docs: Add migration guide from version 2.x to 3.x

UPGRADING.md

@@ -13,8 +13,188 @@ limitations under the License.
 
 # 3.0.0 Migration Guide
 
-TODO
+## New Required Dependencies
 
+Some of the previously optional dependencies are now *required* in `3.x` versions of the
+library, namely
+[google-cloud-bigquery-storage](https://pypi.org/project/google-cloud-bigquery-storage/)
+(minimum version `2.0.0`) and [pyarrow](https://pypi.org/project/pyarrow/) (minimum
+version `3.0.0`).
+
+The behavior of some of the package "extras" has thus also changed:
+ * The `pandas` extra now requires the [db-types](https://pypi.org/project/db-dtypes/)
+   package.
+ * The `bqstorage` extra has been preserved for comaptibility reasons, but it is now a
+   no-op and should be omitted when installing the BigQuery client library.
+
+   **Before:**
+   ```
+   $ pip install google-cloud-bigquery[bqstorage]
+   ```
+
+   **After:**
+   ```
+   $ pip install google-cloud-bigquery
+   ```
+
+ * The `bignumeric_type` extra has been removed, as `BIGNUMERIC` type is now
+   automatically supported. That extra should thus not be used.
+
+   **Before:**
+   ```
+   $ pip install google-cloud-bigquery[bignumeric_type]
+   ```
+
+   **After:**
+   ```
+   $ pip install google-cloud-bigquery
+   ```
+
+
+## Type Annotations
+
+The library is now type-annotated and declares itself as such. If you use a static
+type checker such as `mypy`, you might start getting errors in places where
+`google-cloud-bigquery` package is used.
+
+It is recommended to update your code and/or type annotations to fix these errors, but
+if this is not feasible in the short term, you can temporarily ignore type annotations
+in `google-cloud-bigquery`, for example by using a special `# type: ignore` comment:
+
+```py
+from google.cloud import bigquery  # type: ignore
+```
+
+But again, this is only recommended as a possible short-term workaround if immediately
+fixing the type check errors in your project is not feasible.
+
+## Re-organized Types
+
+The auto-generated parts of the library has been removed, and proto-based types formerly
+found in `google.cloud.bigquery_v2` have been replaced by the new implementation (but
+see the [section](#legacy-types) below).
+
+For example, the standard SQL data types should new be imported from a new location:
+
+**Before:**
+```py
+from google.cloud.bigquery_v2 import StandardSqlDataType
+from google.cloud.bigquery_v2.types import StandardSqlField
+from google.cloud.bigquery_v2.types.standard_sql import StandardSqlStructType
+```
+
+**After:**
+```py
+from google.cloud.bigquery import StandardSqlDataType
+from google.cloud.bigquery.standard_sql import StandardSqlField
+from google.cloud.bigquery.standard_sql import StandardSqlStructType
+```
+
+The `TypeKind` enum defining all possible SQL types for schema fields has been renamed
+and is not nested anymore under `StandardSqlDataType`:
+
+
+**Before:**
+```py
+from google.cloud.bigquery_v2 import StandardSqlDataType
+
+if field_type == StandardSqlDataType.TypeKind.STRING:
+    ...
+```
+
+**After:**
+```py
+
+from google.cloud.bigquery import StandardSqlTypeNames
+
+if field_type == StandardSqlTypeNames.STRING:
+    ...
+```
+
+
+## Issuing queries with `Client.create_job` preserves destination table
+
+The `Client.create_job` method no longer removes the destination table from a
+query job's configuration. Destination table for the query can thus be
+explicitly defined by the user.
+
+
+## Changes to data types when reading a pandas DataFrame
+
+The default dtypes returned by the `to_dataframe` method have changed.
+
+* Now, the BigQuery `BOOLEAN` data type maps to the pandas `boolean` dtype.
+  Previously, this mapped to the pandas `bool` dtype when the column did not
+  contain `NULL` values and the pandas `object` dtype when `NULL` values are
+  present.
+* Now, the BigQuery `INT64` data type maps to the pandas `Int64` dtype.
+  Previously, this mapped to the pandas `int64` dtype when the column did not
+  contain `NULL` values and the pandas `float64` dtype when `NULL` values are
+  present.
+* Now, the BigQuery `DATE` data type maps to the pandas `dbdate` dtype, which
+  is provided by the
+  [db-dtypes](https://googleapis.dev/python/db-dtypes/latest/index.html)
+  package. If any date value is outside of the range of
+  [pandas.Timestamp.min](https://pandas.pydata.org/docs/reference/api/pandas.Timestamp.min.html)
+  (1677-09-22) and
+  [pandas.Timestamp.max](https://pandas.pydata.org/docs/reference/api/pandas.Timestamp.max.html)
+  (2262-04-11), the data type maps to the pandas `object` dtype. The
+  `date_as_object` parameter has been removed.
+* Now, the BigQuery `TIME` data type maps to the pandas `dbtime` dtype, which
+  is provided by the
+  [db-dtypes](https://googleapis.dev/python/db-dtypes/latest/index.html)
+  package.
+
+
+## Changes to data types loading a pandas DataFrame
+
+In the absence of schema information, pandas columns with naive
+`datetime64[ns]` values, i.e. without timezone information, are recognized and
+loaded using the `DATETIME` type.  On the other hand, for columns with
+timezone-aware `datetime64[ns, UTC]` values, the `TIMESTAMP` type is continued
+to be used.
+
+## Changes to `Model`, `Client.get_model`, `Client.update_model`, and `Client.list_models`
+
+The types of several `Model` properties have been changed.
+
+- `Model.feature_columns` now returns a sequence of `google.cloud.bigquery.standard_sql.StandardSqlField`.
+- `Model.label_columns` now returns a sequence of `google.cloud.bigquery.standard_sql.StandardSqlField`.
+- `Model.model_type` now returns a string.
+- `Model.training_runs` now returns a sequence of dictionaries, as recieved from the [BigQuery REST API](https://cloud.google.com/bigquery/docs/reference/rest/v2/models#Model.FIELDS.training_runs).
+
+<a name="legacy-protobuf-types"></a>
+## Legacy Protocol Buffers Types
+
+For compatibility reasons, the legacy proto-based types still exists as static code
+and can be imported:
+
+```py
+from google.cloud.bigquery_v2 import Model  # a sublcass of proto.Message
+```
+
+Mind, however, that importing them will issue a warning, because aside from
+being importable, these types **are not maintained anymore**. They may differ
+both from the types in `google.cloud.bigquery`, and from the types supported on
+the backend.
+
+### Maintaining compatibility with `google-cloud-bigquery` version 2.0
+
+If you maintain a library or system that needs to support both
+`google-cloud-bigquery` version 2.x and 3.x, it is recommended that you detect
+when version 2.x is in use and convert properties that use the legacy protocol
+buffer types, such as `Model.training_runs`, into the types used in 3.x.
+
+Call the [`to_dict`
+method](https://proto-plus-python.readthedocs.io/en/latest/reference/message.html#proto.message.Message.to_dict)
+on the protocol buffers objects to get a JSON-compatible dictionary.
+
+```py
+from google.cloud.bigquery_v2 import Model
+
+training_run: Model.TrainingRun = ...
+training_run_dict = training_run.to_dict()
+```
 
 # 2.0.0 Migration Guide
 

docs/index.rst

@@ -30,7 +30,8 @@ API Reference
 Migration Guide
 ---------------
 
-See the guide below for instructions on migrating to the 2.x release of this library.
+See the guides below for instructions on migrating from older to newer *major* releases
+of this library (from ``1.x`` to ``2.x``, or from ``2.x`` to ``3.x``).
 
 .. toctree::
     :maxdepth: 2