From: Mike Bayer <mike_mp@zzzcomputing.com>
Date: Sun, 9 Nov 2014 03:55:51 +0000 (-0500)
Subject: documenting
X-Git-Tag: rel_0_7_0~43
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=8ab32c426e807c404cfed0b0edd8daf8973deebf;p=thirdparty%2Fsqlalchemy%2Falembic.git

documenting
---

diff --git a/docs/build/tutorial.rst b/docs/build/tutorial.rst
index 652a7197..ce1d8fc1 100644
--- a/docs/build/tutorial.rst
+++ b/docs/build/tutorial.rst
@@ -926,17 +926,81 @@ correspond to the new structure, to transfer the data from the existing
 table to the new one, then drop the old table.  In order to accommodate this
 workflow in a way that is reasonably predictable, while remaining compatible
 with other databases, Alembic provides the "batch" operations context.
-This context provides the table recreate plus data copy operation for SQLite,
-and can also be instructed to use this flow for other databases as well,
-providing an alternative scheme in some cases where the blocking behavior of
-some ALTER statements is to be avoided.
 
 Within this context, a relational table is named, and then a series of
-any number of mutation operations to that table then proceed within the
-block.  When the context is complete, the above mentioned table-recreate
-and data copy operation proceeds, but *only* if appropriate for the target
-database; for other databases, the batch context proceeds using traditional
-ALTER statements, unless the context is set to "always".
+any number of mutation operations to that table may be specified within
+the block.  When the context is complete, a process begins whereby the
+existing table structure is reflected from the database, a new version of
+this table is created with the given changes, data is copied from the
+old table to the new table, and finally the old table is dropped and the
+new one renamed to the original name.
+
+The :meth:`.Operations.batch_alter_table` provides the gateway to this
+process::
+
+    with op.batch_alter_table("some_table") as batch_op:
+        batch_op.add_column(Column('foo', Integer))
+        batch_op.drop_column('bar')
+
+With the above directives, on a SQLite backend we would see SQL like::
+
+    CREATE TABLE _alembic_batch_temp (
+      id INTEGER NOT NULL,
+      foo INTEGER,
+      PRIMARY KEY (id)
+    );
+    INSERT INTO _alembic_batch_temp (id) SELECT some_table.id FROM some_table;
+    DROP TABLE some_table;
+    ALTER TABLE _alembic_batch_temp RENAME TO some_table;
+
+On other backends, we'd see the usual ``ALTER`` statements done as though
+there were no batch directive - the batch context by default only does
+the "move and copy" process if SQLite is in use.   It can be configured
+to run "move and copy" on other backends as well, if desired.
+
+Batch mode with Autogenerate
+----------------------------
+
+The syntax of batch mode is essentially that :meth:`.Operations.batch_alter_table`
+is used to enter a batch block, and the returned :class:`.BatchOperations` context
+works just like the regular :class:`.Operations` context, except that
+the "table name" and "schema name" arguments are omitted.
+
+To support rendering of migration commands in batch mode for autogenerate,
+configure the :paramref:`.EnvironmentContext.configure.render_as_batch`
+flag in ``env.py``::
+
+    context.configure(
+        connection=connection,
+        target_metadata=target_metadata,
+        render_as_batch=True
+    )
+
+Autogenerate will now generate along the lines of::
+
+    def upgrade():
+        ### commands auto generated by Alembic - please adjust! ###
+        with op.batch_alter_table('address', schema=None) as batch_op:
+            batch_op.add_column(sa.Column('street', sa.String(length=50), nullable=True))
+
+
+Batch mode with databases other than SQLite
+--------------------------------------------
+
+There's an odd use case some shops have, where the "move and copy" style
+of migration is useful in some cases for databases that do already support
+ALTER.   There's some cases where an ALTER operation may block access to the
+table for a long time, which might not be acceptable.   "move and copy" can
+be made to work on other backends, though with a few extra caveats.
+
+The batch mode directive will run the "recreate" system regardless of
+backend if the flag ``recreate='always'`` is passed::
+
+    with op.batch_alter_table("some_table", recreate='always') as batch_op:
+        batch_op.add_column(Column('foo', Integer))
+
+
+
 
 TODO: "recreate" doesn't work very well for Postgresql due to constraint naming
 TODO: caveats, beta status