Update: We recommend checking out Buf, which is under active development. There are a ton of docs for getting started, including for migration from Prototool.
Protobuf is one of the best interface description
languages out there - it’s widely adopted, and after over 15 years of use, it’s practically
bulletproof. However, working with Protobuf and maintaining consistency across your Protobuf files
can be a pain - protoc, while being a tool that has stood the test of time, is non-trivial to
use, and the Protobuf community has not developed common standards with regards to stub generation.
Prototool aims to solve this by making working with Protobuf much simpler.
Prototool lets you:
- Handle installation of
protoc and the import of all of the Well-Known Types behind the scenes
in a platform-independent manner.
- Standardize building of your Protobuf files with a common configuration.
- Lint your Protobuf files with common linting rules according to
Google’ Style Guide,
Uber’s V1 Style Guide,
Uber’s V2 Style Guide, or your own set of configured lint rules.
- Format your Protobuf files in a consistent manner.
- Create Protobuf files from a template that passes lint, taking care of
package naming for you.
- Generate stubs using any plugin based on a simple configuration file,
including handling imports of all the Well-Known Types.
- Call gRPC endpoints with ease, taking care of the JSON to binary
conversion for you.
- Check for breaking changes on a per-package basis, verifying that your
API never breaks.
- Output errors and lint failures in a common
file:line:column:message format, making integration
with editors possible, Vim integration is provided out of the box.
Prototool accomplishes this by downloading and calling protoc on the fly for you, handing error
messages from protoc and your plugins, and using the generated FileDescriptorSets for internal
functionality, as well as wrapping a few great external libraries already in the Protobuf
ecosystem. Compiling, linting and formatting commands run in around 3/100ths of second for a single
Protobuf file, or under a second for a larger number (500+) of Protobuf files.
Table Of Contents
Installation
Prototool can be installed on Mac OS X or Linux through a variety of methods.
See install.md for full instructions.
Quick Start
We’ll start with a general overview of the commands. There are more commands, and we will get into]
usage below, but this shows the basic functionality.
prototool help
prototool lint idl/uber # search for all .proto files recursively, obeying exclude_paths in prototool.yaml or prototool.json files
prototool lint # same as "prototool lint .", by default the current directory is used in directory mode
prototool create foo.proto # create the file foo.proto from a template that passes lint
prototool files idl/uber # list the files that will be used after applying exclude_paths from corresponding prototool.yaml or prototool.json files
prototool lint --list-linters # list all current lint rules being used
prototool lint --list-all-lint-groups # list all available lint groups, currently "google" and "uber"
prototool compile idl/uber # make sure all .proto files in idl/uber compile, but do not generate stubs
prototool generate idl/uber # generate stubs, see the generation directives in the config file example
prototool grpc idl/uber --address 0.0.0.0:8080 --method foo.ExcitedService/Exclamation --data '{"value":"hello"}' # call the foo.ExcitedService method Exclamation with the given data on 0.0.0.0:8080
prototool descriptor-set --include-imports idl/uber # generate a FileDescriptorSet for all files under idl/uber, outputting to stdout, a given file, or a temporary file
prototool break check idl/uber --git-branch master # check for breaking changes as compared to the Protobuf definitions in idl/uber on the master branch
Full Example
See the example directory.
The make command make example runs prototool while installing the necessary plugins.
Configuration
Prototool operates using a config file named either prototool.yaml or prototool.json. Only one
of prototool.yaml or prototool.json can exist in a given directory. For non-trivial use, you
should have a config file checked in to at least the root of your repository. It is important
because the directory of an associated config file is passed to protoc as an include directory
with -I, so this is the logical location your Protobuf file imports should start from.
Recommended base config file:
protoc:
version: 3.11.0
lint:
group: uber2
See protoc.md for how Prototool handles working with protoc.
The command prototool config init will generate a config file in the current directory with the
currently recommended options set.
When specifying a directory or set of files for Prototool to operate on, Prototool will search for
config files for each directory starting at the given path, and going up a directory until hitting
root. If no config file is found, Prototool will use default values and operate as if there was a
config file in the current directory, including the current directory with -I to protoc.
If multiple prototool.yaml or prototool.json files are found that match the input directory or
files, an error will be returned.
See etc/config/example/prototool.yaml all available
options.
File Discovery
In most Prototool commands, you will see help along the following lines:
$ prototool help lint
Lint proto files and compile with protoc to check for failures.
Usage:
prototool lint [dirOrFile] [flags]
dirOrFile can take two forms:
- You can specify exactly one directory. If this is done, Prototool goes up until it finds a
prototool.yaml or prototool.json file (or uses the current directory if none is found), and
then uses this config for all .proto files under the given directory recursively, except for
files in the excludes lists in prototool.yaml or prototool.json files.
- You can specify exactly one file. This has the effect as if you specified the directory of this
file (using the logic above), but errors are only printed for that file. This is useful for
e.g. Vim integration.
- You can specify nothing. This has the effect as if you specified the current directory as the
directory.
The idea with “directory builds” is that you often need more than just one file to do a protoc
call, for example if you have types in other files in the same package that are not referenced by
their fully-qualified name, and/or if you need to know what directories to specify with -I to
protoc (by default, the directory of the prototool.yaml or prototool.json file is used).
Command Overview
Let’s go over some of the basic commands.
Create a prototool.yaml file in the current directory with the currently recommended options set.
Pass the --document flag to generate a prototool.yaml file with all other options documented
and commented out.
Pass the --uncomment flag to generate prototool.yaml file with all options documented but
uncommented.
See etc/config/example/prototool.yaml for the config file
that prototool config init --uncomment generates.
Compile your Protobuf files, but do not generate stubs. This has the effect of calling protoc
with -o /dev/null.
Pass the --dry-run flag to see the protoc commands that Prototool runs behind the scenes.
Compile your Protobuf files and generate stubs according to the rules in your prototool.yaml or
prototool.json file.
See etc/config/example/prototool.yaml for all available
options. There are special options available for Golang plugins, and plugins that output a single
file instead of a set of files. Specifically, you can output a single JAR for the built-in protoc
java plugin, and you can output a file with the serialized FileDescriptorSet using the built-in
protoc descriptor_set plugin, optionally also calling --include_imports and/or
--include_source_info.
Pass the --dry-run flag to see the protoc commands that Prototool runs behind the scenes.
See example/proto/prototool.yaml for a full example.
Lint rules can be set using the configuration file. See the configuration at
etc/config/example/prototool.yaml for all available
options. There are three pre-configured groups of rules, the setting of which is integral to the
prototool lint, prototool create, and prototool format commands:
uber2: This lint group follows the V2 Uber Style Guide, and makes some
modifications to more closely follow the Google Cloud APIs file structure, as well as adding even
more rules to enforce more consistent development patterns. This is the lint group we recommend
using.uber1: This lint group follows the V1 Uber Style Guide. For
backwards compatibility reasons, this is the default lint group, however we recommend using the
uber2 lint group.google: This lint group follows the
Google Style Guide. This is a small
group of rules meant to enforce basic naming. The style guide is copied to
etc/style/google/google.proto.
The flag --generate-ignores will help with migrating to a given lint group by generating
the configuration to ignore existing lint failures on a per-file basis.
See lint.md for full instructions.
Format a Protobuf file and print the formatted file to stdout. There are flags to perform different
actions:
-d Write a diff instead.-f Fix the file according to the Style Guide. This will have different behavior if the uber2
lint group is set.-l Write a lint error in the form file:line:column:message if a file is unformatted.-w Overwrite the existing file instead.
Create Protobuf files from a template. With the provided Vim integration, this will automatically
create new files that pass lint when a new file is opened.
See create.md for full instructions.
Print the list of all files that will be used given the input dirOrFile. Useful for debugging.
Protobuf is a great way to represent your APIs and generate stubs in each language you develop
with. As such, Protobuf APIs should be stable so as not to break consumers across repositories.
Even in a monorepo context, making sure that your Protobuf APIs do not introduce breaking
changes is important so that different deployed versions of your services do not have
wire incompatibilities.
Prototool exposes a breaking change detector through the prototool break check command. This will
check your current Protobuf definitions against a past version of your Protobuf definitions to see
if there are any source or wire incompatible changes. Some notes on this command:
- The breaking change detection operates on a per-package basis, not per-file - definitions
can be moved between files within the same Protobuf package without being considered breaking.
- The breaking change detector can either check against a given git branch or tag, or it can check
against a previous state saved with the
prototool break descriptor-set command.
- The breaking change detector understands the concept of beta vs. stable packages, discussed
in the V2 Style Guide. By default, the breaking change
detector will not check beta packages for breaking changes, and will not allow stable packages to
depend on beta packages, however both of these options are configurable in your
prototool.yaml
file.
See breaking.md for full instructions.
Produce a serialized FileDescriptorSet for all Protobuf definitions. By default, the serialized
FileDescriptorSet is printed to stdout. There are a few options:
--include-imports, --include-source-info are analagous to protoc’s --include_imports, --include_source_info flags.--json outputs the FileDescriptorSet as JSON instead of binary.-o writes the FileDescriptorSet to the given output file path.--tmp writes the FileDescriptorset to a temporary file and prints the file path.
The outputted FileDescriptorSet is a merge of all produced FileDescriptorSets for each
Protobuf package compiled.
This command is useful in a few situations.
One such situation is with external gRPC tools such as grpcurl
or ghz. Both tools take a path to a serialized FileDescriptorSet for use to
figure out the request/response structure of RPCs when the gRPC reflection service is not available.
prototool descriptor-set can be used to generate these FileDescriptorSets on the fly.
grpcurl -protoset $(prototool descriptor-set --include-imports --tmp) ...
ghz -protoset $(prototool descriptor-set --include-imports --tmp) ...
You can also just save the file once and not re-compile each time.
prototool descriptor-set --include-imports -o descriptor_set.bin
grpcurl -protoset descriptor_set.bin ...
ghz -protoset descriptor_set.bin ...
Another situation is to use jq to make arbitrary queries on your Protobuf definitions.
For example, if your Protobuf definitions are in path/to/proto, the following will print
all message names.
prototool descriptor-set path/to/proto --json | \
jq '.file[] | select(.messageType != null) | .messageType[] | .name' | \
sort | uniq
Call a gRPC endpoint using a JSON input. What this does behind the scenes:
- Compiles your Protobuf files with
protoc, generating a FileDescriptorSet.
- Uses the
FileDescriptorSet to figure out the request and response type for the endpoint, and to
convert the JSON input to binary.
- Calls the gRPC endpoint.
- Uses the
FileDescriptorSet to convert the resulting binary back to JSON, and prints it out for
you.
See grpc.md for full instructions.
Tips and Tricks
Prototool is meant to help enforce a consistent development style for Protobuf, and as such you
should follow some basic rules:
- Have all your imports start from the directory your
prototool.yaml or prototool.json file is
in. While there is a configuration option protoc.includes to denote extra include directories,
this is not recommended.
- Have all Protobuf files in the same directory use the same
package.
- Do not use long-form
go_package values, ie use foopb, not github.com/bar/baz/foo;foopb.
This helps prototool generate do the best job.
Vim Integration
This repository is a self-contained plugin for use with the
ALE Lint Engine. The Vim integration will currently compile, provide
lint errors, do generation of your stubs, and format your files on save. It will also optionally
create new files from a template when opened.
See vim.md for full instructions.
Stability
Prototool is generally available, and conforms to SemVer, so Prototool will
not have any breaking changes on a given major version, with some exceptions:
- Commands under the
x top-level command are experimental, and may change or be deleted between
minor versions of Prototool. We expect such commands to be promoted to stable within a few minor
releases, however development is still in-progress.
- The output of the formatter may change between minor versions. This has not happened yet, but we
may change the format in the future to reflect things such as max line lengths.
- The breaking change detector’s output format currently does not output filename, line, or column.
This is an expected upgrade in the future, so the output will likely change. This is viewed as
purely an upgrade, so until this is done, do not parse
prototool break check output in scripts.
- The breaking change detector may have additional checks added between minor versions, and
therefore a change that might not have been breaking previously might become a breaking change.
This may become stable in the near future, and at this time we’ll denote that no more checks
will be added.
Development
See development.md for concerns related to Prototool development.
See maintenance.md for maintenance-related tasks.
FAQ
See faq.md for answers to frequently asked questions.
Special Thanks
Prototool uses some external libraries that deserve special mention and thanks for their
contribution to Prototool’s functionality:
- github.com/emicklei/proto - The Golang Protobuf parsing
library that started it all, and is still used for the linting and formatting functionality. We
can’t thank Ernest Micklei enough for his help and putting up with all the
filed issues.
- github.com/jhump/protoreflect - Used for the JSON to
binary and back conversion. Josh Humphries is an amazing developer, thank you so much.
- github.com/fullstorydev/grpcurl - Still used for the
gRPC functionality. Again a thank you to Josh Humphries and the team over at FullStory for their
work.
评论