pokedex/db/load.py

   1 """CSV to database or vice versa."""
   2 import csv
   3 import os.path
   4 import pkg_resources
   5 import re
   6 import sys
   7
   8 from sqlalchemy.orm.attributes import instrumentation_registry
   9 import sqlalchemy.sql.util
  10 import sqlalchemy.types
  11
  12 from pokedex.db import metadata
  13 import pokedex.db.tables as tables
  14
  15
  16 def _wildcard_char_to_regex(char):
  17     """Converts a single wildcard character to the regex equivalent."""
  18
  19     if char == '?':
  20         return '.?'
  21     elif char == '*':
  22         return '.*'
  23     else:
  24         return re.escape(char)
  25
  26 def _wildcard_glob_to_regex(glob):
  27     """Converts a single wildcard glob to a regex STRING."""
  28
  29     # If it looks like a filename, make it not one
  30     if '.' in glob or '/' in glob:
  31         _, filename = os.path.split(glob)
  32         table_name, _ = os.path.splitext(filename)
  33         glob = table_name
  34
  35     return u''.join(map(_wildcard_char_to_regex, glob))
  36
  37 def _wildcards_to_regex(strings):
  38     """Converts a list of wildcard globs to a single regex object."""
  39
  40     regex_parts = map(_wildcard_glob_to_regex, strings)
  41
  42     regex = '^(?:' + '|'.join(regex_parts) + ')$'
  43
  44     return re.compile(regex)
  45
  46
  47 def _get_verbose_prints(verbose):
  48     """If `verbose` is true, returns two functions: one for printing a starting
  49     message, and the other for printing a success or failure message when
  50     finished.
  51
  52     If `verbose` is false, returns two no-op functions.
  53     """
  54
  55     if verbose:
  56         import sys
  57         def print_start(thing):
  58             # Truncate to 66 characters, leaving 10 characters for a success
  59             # or failure message
  60             truncated_thing = thing[0:66]
  61
  62             # Also, space-pad to keep the cursor in a known column
  63             num_spaces = 66 - len(truncated_thing)
  64
  65             print "%s...%s" % (truncated_thing, ' ' * num_spaces),
  66             sys.stdout.flush()
  67
  68         def print_done(msg='ok'):
  69             print msg
  70             sys.stdout.flush()
  71
  72         return print_start, print_done
  73
  74     # Not verbose; return dummies
  75     def dummy(*args, **kwargs):
  76         pass
  77
  78     return dummy, dummy
  79
  80
  81 def load(session, tables=[], directory=None, drop_tables=False, verbose=False):
  82     """Load data from CSV files into the given database session.
  83
  84     Tables are created automatically.
  85
  86     `session`
  87         SQLAlchemy session to use.
  88
  89     `tables`
  90         List of tables to load.  If omitted, all tables are loaded.
  91
  92     `directory`
  93         Directory the CSV files reside in.  Defaults to the `pokedex` data
  94         directory.
  95
  96     `drop_tables`
  97         If set to True, existing `pokedex`-related tables will be dropped.
  98
  99     `verbose`
 100         If set to True, status messages will be printed to stdout.
 101     """
 102
 103     # First take care of verbosity
 104     print_start, print_done = _get_verbose_prints(verbose)
 105
 106
 107     if not directory:
 108         directory = pkg_resources.resource_filename('pokedex', 'data/csv')
 109
 110     if tables:
 111         regex = _wildcards_to_regex(tables)
 112         table_names = filter(regex.match, metadata.tables.keys())
 113     else:
 114         table_names = metadata.tables.keys()
 115
 116     table_objs = [metadata.tables[name] for name in table_names]
 117     table_objs = sqlalchemy.sql.util.sort_tables(table_objs)
 118
 119
 120     # Drop all tables if requested
 121     if drop_tables:
 122         print_start('Dropping tables')
 123         for table in reversed(table_objs):
 124             table.drop(checkfirst=True)
 125         print_done()
 126
 127     for table in table_objs:
 128         table.create()
 129     connection = session.connection()
 130
 131     # Okay, run through the tables and actually load the data now
 132     for table_obj in table_objs:
 133         table_name = table_obj.name
 134         insert_stmt = table_obj.insert()
 135
 136         print_start(table_name)
 137
 138         try:
 139             csvfile = open("%s/%s.csv" % (directory, table_name), 'rb')
 140         except IOError:
 141             # File doesn't exist; don't load anything!
 142             print_done('missing?')
 143             continue
 144
 145         reader = csv.reader(csvfile, lineterminator='\n')
 146         column_names = [unicode(column) for column in reader.next()]
 147
 148         # Self-referential tables may contain rows with foreign keys of other
 149         # rows in the same table that do not yet exist.  Pull these out and add
 150         # them to the session last
 151         # ASSUMPTION: Self-referential tables have a single PK called "id"
 152         deferred_rows = []  # ( row referring to id, [foreign ids we need] )
 153         seen_ids = {}       # primary key we've seen => 1
 154
 155         # Fetch foreign key columns that point at this table, if any
 156         self_ref_columns = []
 157         for column in table_obj.c:
 158             if any(_.references(table_obj) for _ in column.foreign_keys):
 159                 self_ref_columns.append(column)
 160
 161         new_rows = []
 162         def insert_and_commit():
 163             session.connection().execute(insert_stmt, new_rows)
 164             session.commit()
 165             new_rows[:] = []
 166
 167         for csvs in reader:
 168             row_data = {}
 169
 170             for column_name, value in zip(column_names, csvs):
 171                 column = table_obj.c[column_name]
 172                 if column.nullable and value == '':
 173                     # Empty string in a nullable column really means NULL
 174                     value = None
 175                 elif isinstance(column.type, sqlalchemy.types.Boolean):
 176                     # Boolean values are stored as string values 0/1, but both
 177                     # of those evaluate as true; SQLA wants True/False
 178                     if value == '0':
 179                         value = False
 180                     else:
 181                         value = True
 182                 else:
 183                     # Otherwise, unflatten from bytes
 184                     value = value.decode('utf-8')
 185
 186                 # nb: Dictionaries flattened with ** have to have string keys
 187                 row_data[ str(column_name) ] = value
 188
 189             # May need to stash this row and add it later if it refers to a
 190             # later row in this table
 191             if self_ref_columns:
 192                 foreign_ids = [row_data[_.name] for _ in self_ref_columns]
 193                 foreign_ids = [_ for _ in foreign_ids if _]  # remove NULL ids
 194
 195                 if not foreign_ids:
 196                     # NULL key.  Remember this row and add as usual.
 197                     seen_ids[row_data['id']] = 1
 198
 199                 elif all(_ in seen_ids for _ in foreign_ids):
 200                     # Non-NULL key we've already seen.  Remember it and commit
 201                     # so we know the old row exists when we add the new one
 202                     insert_and_commit()
 203                     seen_ids[row_data['id']] = 1
 204
 205                 else:
 206                     # Non-NULL future id.  Save this and insert it later!
 207                     deferred_rows.append((row_data, foreign_ids))
 208                     continue
 209
 210             # Insert row!
 211             new_rows.append(row_data)
 212
 213             # Remembering some zillion rows in the session consumes a lot of
 214             # RAM.  Let's not do that.  Commit every 1000 rows
 215             if len(new_rows) >= 1000:
 216                 insert_and_commit()
 217
 218         insert_and_commit()
 219
 220         # Attempt to add any spare rows we've collected
 221         for row_data, foreign_ids in deferred_rows:
 222             if not all(_ in seen_ids for _ in foreign_ids):
 223                 # Could happen if row A refers to B which refers to C.
 224                 # This is ridiculous and doesn't happen in my data so far
 225                 raise ValueError("Too many levels of self-reference!  "
 226                                  "Row was: " + str(row))
 227
 228             session.connection().execute(
 229                 insert_stmt.values(**row_data)
 230             )
 231             seen_ids[row_data['id']] = 1
 232         session.commit()
 233
 234         print_done()
 235
 236
 237
 238 def dump(session, tables=[], directory=None, verbose=False):
 239     """Dumps the contents of a database to a set of CSV files.  Probably not
 240     useful to anyone besides a developer.
 241
 242     `session`
 243         SQLAlchemy session to use.
 244
 245     `tables`
 246         List of tables to dump.  If omitted, all tables are dumped.
 247
 248     `directory`
 249         Directory the CSV files should be put in.  Defaults to the `pokedex`
 250         data directory.
 251
 252     `verbose`
 253         If set to True, status messages will be printed to stdout.
 254     """
 255
 256     # First take care of verbosity
 257     print_start, print_done = _get_verbose_prints(verbose)
 258
 259
 260     if not directory:
 261         directory = pkg_resources.resource_filename('pokedex', 'data/csv')
 262
 263     if tables:
 264         regex = _wildcards_to_regex(tables)
 265         table_names = filter(regex.match, metadata.tables.keys())
 266     else:
 267         table_names = metadata.tables.keys()
 268
 269     table_names.sort()
 270
 271
 272     for table_name in table_names:
 273         print_start(table_name)
 274         table = metadata.tables[table_name]
 275
 276         writer = csv.writer(open("%s/%s.csv" % (directory, table_name), 'wb'),
 277                             lineterminator='\n')
 278         columns = [col.name for col in table.columns]
 279         writer.writerow(columns)
 280
 281         primary_key = table.primary_key
 282         for row in session.query(table).order_by(*primary_key).all():
 283             csvs = []
 284             for col in columns:
 285                 # Convert Pythony values to something more universal
 286                 val = getattr(row, col)
 287                 if val == None:
 288                     val = ''
 289                 elif val == True:
 290                     val = '1'
 291                 elif val == False:
 292                     val = '0'
 293                 else:
 294                     val = unicode(val).encode('utf-8')
 295
 296                 csvs.append(val)
 297
 298             writer.writerow(csvs)
 299
 300         print_done()