m.abbrev_month_name then
      dt.month = datetime._abbrev_month_names[m.abbrev_month_name:lower()]
    end

    dt.day = m.day_of_month

    if m.hour_12 then
      assert(m.ampm, "You must use %p (AM|PM) with %I (12-hour time)")
      if m.ampm == "PM" then
        dt.hour = 12 + tonumber(m.hour_12)
      else
        dt.hour = tonumber(m.hour_12)
      end
    else
      dt.hour = tonumber(m.hour_24)
    end

    dt.min = tonumber(m.minute)
    dt.sec = tonumber(m.second)
    dt.isdst = false

    if as_epoch then
      return os.time(dt)
    else
      return dt
    end
  else
    return nil
  end
end



-----------------------------------------------------------------------------
-- The database wrapper library
-----------------------------------------------------------------------------
if package.loaded["luasql.sqlite3"] then
  luasql = require "luasql.sqlite3"
end

db = {}
db.__autocommit = {}
db.__schema = {}
db.__conn = {}

db.debug_sql = false



-- NOT LUADOC
-- Converts the type of a lua object to the equivalent type in SQL
function db:_sql_type(value)
  local t = type(value)

  if t == "number" then
    return "REAL"
  elseif t == "nil" then
    return "NULL"
  elseif t == "table" and value._timestamp ~= nil then
    return "INTEGER"
  else
    return "TEXT"
  end
end



-- NOT LUADOC
-- Converts a data value in Lua to its SQL equivalent; notably it will also escape single-quotes to
-- prevent inadvertant SQL injection.
-- called when generating the schema
function db:_sql_convert(value)
  local t = db:_sql_type(value)

  if value == nil then
    return "NULL"
  elseif t == "TEXT" and type(value) == "string" then
    return '"' .. value:gsub("'", "''") .. '"'
  elseif t == "NULL" then
    return "NULL"
  elseif t == "INTEGER" then
    -- With db.Timestamp's, a value of false should be interpreted as nil.
    if value._timestamp == false then
      return "NULL"
    end
    return tostring(value._timestamp)
  else
    return tostring(value)
  end
end



-- NOT LUADOC
-- Given a sheet name and the details of an index, this function will return a unique index name to
-- add to the database. The purpose of this is to create unique index names as indexes are tested
-- for existance on each call of db:create and not only on creation. That way new indexes can be
-- added after initial creation.
function db:_index_name(tbl_name, params)
  local t = type(params)

  if t == "string" then
    return "idx_" .. tbl_name .. "_c_" .. params
  elseif assert(t == "table", "Indexes must be either a string or a table.") then
    local parts = { "idx", tbl_name, "c" }
    for _, v in pairs(params) do
      parts[#parts + 1] = v
    end
    return table.concat(parts, "_")
  end
end



-- NOT LUADOC
-- This function returns true if all of the columns referenced in index_columns also exist within
-- the sheet_columns table array. The purpose of this is to raise an error if someone tries to index
-- a column which doesn't currently exist in the schema.
function db:_index_valid(sheet_columns, index_columns)
  if type(index_columns) == "string" then
    if sheet_columns[index_columns] ~= nil then
      return true
    else
      return false
    end
  else
    for _, v in ipairs(index_columns) do
      if sheet_columns[v] == nil then
        db:echo_sql("\n--> Bad index " .. v)
        return false
      end
    end
  end
  return true
end



-- NOT LUADOC
-- The column_spec is either a string or an indexed table. This function returns either "column" or
-- "column1", "column2" for use in the column specification of INSERT.
function db:_sql_columns(value)
  local colstr = ''
  local t = type(value)

  if t == "table" then
    col_chunks = {}
    for _, v in ipairs(value) do
      -- see https://www.sqlite.org/syntaxdiagrams.html#ordering-term
      if v:lower() == "desc" or v:lower() == "asc" then
        col_chunks[#col_chunks] = col_chunks[#col_chunks] .. " " .. v
      else
        col_chunks[#col_chunks + 1] = '"' .. v:lower() .. '"'
      end
    end

    colstr = table.concat(col_chunks, ',')
  elseif assert(t == "string",
  "Must specify either a table array or string for index, not " .. type(value)) then
    colstr = '"' .. value:lower() .. '"'
  end
  return colstr
end



-- NOT LUADOC
-- This serves as a very similar function to db:_sql_columns, quoting column names properly but for
-- uses outside of INSERTs.
function db:_sql_fields(values)
  local sql_fields = {}

  for k, v in pairs(values) do
    sql_fields[#sql_fields + 1] = '"' .. k .. '"'
  end

  return "(" .. table.concat(sql_fields, ",") .. ")"
end



-- NOT LUADOC
-- This quotes values to be passed into an INSERT or UPDATE operation in a SQL list. Meaning, it turns
-- {x="this", y="that", z=1} into ('this', 'that', 1).
-- It is intelligent with data-types; strings are automatically quoted (with internal single quotes
-- escaped), nil turned into NULL, timestamps converted to integers, and such.
function db:_sql_values(values)
  local sql_values = {}

  for k, v in pairs(values) do
    local t = type(v)
    local s = ""

    if t == "string" then
      s = "'" .. v:gsub("'", "''") .. "'"
    elseif t == "nil" then
      s = "NULL"
    elseif t == "table" and v._timestamp ~= nil then
      if not v._timestamp then
        return "NULL"
      else
        s = "datetime('" .. v._timestamp .. "', 'unixepoch')"
      end
    else
      s = tostring(v)
    end

    sql_values[#sql_values + 1] = s
  end

  return "(" .. table.concat(sql_values, ",") .. ")"
end



--- <b><u>TODO</u></b> db:safe_name(name)
--   On a filesystem level, names are restricted to being alphanumeric only. So, "my_database" becomes
--   "mydatabase", and "../../../../etc/passwd" becomes "etcpasswd". This prevents any possible
--   security issues with database names.
function db:safe_name(name)
  name = name:gsub("[^%ad]", "")
  name = name:lower()
  return name
end



--- Creates and/or modifies an existing database. This function is safe to define at a top-level of a Mudlet
--- script: in fact it is reccommended you run this function at a top-level without any kind of guards.
--- If the named database does not exist it will create it. If the database does exist then it will add
--- any columns or indexes which didn't exist before to that database. If the database already has all the
--- specified columns and indexes, it will do nothing. <br/><br/>
---
--- The database will be called Database_<sanitized database name>.db and will be stored in the
--- Mudlet configuration directory. <br/><br/>
---
--- Database 'tables' are called 'sheets' consistently throughout this documentation, to avoid confusion
--- with Lua tables. <br/><br/>
---
--- The schema table must be a Lua table array containing table dictionaries that define the structure and
--- layout of each sheet. <br/><br/>
---
--- For sheets with unique indexes, you may specify a _violations key to indicate how the db layer handle
--- cases where the unique index is violated. The options you may use are:
---   <pre>
---   FAIL - the default. A hard error is thrown, cancelling the script.
---   IGNORE - The command that would add a record that violates uniqueness just fails silently.
---   REPLACE - The old record which matched the unique index is dropped, and the new one is added to replace it.
---   </pre>
---
--- @usage Example bellow will create a database with two sheets; the first is kills and is used to track every successful kill,
---   with both where and when the kill happened. It has one index, a compound index tracking the combination of name and area.
---   The second sheet has two indexes, but one is unique: it isn't possible to add two items to the enemies sheet with the same name.
---   <pre>
---   local mydb = db:create("combat_log",
---     {
---       kills = {
---         name = "",
---         area = "",
---         killed = db:Timestamp("CURRENT_TIMESTAMP"),
---         _index = {{"name", "area"}}
---       },
---       enemies = {
---         name = "",
---         city = "",
---         reason = "",
---         enemied = db:Timestamp("CURRENT_TIMESTAMP"),
---         _index = { "city" },
---         _unique = { "name" },
---         _violations = "IGNORE"
---       }
---     }
---   )
---   </pre>
---   Note that you have to use double {{ }} if you have composite index/unique constrain.
function db:create(db_name, sheets)
  if not db.__env or (db.__env and db.__env == 'SQLite3 environment (closed)') then
    db.__env = luasql.sqlite3()
  end

  db_name = db:safe_name(db_name)

  if not db.__conn[db_name] or (db.__conn[db_name] and db.__conn[db_name] == 'SQLite3 connection (closed)') or (not io.exists(getMudletHomeDir() .. "/Database_" .. db_name .. ".db")) then
    db.__conn[db_name] = db.__env:connect(getMudletHomeDir() .. "/Database_" .. db_name .. ".db")
    db.__conn[db_name]:setautocommit(false)
    db.__autocommit[db_name] = true
  end

  db.__schema[db_name] = {}

  -- We need to separate the actual column configuration from the meta-configuration of the desired
  -- sheet. {sheet={"column"}} verses {sheet={"column"}, _index={"column"}}. In the former we are
  -- creating a database with a single field; in the latter we are also adding an index on that
  -- field. The db package reserves any key that begins with an underscore to be special and syntax
  -- for its own use.
  for s_name, sht in pairs(sheets) do
    options = {}

    if sht[1] ~= nil then
      -- in case the sheet was provided in the sheet = {"column1", "column2"} format:
      local t = {}               --   assume field types are text, and should default to ""
      for k, v in pairs(sht) do
        t[v] = ""
      end
      sht = t
    else -- sheet provided in the sheet = {"column1" = default} format
      local opts = {}
      for k, v in pairs(sht) do
        if string.starts(k, "_") then
          options[k] = v
          opts[#opts + 1] = k
        end
      end
      for _, v in ipairs(opts) do
        sht[v] = nil
      end
    end

    if not options._violations then
      options._violations = "FAIL"
    end

    db.__schema[db_name][s_name] = { columns = sht, options = options }
    db:_migrate(db_name, s_name)
  end
  return db:get_database(db_name)
end



-- NOT LUADOC
-- The migrate function is meant to upgrade an existing database live, to maintain a consistant
-- and correct set of sheets and fields, along with their indexes. It should be safe to run
-- at any time, and must not cause any data loss. It simply adds to what is there: in perticular
-- it is not capable of removing indexes, columns, or sheets after they have been defined.
function db:_migrate(db_name, s_name)
  local conn = db.__conn[db_name]
  local schema = db.__schema[db_name][s_name]

  local current_columns = {}

  -- The PRAGMA table_info command is a query which returns all of the columns currently
  -- defined in the specified table. The purpose of this section is to see if any new columns
  -- have been added.
  local cur = conn:execute("PRAGMA table_info('" .. s_name .. "')") -- currently broken - LuaSQL bug, needs to be upgraded for new sqlite API

  if type(cur) == "userdata" then
    local row = cur:fetch({}, "a")
    if row then
      while row do
        current_columns[row.name] = row.type
        row = cur:fetch({}, "a")
      end
    else
      ---------------  GETS ALL COLUMNS FROM SHEET IF IT EXISTS
      db:echo_sql("SELECT * FROM " .. s_name)
      local get_sheet_cur = conn:execute("SELECT * FROM " .. s_name)  -- select the sheet

      if get_sheet_cur and get_sheet_cur ~= 0 then
        local row = get_sheet_cur:fetch({}, "a") -- grab the first row, if any
        if not row then
          -- if no first row then
          local tried_cols, contains, found_something, col = {}, table.contains, false

          while not found_something do
            -- guarded by the error below from infinite looping
            col = false
            for k, v in pairs(schema.columns) do
              -- look through sheet schema to find the first column that is text
              if type(k) == "number" then
                if string.sub(v, 1, 1) ~= "_" and not contains(tried_cols, v) then
                  col = v break
                end
              else
                if string.sub(k, 1, 1) ~= "_" and type(v) == "string" and not contains(tried_cols, k) then
                  col = k break
                end
              end
            end

            if not col then
              error("db:_migrate: cannot find a suitable column for testing a new row with.")
            end

            -- add row with found column set as "test"
            db:add({ _db_name = db_name, _sht_name = s_name }, { [col] = "test" })

            db:echo_sql("SELECT * FROM " .. s_name)
            local get_row_cur = conn:execute("SELECT * FROM " .. s_name) -- select the sheet
            row = get_row_cur:fetch({}, "a") -- grab the newly created row
            get_row_cur:close()

            -- delete the newly created row. If we picked a row that doesn't exist yet and we're
            -- trying to add, the delete will fail - remember this, and try another row
            local worked, msg = pcall(db.delete, db, { _db_name = db_name, _sht_name = s_name }, db:eq({ database = db_name, sheet = s_name, name = col, type = "string" }, "test"))

            if not worked then
              tried_cols[#tried_cols + 1] = col
            else
              found_something = true
            end
          end
        end

        if row then
          -- add each column from row to current_columns table
          for k, v in pairs(row) do
            current_columns[k] = ""
          end
        end
        get_sheet_cur:close()
      end
    end
  end

  if type(cur) == "userdata" then
    cur:close()
  end

  -- The SQL definition of a column is:
  --    "column_name" column_type NULL
  -- The db module does not presently support columns that are required. Everything is optional,
  -- everything may be NULL / nil.
  -- If you specify a column's type, you also specify its default value.
  if table.is_empty(current_columns) then
    -- At this point, we know that the specified table does not exist in the database and so we
    -- should create it.

    -- Every sheet has an implicit _row_id column. It is not presently (and likely never will be)
    -- supported to define the primary key of any sheet.
    local sql = db:_build_create_table_sql(schema, s_name)
    db:echo_sql(sql)
    conn:execute(sql)

  else
    -- At this point we know that the sheet already exists, but we are concerned if the current
    -- definition includes columns which may be added.
    local missing = {}

    for k, v in pairs(schema.columns) do

      -- Here we test it a given column exists in the sheet already, and if not, we add that
      -- column.
      if not current_columns[k] then
        missing[#missing + 1] = { name = k, default = v }
      end
    end

    if #missing > 0 and
    table.size(current_columns) + #missing == table.size(schema.columns) + 1
    -- We have changes and when we did those changes, we have exactly
    -- the number of columns we need. The "+1" is for the _row_id
    -- which is not in the schema.
    then
      local sql_add = 'ALTER TABLE %s ADD COLUMN "%s" %s NULL DEFAULT %s'
      for _, v in ipairs(missing) do
        local t = db:_sql_type(v.default)
        local def = db:_sql_convert(v.default)
        local sql = sql_add:format(s_name, v.name, t, def)
        conn:execute(sql)
        db:echo_sql(sql)
      end
    elseif
    #missing + table.size(current_columns) > table.size(schema.columns) + 1
    -- if we add all missing columns and we have more columns than we want
    -- then there are currently some columns we don't want anymore.
    then
      local get_create = "SELECT sql FROM sqlite_master " ..
      "WHERE type = 'table' AND " ..
      "name = '" .. s_name .. "'"
      local ret_str
      cur, ret_str = conn:execute(get_create)
      assert(cur, ret_str)
      if type(cur) ~= "number" then
        local row = cur:fetch({}, "a");
        cur:close()
        local create_tmp = row.sql:gsub(s_name, s_name .. "_bak")
        local sql_chunks = {}
        local fields = { "_row_id" }
        local sql

        create_tmp = create_tmp:gsub("TABLE", "TEMPORARY TABLE")

        for k, _ in pairs(schema.columns) do
          fields[#fields + 1] = string.format('"%s"', k)
        end
        local fields_sql = table.concat(fields, ", ")

        sql_chunks[#sql_chunks + 1] = create_tmp .. ";"
        sql_chunks[#sql_chunks + 1] = "INSERT INTO " .. s_name .. "_bak " ..
        "SELECT * FROM " .. s_name .. ";"
        sql_chunks[#sql_chunks + 1] = "DROP TABLE " .. s_name .. ";"
        sql_chunks[#sql_chunks + 1] = db:_build_create_table_sql(schema,
        s_name) .. ";"
        sql_chunks[#sql_chunks + 1] = string.format(
        "INSERT INTO %s SELECT %s FROM %s_bak;", s_name, fields_sql,
        s_name)
        sql_chunks[#sql_chunks + 1] = "DROP TABLE " .. s_name .. "_bak;"

        for _, sql in ipairs(sql_chunks) do
          db:echo_sql(sql)
          local ret, str = conn:execute(sql)
          assert(ret, str)
        end
      end
    end
  end

  -- On every invocation of db:create we run the code that creates indexes, as that code will
  -- do nothing if the specific indexes already exist. This is enforced by the db:_index_name
  -- function creating a unique index.
  --
  -- Note that in no situation will an existing index be deleted.

  -- make up current_columns, as pragma_info currently does not populate it, due to luasql bug
  for key, value in pairs(schema.columns) do
    current_columns[key] = db:_sql_type(value)
  end

  db:_migrate_indexes(conn, s_name, schema, current_columns)
  db:echo_sql("COMMIT")
  conn:commit()
  conn:execute("VACUUM")
end

function db:_build_create_table_sql(schema, s_name)

  local sql_column = ', "%s" %s NULL'
  local sql_column_default = sql_column .. ' DEFAULT %s'


  local sql_chunks = { "CREATE TABLE ", s_name, '("_row_id" INTEGER PRIMARY KEY AUTOINCREMENT' }

  -- We iterate over every defined column, and add a line which creates it.
  for key, value in pairs(schema.columns) do
    local sql = ""
    if value == nil then
      sql = sql_column:format(key, db:_sql_type(value))
    else
      sql = sql_column_default:format(key, db:_sql_type(value), db:_sql_convert(value))
    end
    if (type(schema.options._unique) == "table" and table.contains(schema.options._unique, key))
    or (type(schema.options._unique) == "string" and schema.options._unique == key) then
      sql = sql .. " UNIQUE"
    end
    sql_chunks[#sql_chunks + 1] = sql
  end

  sql_chunks[#sql_chunks + 1] = ")"

  return table.concat(sql_chunks, "")
end


-- NOT LUADOC
-- Creates any indexes which do not yet exist in the given database.
function db:_migrate_indexes(conn, s_name, schema, current_columns)
  local sql_create_index = "CREATE %s IF NOT EXISTS %s ON %s (%s);"
  local opt = { _unique = "UNIQUE INDEX", _index = "INDEX" } -- , _check = "CHECK"}

  for option_type, options in pairs(schema.options) do
    if option_type == "_unique" or option_type == "_index" then
      for _, value in pairs(options) do

        -- If an index references a column which does not presently exist within the schema
        -- this will fail.

        if db:_index_valid(current_columns, value) then
          --assert(db:_index_valid(current_columns, value),
          --      "In sheet "..s_name.." an index field is specified that does not exist.")

          local sql = sql_create_index:format(
          opt[option_type], db:_index_name(s_name, value), s_name, db:_sql_columns(value)
          )
          db:echo_sql(sql)
          conn:execute(sql)
        end
      end
    end
  end
end



--- Adds one or more new rows to the specified sheet. If any of these rows would violate a UNIQUE index,
--- a lua error will be thrown and execution will cancel. As such it is advisable that if you use a UNIQUE
--- index, you test those values before you attempt to insert a new row. <br/><br/>
---
--- Each table is a series of key-value pairs to set the values of the sheet, but if any keys do not exist
--- then they will be set to nil or the default value. As you can see, all fields are optional.
---
--- @usage Adding one record.
---   <pre>
---   db:add(mydb.enemies, {name="Bob Smith", city="San Francisco"})
---   </pre>
--- @usage Adding multiple records.
---   <pre>
---   db:add(mydb.enemies,
---     {name="John Smith", city="San Francisco"},
---     {name="Jane Smith", city="San Francisco"},
---     {name="Richard Clark"}
---   )
---   </pre>
function db:add(sheet, ...)
  local db_name = sheet._db_name
  local s_name = sheet._sht_name
  assert(s_name, "First argument to db:add must be a proper Sheet object.")

  local conn = db.__conn[db_name]
  local sql_insert = "INSERT OR %s INTO %s %s VALUES %s"

  for _, t in ipairs({ ... }) do
    if t._row_id then
      -- You are not permitted to change a _row_id
      t._row_id = nil
    end

    local sql = sql_insert:format(db.__schema[db_name][s_name].options._violations, s_name, db:_sql_fields(t), db:_sql_values(t))
    db:echo_sql(sql)

    local result, msg = conn:execute(sql)
    if not result then
      return nil, msg
    end
  end
  if db.__autocommit[db_name] then
    conn:commit()
  end
  return true
end



--- Execute SQL select query against database. This only useful for some very specific cases. <br/>
--- Use db:fetch if possible instead - this function should not be normally used!
---
--- @release post Mudlet 1.1.1 (<b><u>TODO update before release</u></b>)
---
--- @usage Following will select all distinct area from my kills DB.
---   <pre>
---   db:fetch_sql(mydb.kills, "SELECT distinct area FROM kills")
---   </pre>
---
--- @see db:fetch
function db:fetch_sql(sheet, sql)
  local db_name = sheet._db_name
  local conn = db.__conn[db_name]

  db:echo_sql(sql)
  local cur = conn:execute(sql)

  -- if we had a syntax error in our SQL, cur will be nil
  if cur and cur ~= 0 then
    local results = {}
    local row = cur:fetch({}, "a")

    while row do
      results[#results + 1] = db:_coerce_sheet(sheet, row)
      row = cur:fetch({}, "a")
    end
    cur:close()
    return results
  else
    return nil
  end
end



--- Returns a table array containing a table for each matching row in the specified sheet. All arguments
--- but sheet are optional. If query is nil, the entire contents of the sheet will be returned. <br/><br/>
---
--- Query is a string which should be built by calling the various db: expression functions, such as db:eq,
--- db:AND, and such. You may pass a SQL WHERE clause here if you wish, but doing so is very dangerous.
--- If you don't know SQL well, its best to build the expression.<br/><br/>
---
--- Query may also be a table array of such expressions, if so they will be AND'd together implicitly.<br/><br/>
---
--- The results that are returned are not in any guaranteed order, though they are usually the same order
--- as the records were inserted. If you want to rely on the order in any way, you must pass a value to the
--- order_by field. This must be a table array listing the fields you want to sort by.
--- It can be { mydb.kills.area }, or { mydb.kills.area, mydb.kills.name } <br/><br/>
---
--- The results are returned in ascending (smallest to largest) order; to reverse this pass true into the final field.
---
--- @usage The first will fetch all of your enemies, sorted first by the city they reside in and then by their name.
---   <pre>
---   db:fetch(mydb.enemies, nil, {mydb.enemies.city, mydb.enemies.name})
---   </pre>
--- @usage The second will fetch only the enemies which are in San Francisco.
---   <pre>
---   db:fetch(mydb.enemies, db:eq(mydb.enemies.city, "San Francisco"))
---   </pre>
--- @usage The third will fetch all the things you've killed in Undervault which have Drow in their name.
---   <pre>
---   db:fetch(mydb.kills,
---      {
---         db:eq(mydb.kills.area, "Undervault"),
---         db:like(mydb.kills.name, "%Drow%")
---      }
---   )
---   </pre>
---
--- @see db:fetch_sql
function db:fetch(sheet, query, order_by, descending)
  local s_name = sheet._sht_name

  local sql = "SELECT * FROM " .. s_name

  if query then
    if type(query) == "table" then
      sql = sql .. " WHERE " .. db:AND(unpack(query))
    else
      sql = sql .. " WHERE " .. query
    end
  end

  if order_by then
    local o = {}
    for _, v in ipairs(order_by) do
      assert(v.name, "You must pass field instances (as obtained from yourdb.yoursheet.yourfield) to sort.")
      o[#o + 1] = v.name

      if descending then
        o[#o + 1] = "DESC"
      end
    end

    sql = sql .. " ORDER BY " .. db:_sql_columns(o)
  end

  return db:fetch_sql(sheet, sql)
end



--- Returns the result of calling the specified aggregate function on the field and its sheet. <br/><br/>
---
--- The supported aggregate functions are:
---   <pre>
---   COUNT - Returns the total number of records that are in the sheet or match the query.
---   AVG   - Returns the average of all the numbers in the specified field.
---   MAX   - Returns the highest number in the specified field.
---   MIN   - Returns the lowest number in the specified field.
---   TOTAL - Returns the value of adding all the contents of the specified field.
---   </pre>
---
--- @param query optional
---
--- @usage Example:
---   <pre>
---   local mydb = db:get_database("my database")
---   echo(db:aggregate(mydb.enemies.name, "count"))
---   </pre>
function db:aggregate(field, fn, query, distinct)
  local db_name = field.database
  local s_name = field.sheet
  local conn = db.__conn[db_name]

  assert(type(field) == "table", "Field must be a field reference.")
  assert(field.name, "Field must be a real field reference.")

  local sql_chunks = { "SELECT", fn, "(", distinct and "DISTINCT" or "", field.name, ")", "AS", fn, "FROM", s_name }

  if query then
    sql_chunks[#sql_chunks + 1] = "WHERE"
    if type(query) == "table" then
      sql_chunks[#sql_chunks + 1] = db:AND(unpack(query))
    else
      sql_chunks[#sql_chunks + 1] = query
    end
  end

  local sql = table.concat(sql_chunks, " ")

  db:echo_sql(sql)
  local cur = conn:execute(sql)

  if cur ~= 0 then
    local row = cur:fetch({}, "a")
    local count = row[fn]
    cur:close()

    -- give back the correct data type. see http://www.sqlite.org/lang_aggfunc.html
    if (fn:upper() ~= "MIN" and fn:upper() ~= "MAX") or field.type == "number" then
      return tonumber(count)
    end
    if field.type == "string" then
      return count
    end
    -- Only datetime left
    -- the value, count, is currently in a UTC timestamp
    local localtime = datetime:parse(count, nil, true)
    -- convert it into a UTC timestamp as datetime:parse parses it in the local time context
    count = db:Timestamp(localtime + datetime:calculate_UTCdiff(localtime))
    return count
  else
    return 0
  end
end



--- Deletes rows from the specified sheet. The argument for query tries to be intelligent: <br/>
--- * if it is a simple number, it deletes a specific row by _row_id <br/>
--- * if it is a table that contains a _row_id (e.g., a table returned by db:get) it deletes just that record. <br/>
--- * Otherwise, it deletes every record which matches the query pattern which is specified as with db:get. <br/>
--- * If the query is simply true, then it will truncate the entire contents of the sheet. <br/>
---
--- @usage When passed an actual result table that was obtained from db:fetch, it will delete the record for that table.
---   <pre>
---   enemies = db:fetch(mydb.enemies)
---   db:delete(mydb.enemies, enemies[1])
---   </pre>
--- @usage When passed a number, will delete the record for that _row_id. This example shows getting the row id from a table.
---   <pre>
---   enemies = db:fetch(mydb.enemies)
---   db:delete(mydb.enemies, enemies[1]._row_id)
---   </pre>
--- @usage As above, but this example just passes in the row id directly.
---   <pre>
---   db:delete(mydb.enemies, 5)
---   </pre>
--- @usage Here, we will delete anything which matches the same kind of query as db:fetch uses - namely,
---   anyone who is in the city of San Francisco.
---   <pre>
---   db:delete(mydb.enemies, db:eq(mydb.enemies.city, "San Francisco"))
---   </pre>
--- @usage And finally, we will delete the entire contents of the enemies table.
---   <pre>
---   db:delete(mydb.enemies, true)
---   </pre>
function db:delete(sheet, query)
  local db_name = sheet._db_name
  local s_name = sheet._sht_name

  local conn = db.__conn[db_name]

  assert(query, "must pass a query argument to db:delete()")
  if type(query) == "number" then
    query = "_row_id = " .. tostring(query)
  elseif type(query) == "table" then
    assert(query._row_id, "Passed a non-result table to db:delete, need a _row_id field to continue.")
    query = "_row_id = " .. tostring(query._row_id)
  end

  local sql = "DELETE FROM " .. s_name

  if query ~= true then
    sql = sql .. " WHERE " .. query
  end

  db:echo_sql(sql)
  assert(conn:execute(sql))
  if db.__autocommit[db_name] then
    conn:commit()
  end
end



--- Merges the specified table array into the sheet, modifying any existing rows and adding any that don't exist.
---
--- This function is a convenience utility that allows you to quickly modify a sheet, changing
--- existing rows and add new ones as appropriate. It ONLY works on sheets which have a unique
--- index, and only when that unique index is only on a single field. For more complex situations
--- you'll have to do the logic yourself.
---
--- The table array may contain tables that were either returned previously by db:fetch, or new tables
--- that you've constructed with the correct fields, or any mix of both. Each table must have a value
--- for the unique key that has been set on this sheet.
---
--- @usage For example, consider this database:
---   <pre>
---   local mydb = db:create("peopledb",
---     {
---       friends = {
---         name = "",
---         race = "",
---         level = 0,
---         city = "",
---         _index = { "city" },
---         _unique = { "name" }
---       }
---     }
---   )
---   </pre>
---
---   Here you have a database with one sheet, which contains your friends, their race, level,
---   and what city they live in. Let's say you want to fetch everyone who lives in San Francisco, you could do:
---   <pre>
---   local results = db:fetch(mydb.friends, db:eq(mydb.friends.city, "San Francisco"))
---   </pre>
---
---   The tables in results are static, any changes to them are not saved back to the database.
---   But after a major radioactive cataclysm rendered everyone in San Francisco a mutant,
---   you could make changes to the tables as so:
---   <pre>
---   for _, friend in ipairs(results) do
---     friend.race = "Mutant"
---   end
---   </pre>
---
---   If you are also now aware of a new arrival in San Francisco, you could add them to that existing table array:
---   <pre>
---   results[#results+1] = {name="Bobette", race="Mutant", city="San Francisco"}
---   </pre>
---
---   And commit all of these changes back to the database at once with:
---   <pre>
---   db:merge_unique(mydb.friends, results)
---   </pre>
---
---   The db:merge_unique function will change the 'city' values for all the people who we previously fetched, but then add a new record as well.
function db:merge_unique(sheet, tables)
  assert(type(tables) == "table", "db:merge_unique: missing the required table of data to merge")

  local db_name = sheet._db_name
  local s_name = sheet._sht_name

  local unique_options = db.__schema[db_name][s_name].options._unique
  assert(unique_options, "db:merge_unique only works on a sheet with a unique index.")
  assert(#unique_options == 1, "db:merge_unique only works on a sheet with a single unique index.")

  local unique_index = unique_options[1]
  local unique_key = ""
  if type(unique_index) == "table" then
    assert(#unique_index == 1, "db:merge_unique currently only supports sheets with a single unique index with a single column.")
    unique_key = unique_index[1]
  else
    unique_key = unique_index
  end

  db:echo_sql(":: Unique index = " .. unique_key)

  local conn = db.__conn[db_name]
  local mydb = db:get_database(db_name)
  mydb:_begin()

  for _, tbl in ipairs(tables) do
    assert(tbl[unique_key], "attempting to db:merge_unique with a table that does not have the unique key.")

    local results = db:fetch(sheet, db:eq(sheet[unique_key], tbl[unique_key]))
    if results and results[1] then
      local t = results[1]
      for k, v in pairs(tbl) do
        t[k] = v
      end

      db:update(sheet, t)
    else
      db:add(sheet, tbl)
    end
  end

  mydb:_commit()
  mydb:_end()
end



--- This function updates a row in the specified sheet, but only accepts a row which has been previously
--- obtained by db:fetch. Its primary purpose is that if you do a db:fetch, then change the value of a field
--- or tow, you can save back that table.
---
--- @usage This obtains a database reference, and queries the friends sheet for someone named Bob. As this
---   returns a table array containing only one item, it assigns that one item to the local variable named bob.
---   We then change the notes on Bob, and pass it into db:update() to save the changes back.
---   <pre>
---   local mydb = db:get_database("my database")
---   local bob = db:fetch(mydb.friends, db:eq(mydb.friends.name, "Bob"))[1]
---   bob.notes = "He's a really awesome guy."
---   db:update(mydb.friends, bob)
---   </pre>
function db:update(sheet, tbl)
  assert(tbl._row_id, "Can only update a table with a _row_id")
  assert(not table.is_empty(tbl), "An empty table was passed to db:update")

  local db_name = sheet._db_name
  local s_name = sheet._sht_name

  local conn = db.__conn[db_name]

  local sql_chunks = { "UPDATE OR", db.__schema[db_name][s_name].options._violations, s_name, "SET" }

  local set_chunks = {}
  local set_block = [["%s" = %s]]

  for k, v in pairs(db.__schema[db_name][s_name]['columns']) do
    if tbl[k] then
      local field = sheet[k]
      set_chunks[#set_chunks + 1] = set_block:format(k, db:_coerce(field, tbl[k]))
    end
  end

  sql_chunks[#sql_chunks + 1] = table.concat(set_chunks, ",")
  sql_chunks[#sql_chunks + 1] = "WHERE _row_id = " .. tbl._row_id

  local sql = table.concat(sql_chunks, " ")
  db:echo_sql(sql)
  assert(conn:execute(sql))
  if db.__autocommit[db_name] then
    conn:commit()
  end
end



--- The db:set function allows you to set a certain field to a certain value across an entire sheet.
--- Meaning, you can change all of the last_read fields in the sheet to a certain value, or possibly only
--- the last_read fields which are in a certain city. The query argument can be any value which is appropriate
--- for db:fetch, even nil which will change the value for the specified column for EVERY row in the sheet.
---
--- For example, consider a situation in which you are tracking how many times you find a certain
--- type of egg during Easter. You start by setting up your database and adding an Eggs sheet, and
--- then adding a record for each type of egg.
---   <pre>
---   local mydb = db:create("egg database", {eggs = {color = "", last_found = db.Timestamp(false), found = 0}})
---   db:add(mydb.eggs,
---     {color = "Red"},
---     {color = "Blue"},
---     {color = "Green"},
---     {color = "Yellow"},
---     {color = "Black"}
---   )
---   </pre>
---
--- Now, you have three columns. One is a string, one a timestamp (that ends up as nil in the database),
--- and one is a number. <br/><br/>
---
--- You can then set up a trigger to capture from the mud the string, "You pick up a (.*) egg!", and you
--- end up arranging to store the value of that expression in a variable called "myegg". <br/><br/>
---
--- To increment how many we found, we will do this:
---   <pre>
---   myegg = "Red" -- We will pretend a trigger set this.
---   db:set(mydb.eggs.found, db:exp("found + 1"), db:eq(mydb.eggs.color, myegg))
---   db:set(mydb.eggs.last_found, db.Timestamp("CURRENT_TIMESTAMP"), db:eq(mydb.eggs.color, myegg))
---   </pre>
---
--- This will go out and set two fields in the Red egg sheet; the first is the found field, which will
--- increment the value of that field (using the special db:exp function). The second will update the
--- last_found field with the current time. <br/><br/>
---
--- Once this contest is over, you may wish to reset this data but keep the database around.
--- To do that, you may use a more broad use of db:set as such:
---   <pre>
---   db:set(mydb.eggs.found, 0)
---   db:set(mydb.eggs.last_found, nil)
---   </pre>
function db:set(field, value, query)
  local db_name = field.database
  local s_name = field.sheet

  local conn = db.__conn[db_name]

  local sql_update = [[UPDATE OR %s %s SET "%s" = %s]]
  if query then
    sql_update = sql_update .. [[ WHERE %s]]
  end

  local sql = sql_update:format(db.__schema[db_name][s_name].options._violations, s_name, field.name, db:_coerce(field, value), query)

  db:echo_sql(sql)
  assert(conn:execute(sql))
  if db.__autocommit[db_name] then
    conn:commit()
  end
end



--- This is a debugging function, which echos any SQL commands if db.debug_sql is true.
--- You should not call this function directly from Mudlet.
---
--- @usage Set following lua variable to enable SQL echos.
---   <pre>
---   db.debug_sql=true
---   </pre>
function db:echo_sql(sql)
  if db.debug_sql then
    print(sql)
  end
end



-- NOT LUADOC
-- After a table so retrieved from the database, this function coerces values to
-- their proper types. Specifically, numbers and datetimes become the proper
-- types.
function db:_coerce_sheet(sheet, tbl)
  if tbl then
    tbl._row_id = tonumber(tbl._row_id)

    for k, v in pairs(tbl) do
      if k ~= "_row_id" then
        local field = sheet[k]
        if field.type == "number" then
          tbl[k] = tonumber(tbl[k]) or tbl[k]
        elseif field.type == "datetime" then
          -- the value, tbl[k], is currently in a UTC timestamp
          local localtime = datetime:parse(tbl[k], nil, true)
          -- convert it into a UTC timestamp as datetime:parse parses it in the local time context
          tbl[k] = db:Timestamp(localtime + datetime:calculate_UTCdiff(localtime))
        end
      end
    end
    return tbl
  end
end



-- NOT LUADOC
-- The function converts a Lua value into its SQL representation, depending on the
-- type of the specified field. Strings will be single-quoted (and single-quotes
-- within will be properly escaped), numbers will be rendered properly, and such.
function db:_coerce(field, value)
  if field.type == "number" then
    return tonumber(value) or "'" .. value .. "'"
  elseif field.type == "datetime" then
    if value._timestamp == false then
      return "NULL"
    else
      return "datetime('" .. value._timestamp .. "', 'unixepoch')" or "'" .. value .. "'"
    end
  else
    return "'" .. tostring(value):gsub("'", "''") .. "'"
  end
end



--- Returns a database expression to test if the field in the sheet is equal to the value.
---
--- @see db:fetch
function db:eq(field, value, case_insensitive)
  local fieldname = field.name
  -- escape column names as per https://www.sqlite.org/lang_expr.html
  fieldname = '"' .. fieldname:gsub("'", "''") .. '"'
  if case_insensitive then
    local v = db:_coerce(field, value):lower()
    return "lower(" .. fieldname .. ") == " .. v
  else
    local v = db:_coerce(field, value)
    return fieldname .. " == " .. v
  end
end



--- Returns a database expression to test if the field in the sheet is NOT equal to the value.
---
--- @see db:fetch
function db:not_eq(field, value, case_insensitive)
  if case_insensitive then
    local v = db:_coerce(field, value):lower()
    return "lower(" .. field.name .. ") != " .. v
  else
    local v = db:_coerce(field, value)
    return field.name .. " != " .. v
  end
end



--- Returns a database expression to test if the field in the sheet is less than the value.
---
--- @see db:fetch
function db:lt(field, value)
  local v = db:_coerce(field, value)
  return field.name .. " < " .. v
end



--- Returns a database expression to test if the field in the sheet is less than or equal to the value.
---
--- @see db:fetch
function db:lte(field, value)
  local v = db:_coerce(field, value)
  return field.name .. " <= " .. v
end



--- Returns a database expression to test if the field in the sheet is greater than to the value.
---
--- @see db:fetch
function db:gt(field, value)
  local v = db:_coerce(field, value)
  return field.name .. " > " .. v
end



--- Returns a database expression to test if the field in the sheet is greater than or equal to the value.
---
--- @see db:fetch
function db:gte(field, value)
  local v = db:_coerce(field, value)
  return field.name .. " >= " .. v
end



--- Returns a database expression to test if the field in the sheet is nil.
---
--- @see db:fetch
function db:is_nil(field)
  return field.name .. " IS NULL"
end



--- Returns a database expression to test if the field in the sheet is not nil.
---
--- @see db:fetch
function db:is_not_nil(field)
  return field.name .. " IS NOT NULL"
end



--- Returns a database expression to test if the field in the sheet matches the specified pattern. <br/><br/>
---
--- LIKE patterns are not case-sensitive, and allow two wild cards. The first is an underscore which matches
--- any single one character. The second is a percent symbol which matches zero or more of any character.
---   <pre>
---   LIKE with "_" is therefore the same as the "." regular expression.
---   LIKE with "%" is therefore the same as ".*" regular expression.
---   </pre>
---
--- @see db:not_like
--- @see db:fetch
function db:like(field, value)
  local v = db:_coerce(field, value)
  return field.name .. " LIKE " .. v
end



--- Returns a database expression to test if the field in the sheet does not match the specified pattern.
---
--- LIKE patterns are not case-sensitive, and allow two wild cards. The first is an underscore which matches
--- any single one character. The second is a percent symbol which matches zero or more of any character.
---   <pre>
---   LIKE with "_" is therefore the same as the "." regular expression.
---   LIKE with "%" is therefore the same as ".*" regular expression.
---   </pre>
---
--- @see db:like
--- @see db:fetch
function db:not_like(field, value)
  local v = db:_coerce(field, value)
  return field.name .. " NOT LIKE " .. v
end



--- Returns a database expression to test if the field in the sheet is a value between lower_bound and upper_bound.
--- This only really makes sense for numbers and Timestamps.
---
--- @see db:not_between
--- @see db:fetch
function db:between(field, left_bound, right_bound)
  local x = db:_coerce(field, left_bound)
  local y = db:_coerce(field, right_bound)
  return field.name .. " BETWEEN " .. x .. " AND " .. y
end



--- Returns a database expression to test if the field in the sheet is NOT a value between lower_bound and upper_bound.
--- This only really makes sense for numbers and Timestamps.
---
--- @see db:between
--- @see db:fetch
function db:not_between(field, left_bound, right_bound)
  local x = db:_coerce(field, left_bound)
  local y = db:_coerce(field, right_bound)
  return field.name .. " NOT BETWEEN " .. x .. " AND " .. y
end



--- Returns a database expression to test if the field in the sheet is one of the values in the table array. <br/><br/>
---
--- First, note the trailing underscore carefully! It is required.
---
--- @usage The following example illustrates the use of <b>in_</b>:
---   This will obtain all of your kills which happened in the Undervault, Hell or Purgatory. Every db:in_ expression
---   can be written as a db:OR, but that quite often gets very complex.
---   <pre>
---   local mydb = db:get_database("my database")
---   local areas = {"Undervault", "Hell", "Purgatory"}
---   db:fetch(mydb.kills, db:in_(mydb.kills.area, areas))
---   </pre>
---
--- @see db:fetch
function db:in_(field, tbl)
  local parts = {}
  for _, v in ipairs(tbl) do
    parts[#parts + 1] = db:_coerce(field, v)
  end

  return field.name .. " IN (" .. table.concat(parts, ",") .. ")"
end



--- Returns a database expression to test if the field in the sheet is not one of the values in the table array.
---
--- @see db:in_
--- @see db:fetch
function db:not_in(field, tbl)
  local parts = {}
  for _, v in ipairs(tbl) do
    parts[#parts + 1] = db:_coerce(field, v)
  end

  return field.name .. " NOT IN (" .. table.concat(parts, ",") .. ")"
end



--- Returns the string as-is to the database. <br/><br/>
---
--- Use this function with caution, but it is very useful in some circumstances. One of the most
--- common of such is incrementing an existing field in a db:set() operation, as so:
---   <pre>
---   db:set(mydb.enemies, db:exp("kills + 1"), db:eq(mydb.enemies.name, "Ixokai"))
---   </pre>
---
--- This will increment the value of the kills field for the row identified by the name Ixokai. <br/><br/>
---
--- But there are other uses, as the underlining database layer provides many functions you can call
--- to do certain things. If you want to get a list of all your enemies who have a name longer then
--- 10 characters, you may do:
---   <pre>
---   db:fetch(mydb.enemies, db:exp("length(name) > 10"))
---   </pre>
---
--- Again, take special care with this, as you are doing SQL syntax directly and the library can't
--- help you get things right.
---
--- @see db:fetch
function db:exp(text)
  return text
end



--- Returns a compound database expression that combines all of the simple expressions passed into it.
--- These expressions should be generated with other db: functions such as db:eq, db:like, db:lt and the like. <br/><br/>
---
--- This compound expression will only find items in the sheet if all sub-expressions match.
---
--- @see db:fetch
function db:AND(...)
  local parts = {}

  for _, expression in ipairs({ ... }) do
    parts[#parts + 1] = "(" .. expression .. ")"
  end

  return "(" .. table.concat(parts, " AND ") .. ")"
end



--- Returns a compound database expression that combines both of the simple expressions passed into it.
--- These expressions should be generated with other db: functions such as db:eq, db:like, db:lt and the like. <br/><br/>
---
--- This compound expression will find any item that matches either the first or the second sub-expression.
---
--- @see db:fetch
function db:OR(left, right)
  if not string.starts(left, "(") then
    left = "(" .. left .. ")"
  end

  if not string.starts(right, "(") then
    right = "(" .. right .. ")"
  end

  return left .. " OR " .. right
end



--- <b><u>TODO</u></b>
function db:close()
  for _, c in pairs(db.__conn) do
    c:close()
  end
  db.__conn = {}
  db.__env:close()
  db.__env = nil
end



-- Timestamp support
db.__Timestamp = {}



db.__TimestampMT = {
  __index = db.__Timestamp
}


function db.__Timestamp:as_string(format)
  if not format then
    format = "%m-%d-%Y %H:%M:%S"
  end

  -- given how we have an as_string function, having to wrap it in tostring() is a bit silly. So in this case, return nil as a string
  if type(self._timestamp) ~= "number" then
    return "nil", "db.Timestamp:as_string: timestamp seems to be invalid and isn't a number"
  end

  return os.date(format, self._timestamp)
end

function db.__Timestamp:as_table()
  if type(self._timestamp) ~= "number" then
    return nil, "db.Timestamp:as_table: timestamp seems to be invalid and isn't a number"
  end

  return os.date("*t", self._timestamp)
end

function db.__Timestamp:as_number()
  if type(self._timestamp) ~= "number" then
    return nil, "db.Timestamp:as_number: timestamp seems to be invalid and isn't a number"
  end

  return self._timestamp
end

function db.__Timestamp:set(timestamp)
  assert(type(timestamp) == "number", "db.Timestamp:set: timestamp needs to be a number")

  self._timestamp = timestamp
end


--- <b><u>TODO</u></b>
function db:Timestamp(ts, fmt)
  local dt = {}
  if type(ts) == "table" then
    dt._timestamp = os.time(ts)
  elseif type(ts) == "number" then
    dt._timestamp = ts
  elseif type(ts) == "string" and
  assert(ts == "CURRENT_TIMESTAMP", "The only strings supported by db.DateTime:new is CURRENT_TIMESTAMP") then
    dt._timestamp = "CURRENT_TIMESTAMP"
  elseif ts == nil then
    dt._timestamp = false
  else
    assert(nil, "Invalid value passed to db.Timestamp()")
  end
  return setmetatable(dt, db.__TimestampMT)
end



db.Field = {}
db.__FieldMT = {
  __index = db.Field
}



db.Sheet = {}
db.__SheetMT = {
  __index = function(t, k)
    local v = rawget(db.Sheet, k)
    if v then
      return v
    end

    local db_name = rawget(t, "_db_name")
    local sht_name = rawget(t, "_sht_name")
    local f_name = k

    local errormsg = "Attempt to access field %s in sheet %s in database %s that does not exist."

    local field = db.__schema[db_name][sht_name]['columns'][f_name]
    if assert(field, errormsg:format(k, sht_name, db_name)) then
      type_ = type(field)
      if type_ == "table" and field._timestamp then
        type_ = "datetime"
      end

      rt = setmetatable({ database = db_name, sheet = sht_name, type = type_, name = f_name }, db.__FieldMT)
      rawset(t, k, rt)
      return rt
    end

  end
}



db.Database = {}
db.__DatabaseMT = {
  __index = function(t, k)
    local v = rawget(t, k)
    if v then
      return v
    end

    local v = rawget(db.Database, k)
    if v then
      return v
    end

    local db_name = rawget(t, "_db_name")
    if assert(db.__schema[db_name][k], "Attempt to access sheet '" .. k .. "'in db '" .. db_name .. "' that does not exist.") then
      rt = setmetatable({ _db_name = db_name, _sht_name = k }, db.__SheetMT)
      rawset(t, k, rt)
      return rt
    end
  end
}



function db.Database:_begin()
  db.__autocommit[self._db_name] = false
end



function db.Database:_commit()
  local conn = db.__conn[self._db_name]
  conn:commit()
end



function db.Database:_rollback()
  local conn = db.__conn[self._db_name]
  conn:rollback()
end



function db.Database:_end()
  db.__autocommit[self._db_name] = true
end



function db.Database:_drop(s_name)
  local conn = db.__conn[self._db_name]
  local schema = db.__schema[self._db_name]

  if schema.options._index then
    for _, value in schema.options._index do
      conn:execute("DROP INDEX IF EXISTS " .. db:_index_name(s_name, value))
    end
  end

  if schema.options._unique then
    for _, value in schema.options._unique do
      conn:execute("DROP INDEX IF EXISTS " .. db:_index_name(s_name, value))
    end
  end

  conn:execute("DROP TABLE IF EXISTS " .. s_name)
  conn:commit()
end



--- Returns a reference of an already existing database. This instance can be used to get references
--- to the sheets (and from there, fields) that are defined within the database. You use these
--- references to construct queries. <br/><br/>
---
--- These references do not contain any actual data, they only point to parts of the database structure.
---
--- @usage If a database has a sheet named enemies, you can obtain a reference to that sheet by simply doing:
---   <pre>
---   local mydb = db:get_database("my database")
---   local enemies_ref = mydb.enemies
---   local name_ref = mydb.enemies.name
---   </pre>
function db:get_database(db_name)
  db_name = db:safe_name(db_name)
  assert(db.__schema[db_name], "Attempt to access database that does not exist.")

  db_inst = { _db_name = db_name }
  return setmetatable(db_inst, db.__DatabaseMT)
end

--- Queries for database content matching the given example. Different fields of
--- the example are AND connected.
--- </br>
--- Field values should be strings and can contain the following values:
--- <ul>
---   <li>literal strings to search for
---   <li>comparison terms prepended with &lt;, &gt;, &gt;=, &lt;=, !=, &lt;&gt;
---       for number and date comparisons
---   <li>ranges with :: between lower and upper bound
---   <li>different single values combined by || as OR
---   <li>strings containing % for a single and _ for multiple wildcard
---       characters
--- </ul>
--- <br/>
--- @param database Reference to the database that should be queried.
--- @param example  Query prototype that should be searched for.
--- @usage This example shows, how to use this function:
---   <pre>
---      mydb = db:create("mydb",
---        {
---          sheet = {
---            name = "", id = 0, city = "",
---            _index = { "name" },
---            _unique = { "id" },
---            _violations = "FAIL"
---          }
---        })
---      test_data = {
---        {name="Ixokai", city="Magnagora", id=1},
---        {name="Vadi", city="New Celest", id=2},
---        {name="Heiko", city="Hallifax", id=3},
---        {name="Keneanung", city="Hashan", id=4},
---        {name="Carmain", city="Mhaldor", id=5},
---        {name="Ixokai", city="Hallifax", id=6},
---      }
---      db:add(mydb.sheet, unpack(test_data))
---      res = db:query_by_example(mydb.sheet, { name = "Ixokai"})
---      display(res)
---      --[[
---      Prints
---      {
---        {
---          id = 1,
---          name = "Ixokai",
---          city = "Magnagora"
---        },
---        {
---          id = 6,
---          name = "Ixokai",
---          city = "Hallifax"
---        }
---      }
---      --]]
---   </pre>
---
--- @usage This example shows, how to combine two fields:
---   <pre>
---      mydb = db:create("mydb",
---        {
---          sheet = {
---            name = "", id = 0, city = "",
---            _index = { "name" },
---            _unique = { "id" },
---            _violations = "FAIL"
---          }
---        })
---      test_data = {
---        {name="Ixokai", city="Magnagora", id=1},
---        {name="Vadi", city="New Celest", id=2},
---        {name="Heiko", city="Hallifax", id=3},
---        {name="Keneanung", city="Hashan", id=4},
---        {name="Carmain", city="Mhaldor", id=5},
---      }
---      db:add(mydb.sheet, unpack(test_data))
---      res = db:query_by_example(mydb.sheet, { name = "Ixokai", id = "1"})
---      display(res)
---      --[[
---      Prints
---      {
---        id = 1,
---        name = "Ixokai",
---        city = "Magnagora"
---      }
---      --]]
---   </pre>
function db:query_by_example(database, example)

  if table.is_empty(example) then
    return nil
  end

  local topLevel = {}
  local find = string.find
  local match = string.match

  for key, value in pairs(example) do

    value = string.trim(value)

    local op, exp = match(value, "^%s*([<>=!]*)%s*(.*)$")

    if op == "<" then
      topLevel[#topLevel + 1] = db:lt(database[key], exp)
    elseif op == ">" then
      topLevel[#topLevel + 1] = db:gt(database[key], exp)
    elseif op == ">=" then
      topLevel[#topLevel + 1] = db:gte(database[key], exp)
    elseif op == "<=" then
      topLevel[#topLevel + 1] = db:lte(database[key], exp)
    elseif op == "!=" or op == "<>" then
      if find(exp, "__NULL__", 1, true) then
        topLevel[#topLevel + 1] = db:is_not_nil(database[key])
      else
        topLevel[#topLevel + 1] = db:not_eq(database[key], exp)
      end
    else
      if find(value, "%s*||%s*") then
        topLevel[#topLevel + 1] = db:in_(database[key], string.split(value,
        "%s*||%s*"))
      elseif find(value, "__NULL__", 1, true) then
        topLevel[#topLevel + 1] = db:is_nil(database[key])
      elseif find(value, "_", 1, true) or find(value, "%", 1, true) then
        topLevel[#topLevel + 1] = db:like(database[key], value)
      elseif find(value, "::", 1, true) then
        topLevel[#topLevel + 1] = db:between(database[key], match(value, "^(.-)::(.+)$"))
      else
        topLevel[#topLevel + 1] = db:eq(database[key], value)
      end
    end

  end

  return db:AND(unpack(topLevel))
end