Feature request: ability to specify output package name to protoc

[geigerj]

Context: googleapis/artman#97. This currently affects Python only.

What

We'd like the ability to specify the output package name in calls to protoc (actually, we are currently using the gRPC Python wrapper grpc.tools.protoc). Currently the Python package corresponds exactly to the proto directory structure, for example, foo/bar/baz.proto generates the Python module foo.bar.baz_pb2. It should be possible to substitute an arbitrary value for foo.bar. Currently this only seems to be possible by moving the input proto to a different directory; ideally this could just be an argument to protoc.

Why

We do code generation from googleapis/googleapis, which by convention keeps protos in a directory structure the same as the proto package. Sometimes the proto package is not usable as a Python package (for example, it may already be used by google-cloud-python). Ideally we shouldn't need to move the protos in googleapis/googleapis around just for Python code generation.

[bjwatson]

A related issue is that internal imports use the proto package rather than the directory where the proto file is found, so we have to manually fix these imports when we do this "change proto directory" trick to change the Python package. An example is the dependency of image_annotator_pb2.py on geometry_pb2.py in Cloud Vision v1.

[xfxyjwf]

@haberman

[xfxyjwf]

@bjwatson I don't quite understand the issue you mentioned. What do you mean by "internal imports use the proto package"? For example:

// foo/bar/baz.proto
package a.b.c;
message X {}

"a.b.c" is the proto package. foo.bar is the python package which is the same as the file path. What 's the "internal imports" you are referring to?

[bjwatson]

@xfxyjwf Take for example Cloud Vision image_annotator.proto.

Move it into a path like google/cloud/grpc/vision/v1 (which we have to do to for technical reasons related to our Python google-cloud and GAPIC client libraries), and then use python -m grpc.tools.protoc to build google/cloud/grpc/vision/v1/image_annotator_pb2.py.

You will see that everything builds correctly for the new path (google/cloud/grpc/vision/v1), except for one line which says:

# note no 'grpc' is in the path
from google.cloud.vision.v1 import geometry_pb2 as google_dot_cloud_dot_vision_dot_v1_dot_geometry__pb2

Maybe this won't be an issue once Jacob's feature request here is implemented (and we don't have to move the proto to a different directory), but I want to make sure this issue with our current workaround is understood.

[xfxyjwf]

@bjwatson Have you updated the import statement in image_annotator.proto to point to the correct path of "google/cloud/grpc/vision/v1/geometry.proto"? If you haven't, what you observed is the right behavior.

[bjwatson]

@xfxyjwf Yes, I have to manually update the import statement every time I regenerate image_annotator_pb2.py. We will work around this in the short term by scripting it, but the overarching purpose of this ticket is to future-proof what we're doing.

Basically, we need to be able to reliably build Python code that has a different path than the proto package. We have a recipe that allows us to do this right now, but there's no guarantee that future protobuf and/or gRPC changes won't break it.

[sigmonsays]

+1

[xfxyjwf]

We talked about this issue in a protobuf team meeting and it was pointed out that supporting an alternative python package requires build tool changes. Right now Google's internal build tool blaze assumes a .proto file foo/bar.proto will generate foo/bar_pb2.py. There needs a way to teach blaze that certain .proto files are generating python code in a different package. I can imagine it's not going to be an easy change. Comparing the complexity (update both protobuf and blaze and potentially lots of other tools that analyzes BUILD files) against its benefit (give some people an easier way to avoid name conflict) I would say it's not worthwhile to support.

[bjwatson]

@xfxyjwf Can you hold this issue open for a week or so? I'd like to discuss the trade-offs with you, but I have a few other urgent items I need to address first.

[xfxyjwf]

@bjwatson sure, I will keep this issue open for a while.