Contributing to FIDL

Overview

The FIDL toolchain is composed of roughly three parts:

1. Front-end, a.k.a. fidlc
   • Parses and validates .fidl files
   • Calculates size, alignment, and offset of various structures
   • Produces a JSON IR (Intermediate Representation)
2. Back-end
   • Works off the IR (except the C back-end)
   • Produces target language specific code, which ties into the libraries for that language
3. Runtime Libraries
   • Implement encoding/decoding/validation of messages
   • Method dispatching mechanics

Code location

Compiler front-end

The front-end lives at //zircon/tools/fidl/, with tests in //zircon/system/utest/fidl-compiler/.

Compiler back-ends

Note: The tests column refers to hand-written tests that exercise both the generated code and the runtime libraries. There are also other tests, like unit tests for the codegen itself and GIDL generated tests, refer to the tests section for details.

Supporting code for the target specific backends is located in /tools/fidl/lib/fidlgen.

Testing tools

GIDL

GIDL is a tool used to create general “write once, generate for every backend” programs. Currently GIDL is used to generate encode or decode tests (“conformance tests”) as well as benchmarks.

The actual test targets for the conformance tests in each backend are generally defined alongside the corresponding tests for that backend. Refer to the Bindings tests section for details.

Source compatibility

Source compatibility tests are used to test FIDL's source compatibility guarantees. They are found in /src/tests/fidl/source_compatibility.

Compatibility

Compatibility tests are integration tests that run FIDL clients and servers from different bindings with each other in order to test that they are compatible. Compatibility tests are found at /src/tests/fidl/compatibility/.

Dangerous identifiers

Dangerous identifier tests are found in /src/tests/fidl/dangerous_identifiers and /topaz/tests/fidl-dangerous-identifiers.

Other

Some other FIDL related areas are:

Note: The FIDL team does not necessarily own all of these areas, but they may need to be updated when making changes to the FIDL API surface, such as when changing the FIDL JSON IR. Refer to the respective READMEs and OWNERS files for details.

Other FIDL tools

A number of FIDL tools are located in the fidl-misc repo. To clone this repo, run

git clone https://fuchsia.googlesource.com/fidl-misc

It is then recommended to export the path to this directory, to make setting aliases easier:

export FIDLMISC_DIR=...

Common development tools

This is a crowdsourced section from the FIDL team on useful tools that they use for working on the FIDL codebase.

IDEs

Most of the FIDL team uses VSCode for development. Some useful plugins and workflows:

• The remote ssh feature works really well for doing remote work from your laptop.

  • Setting up tmux or screen is also helpful for remote work, to preserve history and manage multiple sessions in the shell.

• The Fuchsia documentation has instructions for setting up language servers:

  • clangd for c++
  • rust-analyzer for rust

• The rewrap extension is useful for automatically reflowing lines to a certain length (e.g. when editing markdown files).

• To get automatic syntax highlighting for the bindings golden files, update the file.associations setting:

  "files.associations": {
        "*.json.golden": "json",
        "*.rs.golden": "rust",
        "*.cc.golden": "cpp",
        "*.h.golden": "cpp",
        "*.go.golden": "go",
        "*.dart.golden": "dart",
  },
  

Commit message style guide

References to other CLs

To reference another Gerrit change in a commit message, always use the Change-ID.

Using the Change-ID is preferred since:

• The git SHA is only known after a change is merged, and while guidance could be given to use the Change-Id in one case, and the git SHA in the other, we prefer a uniform guidance.
• The link to the change is assigned by Gerrit, and is not part of the persistent history of the repository. Should we change the review mechanism, the Change-Id will continue to be part of the recorded history, whereas change's number will not. There are also rare occurrences where change numbers may be lost, e.g. due to re-indexing issues.

For instance, to refer to the change that added FTP-042 we should use I32b966810d21a249647887fa45b61720ad01714c, and not the git SHA 5d40ee8c42d1b0e4d8b690786da12a0a947c1aaa or the link to the change, https://fuchsia-review.googlesource.com/c/fuchsia/+/284569.

Multi-step change

When executing a change which requires multiple steps across various repositories, for instance to soft transition APIs defined in one repository and used in others, it is preferred to reference the last step taken, and the next step to be taken so that reviewers and those looking at the log can understand and navigate the totality of the change. When possible, it is encouraged to provide all steps to complete the migration in each commit log (but that may be impractical in some cases).

For instance:

• Align Go union unknown data APIs
• Support for flexible enums

C++ style guide

We follow the Fuchsia C++ Style Guide, with additional rules to further remove ambiguity around the application or interpretation of guidelines.

Comments

Comments must respect 80 columns line size limit, unlike code which can extend to 100 lines size limit.

Lambda captures

• If a lambda escapes the current scope, capture all variables explicitly.
• If the lambda is local (does not escape the current scope), prefer using a default capture by reference (“[&]”).

Seeing [&] is a strong signal that the lambda exists within the current scope only, and can be used to distinguish local from non-local lambdas.

// Correct.
std::set<const flat::Library*, LibraryComparator> dependencies;
auto add_dependency = [&](const flat::Library* dep_library) {
  if (!dep_library->HasAttribute("Internal")) {
    dependencies.insert(dep_library);
  }
};

General setup

Fuchsia setup

Read the Fuchsia Getting Started guide first.

fx set

If you are working on the FIDL toolchain, use:

fx set core.x64 --with //bundles/fidl:tests --with //topaz/bundles/fidl:tests

If you are working on an LSC:

fx set terminal.x64 --with //bundles:kitchen_sink \
                    --with //topaz/packages/tests:all \
                    --with //sdk:modular_testing

symbolizer

To symbolize backtraces, you'll need a symbolizer in scope:

export ASAN_SYMBOLIZER_PATH="$(find `pwd` -name llvm-symbolizer | grep clang | head -1)"

Compiling and running tests

We provide mostly one-liners to run tests for the various parts. When in doubt, refer to the “Test:” comment in the git commit message; we do our best to describe the commands used to validate our work there.

Tests are run using the fidldev tool. Examples assume that the fidldev script is somewhere on your PATH, e.g. by adding an alias:

alias fidldev=$FIDLMISC_DIR/fidldev/fidldev.py

fidlc

# optional; builds fidlc for the host with ASan <https://github.com/google/sanitizers/wiki/AddressSanitizer>
fx set core.x64 --variant=host_asan

# build fidlc
fx ninja -C $(cat .fx-build-dir) host_x64/fidlc

If you're doing extensive edit-compile-test cycles on fidlc, building with fewer optimizations can make a significant difference in the build speed. To do this, change the optimization setting in zircon/public/gn/config/levels.gni from default to debug or none.

Warning: The kernel is not regularly tested with debug, and only supports none for building. Running with none can cause kernel panics from stack overflows in the kernel.

To avoid accidentally committing this change, run:

git update-index --skip-worktree zircon/public/gn/config/levels.gni

If you want to allow the changes to be committed again, run:

git update-index --no-skip-worktree zircon/public/gn/config/levels.gni

fidlc tests

fidlc tests are at:

• //zircon/system/utest/fidl-compiler/.
• //src/lib/fidl/c/walker_tests/.
• //src/lib/fidl/c/coding_tables_tests/.
• //src/lib/fidl/c/simple_tests (C runtime tests).

To build and run fidlc tests:

fidldev test fidlc

To run a specific test case, use the --case flag with the fidlc test binary. The binary can be located by running fidldev test --dry-run --no-regen fidlc.

$FUCHSIA_DIR/out/default/host_x64/fidl-compiler --gtest_filter 'EnumsTests.*'

To easily run tests in a debug build:

fx set core.x64 --variant=host_asan --with //bundles/fidl:tests
export ASAN_SYMBOLIZER_PATH="$(find `pwd` -name llvm-symbolizer | grep clang | head -1)"
fx ninja -C out/default host_x64-asan/exe.unstripped/fidl-compiler && \
    ./out/default/host_x64-asan/exe.unstripped/fidl-compiler --gtest_filter 'EnumsTests.*'

To regenerate the fidlc JSON goldens:

fidldev regen fidlc

These “golden” files are examples of what kind of JSON IR fidlc produces and are used to track changes. It is required to regenerate the golden files each time the JSON IR is changed in any way, otherwise the json_generator_tests fails.

fidlgen (LLCPP, HLCPP, Rust, Go)

Build:

fx build tools/fidl

Run:

$FUCHSIA_DIR/out/default/host_x64/fidlgen_{llcpp, hlcpp, rust, go}

Some example tests you can run:

fx test fidlgen_hlcpp_golden_tests
fx test fidlgen_golang_lib_tests
fidldev test --no-regen fidlgen

To regenerate the goldens:

fidldev regen fidlgen

fidlgen_dart

Build:

fx ninja -C out/default host_x64/fidlgen_dart

Run:

$FUCHSIA_DIR/out/default/host_x64/fidlgen_dart

Some example tests you can run:

fidldev test --no-regen fidlgen_dart

To regenerate the goldens:

fidldev regen fidlgen_dart

Bindings

fidldev supports tests for each of the bindings. Some of the bindings tests run on device and require having Fuchsia running in an emulator. Here are the steps:

Tab 1> fx build && fx serve-updates

Tab 2> fx qemu -kN

The -k flag enables KVM. It is not required, but the emulator is much slower without it. The -N flag enables networking.

The bindings tests can then be run with fidldev:

fidldev test --no-regen hlcpp
fidldev test --no-regen llcpp
fidldev test --no-regen c
fidldev test --no-regen go
fidldev test --no-regen rust
fidldev test --no-regen dart

Alternatively, run fidldev with no arguments to test files that have changed:

fidldev test

To run a specific test or to pass flags to a specific test, run fidldev with the --dry-run, --no-build, --no-regen flags to obtain the desired test commands.

Compatibility test

Details about how the compatibility tests work and where the code is located can be found in the README at //src/tests/fidl/compatibility.

To run the compatibility tests, you first need to have Fuchsia running in an emulator:

Tab 1> fx build && fx serve

Tab 2> fx qemu -kN

To run the compatibility tests:

Tab 3> fx set core.x64 --with //src/tests/fidl/compatibility
Tab 3> fx test fidl-compatibility-test

GIDL

To rebuild GIDL:

fx build host-tools/gidl

Measure tape

fx set core.x64 --with //tools/fidl/measure-tape/src:host
fx build

All tests

Bindings tests

On device tests generally have greater coverage than host tests, due to support for only running a subset of features on host. However, host tests can be useful for debugging issues that prevent boot of the device.

On device

Note: While fx test fidl_bindings_test prints test names as they run, it does not show stack traces for test failures. To see those, look at the fx qemu or fx log output.

Host

Fidlgen tests

Other

All benchmarks

Benchmarks can either be run directly or through one of two test runners: fuchsia_benchmarks (old), SL4F (new).

Benchmarks on chromeperf are currently generated through the fuchsia_benchmarks runner but are transitioning to SL4F. During this transition, benchmarks should be integrated in both systems.

Directly running benchmarks

Ensure that the benchmarks are included in your build:

fx set core.x64 --with //src/tests/benchmarks

You will need to fx build and restart qemu for the packages to be available.

Available benchmarks:

Running all benchmarks with SL4F benchmark runner

This runs benchmarks the same way they are run on CQ. SL4F requires the terminal.x64 product. Use fx set to switch products:

fx set terminal.x64

To run all FIDL tests, use:

fx test --e2e fidl_microbenchmarks_test

All regen commands

Compiling with ninja

In some cases, GN can build many unneeded targets. You can build a specific target with ninja instead of GN. In most cases, you can grep for the binary name to determine the ninja invocation.

For example, you can grep for fidlgen_dart:

fx ninja -C out/default -t targets all | grep -e 'fidlgen_dart:'

This example outputs a list of ninja targets which includes host_x64/fidlgen_dart. Therefore, to build fidlgen_dart run the following ninja command:

fx ninja -C out/default host_x64/fidlgen_dart

Debugging (host)

There are several ways of debugging issues in host binaries. This section gives instructions for the example case where fidlc --files test.fidl is crashing:

• GDB
• Asan
• Valgrind

Note: Even with all optimizations turned off, the binaries in out/default/host_x64 are stripped. For debugging, you should use the binaries with the .debug suffix, such as out/default.zircon/host-x64-linux-clang/obj/tools/fidl/fidlc.debug.

GDB

Start GDB:

gdb --args out/default.zircon/host-x64-linux-clang/obj/tools/fidl/fidlc.debug --files test.fidl

Then, enter “r” to start the program.

ASan

Ensure you are compiling with ASan enabled:

fx set core.x64 --variant=host_asan
fx build host_x64/fidlc

Then run out/default/host_x64/fidlc --files test.fidl. That binary should be the same as out/default.zircon/host-x64-linux-asan/obj/tools/fidl/fidlc.

Valgrind

On Google Linux machines, you may need to install a standard version of Valgrind instead of using the pre-installed binary:

sudo apt-get install valgrind

Then:

valgrind -v -- out/default.zircon/host-x64-linux-clang/obj/tools/fidl/fidlc.debug --files test.fidl

Workflows

Go fuchsia.io and fuchsia.net

To update all the saved fidlgen files, run the following command, which automatically searches for and generates the necessary go files:

fx exec third_party/go/regen-fidl

FAQs

Why is the C back-end different than all other back-ends?

TBD

Why is fidlc in the zircon repo?

TBD

Why aren't all back-ends in one tool?

We'd actually like all back-ends to be in separate tools!

Down the road, we plan to have a script over all the various tools (fidlc, fidlfmt, the various back-ends) to make all things accessible easily, and manage the chaining of these things. For instance, it should be possible to generate Go bindings in one command such as:

fidl gen --library my_library.fidl --binding go --out-dir go/src/my/library

Or format a library in place with:

fidl fmt --library my_library.fidl -i