In the fetch and checkout build modes, hook/build.dart provides:
- A dynamic library with all of BoringSSL when linking is disabled (
dart run,dart test, and Flutter debug builds). - A static library when linking is enabled (
dart build, and Flutter profile and release builds), whichhook/link.dartlinks into a tree-shaken dynamic library containing only the functions recorded via@RecordUse().
Every GitHub Release contains both the dynamic and the static library for each
prebuilt target in lib/src/hook_helpers/targets.dart.
The package:boring binaries
workflow builds the prebuilt libraries (tool/precompile_binaries.dart) from
the vendored BoringSSL whenever a change to the native sources (src/,
third_party/boringssl/, or tool/REVISION) lands on main, attaches them to
a boring-binaries-<date>-<commit> pre-release with a provenance attestation
(gh attestation verify <file> --repo dart-lang/tools), and opens a pull
request (tool/regenerate_hashes.dart) that pins their SHA-256 hashes in
lib/src/hook_helpers/hashes.dart.
CI (tool/check_prebuilt_binaries.sh) checks that a release of this package
pins libraries built from the sources it vendors, so fetch and checkout
produce the same BoringSSL.
Run ./tool/bump-boringssl-revision.sh [revision] to update the vendored
BoringSSL under third_party/boringssl/, record the commit in tool/REVISION,
and regenerate the FFI bindings (tool/ffigen.dart). The generated job in
boring.yaml runs on ubuntu-24.04 and
verifies that all generated files and hashes.dart
are up to date.
The refresh job in the same workflow runs the script weekly against the
latest BoringSSL revision and opens a pull request with the result.
test/conformance/x509_limbo_test.dart
runs the x509-limbo path validation suite against the
vendored BoringSSL, and
x509_limbo_expected_failures.txt
records the testcases BoringSSL is known to disagree with. The x509-limbo job
fails on a divergence that the list does not record, and on a listed testcase
that started agreeing, so the list cannot go stale.
BoringSSL's X.509 behavior changes between revisions. After updating it, run
X509_LIMBO_REGENERATE=1 ./tool/run_x509_limbo_tests.shto regenerate the list, and read the diff to see the change in behavior. The
refresh job does this for its pull requests, so a change to the list there is
part of what to review.