X-Git-Url: http://git.veekun.com/zzz-pokedex.git/blobdiff_plain/bad044d1d8753cf96a3cf0dc2236ffcb416dcc47..638bc0eb815f3cb8bff0bfc027d4d0d931ab7273:/pokedex/__init__.py

diff --git a/pokedex/__init__.py b/pokedex/__init__.py
index 138c58b..02b8ce5 100644
--- a/pokedex/__init__.py
+++ b/pokedex/__init__.py
@@ -1,7 +1,10 @@
 # encoding: utf8
 import sys
 
-from .db import connect, metadata
+from sqlalchemy.exc import IntegrityError
+import sqlalchemy.types
+
+from .db import connect, metadata, tables as tables_module
 
 def main():
     if len(sys.argv) <= 1:
@@ -18,40 +21,114 @@ def main():
         help()
 
 
-def csvimport(engine_uri, dir='.'):
+def csvimport(engine_uri, directory='.'):
     import csv
 
     from sqlalchemy.orm.attributes import instrumentation_registry
 
-    session = connect(engine_uri)
+    # Use autocommit in case rows fail due to foreign key incest
+    session = connect(engine_uri, autocommit=True, autoflush=False)
 
     metadata.create_all()
 
-    # This is a secret attribute on a secret singleton of a secret class that
-    # appears to hopefully contain all registered classes as keys.
-    # There is no other way to accomplish this, as far as I can tell.
-    # Fuck.
-    for table in sorted(instrumentation_registry.manager_finders.keys(),
-                        key=lambda self: self.__table__.name):
-        table_name = table.__table__.name
-        print table_name
-
-        reader = csv.reader(open("%s/%s.csv" % (dir, table_name), 'rb'), lineterminator='\n')
-        columns = [unicode(column) for column in reader.next()]
+    # SQLAlchemy is retarded and there is no way for me to get a list of ORM
+    # classes besides to inspect the module they all happen to live in for
+    # things that look right.
+    table_base = tables_module.TableBase
+    orm_classes = {}  # table object => table class
+
+    for name in dir(tables_module):
+        # dir() returns strings!  How /convenient/.
+        thingy = getattr(tables_module, name)
+
+        if not isinstance(thingy, type):
+            # Not a class; bail
+            continue
+        elif not issubclass(thingy, table_base):
+            # Not a declarative table; bail
+            continue
+        elif thingy == table_base:
+            # Declarative table base, so not a real table; bail
+            continue
+
+        # thingy is definitely a table class!  Hallelujah.
+        orm_classes[thingy.__table__] = thingy
+
+    # Okay, run through the tables and actually load the data now
+    for table_obj in metadata.sorted_tables:
+        table_class = orm_classes[table_obj]
+        table_name = table_obj.name
+
+        # Print the table name but leave the cursor in a fixed column
+        print table_name + '...', ' ' * (40 - len(table_name)),
+        sys.stdout.flush()
+
+        try:
+            csvfile = open("%s/%s.csv" % (directory, table_name), 'rb')
+        except IOError:
+            # File doesn't exist; don't load anything!
+            print 'no data!'
+            continue
+
+        reader = csv.reader(csvfile, lineterminator='\n')
+        column_names = [unicode(column) for column in reader.next()]
+
+        # Self-referential tables may contain rows with foreign keys of
+        # other rows in the same table that do not yet exist.  We'll keep
+        # a running list of these and try inserting them again after the
+        # rest are done
+        failed_rows = []
 
         for csvs in reader:
-            row = table()
-
-            for column, value in zip(columns, csvs):
-                value = value.decode('utf-8')
-                setattr(row, column, value)
-
-            session.add(row)
-
-        session.commit()
-
-
-def csvexport(engine_uri, dir='.'):
+            row = table_class()
+
+            for column_name, value in zip(column_names, csvs):
+                column = table_obj.c[column_name]
+                if column.nullable and value == '':
+                    # Empty string in a nullable column really means NULL
+                    value = None
+                elif isinstance(column.type, sqlalchemy.types.Boolean):
+                    # Boolean values are stored as string values 0/1, but both
+                    # of those evaluate as true; SQLA wants True/False
+                    if value == '0':
+                        value = False
+                    else:
+                        value = True
+                else:
+                    # Otherwise, unflatten from bytes
+                    value = value.decode('utf-8')
+
+                setattr(row, column_name, value)
+
+            try:
+                session.add(row)
+                session.flush()
+            except IntegrityError as e:
+                failed_rows.append(row)
+
+        # Loop over the failed rows and keep trying to insert them.  If a loop
+        # doesn't manage to insert any rows, bail.
+        do_another_loop = True
+        while failed_rows and do_another_loop:
+            do_another_loop = False
+
+            for i, row in enumerate(failed_rows):
+                try:
+                    session.add(row)
+                    session.flush()
+
+                    # Success!
+                    del failed_rows[i]
+                    do_another_loop = True
+                except IntegrityError as e:
+                    pass
+
+        if failed_rows:
+            print len(failed_rows), "rows failed"
+        else:
+            print 'loaded'
+
+def csvexport(engine_uri, directory='.'):
     import csv
     session = connect(engine_uri)
 
@@ -59,7 +136,8 @@ def csvexport(engine_uri, dir='.'):
         print table_name
         table = metadata.tables[table_name]
 
-        writer = csv.writer(open("%s/%s.csv" % (dir, table_name), 'wb'), lineterminator='\n')
+        writer = csv.writer(open("%s/%s.csv" % (directory, table_name), 'wb'),
+                            lineterminator='\n')
         columns = [col.name for col in table.columns]
         writer.writerow(columns)