Integrate Rust -> BSW logger bridge by felixm-acc · Pull Request #480 · eclipse-openbsw/openbsw

felixm-acc · 2026-06-12T06:30:11Z

This PR improves Rust support and offline/containerized build reliability:

Rust logging bridge: adds a bridge so Rust code can log through the existing C++ logger.
Offline builds: vendors first external Rust dependencies so builds work without network access.
Toolchain: bumps the default Rust version to 1.96.0.
Container UID handling: makes the cargo and rustup homes accessible to the runtime UID injected into
the dev container.
Repo hygiene: adds a .gitattributes file (for exclusion from vendored 3rdparty code from SLOC stats on main page and Github diff view)

ThoFrankACN · 2026-06-12T06:35:20Z

 2. The Rust compiler

-Install Rust 1.90.0 to be compatible with the CI builds:
+Install Rust 1.96.0 to be compatible with the CI builds:


The ubuntu version is defined only once. Probably we should use the same approach for rust:

:prop:`tool:ubuntu_version`

done. There's even a nice mechanism already in place in doc/dev/conf.py to replace x.x in code blocks with the correct version number depending on which marker string is nearby.

ThoFrankACN

Really nice work!

I only have some minor feedback. In my opinion the vendoring approach is good. We have everything available for an offline build and can track changes, but using dependencies from crates.io is still simple.

ThoFrankACN · 2026-06-12T06:36:32Z

+    if it is unset the message is dropped. The crate name is prepended to the
+    message so the originating module is still visible.
+
+.. warning::


Should we also warn about the performance implications of double-buffering here?

And not sure how cpp behaves, but the formatting overhead for rust messages is still there even if the cpp logger component is disabled.

Good points, I'll add the two caveats.
I'll also check how it's done in C++, maybe some performance gains hidden there by disabling serialization if it's not going to get printed anyway.

In the bsw logger it's gated through if (isEnabled(componentIndex, _level)) in Logger.h. I can also (double) check isEnabled via FFI from Rust before doing the serialization. Maybe with some profiling of some format heavy logs to see whether that optimization is worth it.

Test log line with some formatting.

"rx id={:#05x} dlc={} ts={:>8}ms seq={:05} crc={:#06x}"
-> "rx id=0x1a4 dlc=8 ts= 123456ms seq=00042 crc=0xbeef"

Results (200k iterations/run, 3 runs)

gated naive (always format) delta

disabled component ~6.5 ns/call ~160 ns/call saving ~155 ns (~24x)

enabled component ~160 ns/call ~160 ns/call overhead ~ 0

definitely seems to be worth it to add the additional is_enabled FFI call also to Rust to skip serializing entirely when the logging component is disabled.

ThoFrankACN · 2026-06-12T06:37:04Z

+            for &b in s {
+                self.buffer[self.bytes_used] = b;
+                self.bytes_used += 1;
+            }
+            Ok(())
+        } else {
+            // not enough space
+            // only copy what fits
+            let s = &s[0..(LOG_BUF_SIZE - self.bytes_used - 1)];
+            for &b in s {
+                self.buffer[self.bytes_used] = b;
+                self.bytes_used += 1;
+            }


Maybe the loops should be replaced by copy_from_slice.

ThoFrankACN · 2026-06-12T06:37:21Z

+pub use component::LoggerComponent;
+
+#[doc(hidden)]
+pub use paste as __paste;


Why is there the need to re-export paste?

It's used in the expansion of the log macro. If we don't re-export then it's not available at the callsite of the macro and every caller of our logging macros would need to add a dependency to paste in their Cargo.toml. Similar to https://stackoverflow.com/questions/65164156/how-to-set-dependencies-for-a-macro-in-rust#:~:text=lib%2Ers%2E-,Library,attribute

/opt/cargo and /opt/rustup are filled as root during the image build (rustup setup and cargo install cbindgen), but the container runs under the UID injected via DOCKER_UID, which then cannot read or lock the cargo registry. This caused the following CI error: error: failed to open `/opt/cargo/registry/cache/.../equivalent-1.0.2.crate` Caused by: Permission denied (os error 13) chmod the cargo and rustup homes world-accessible after install, the same fix already applied to /home/build/.cache. Change-Id: I6cd61820097a6f56d030c0972ca25b0e2dd88320

Introduce the openbsw-logger crate (libs/bsw/logger/rust) so Rust code can emit log messages through the existing C++ ::util::logger::Logger. First-party Rust code uses per-component macros: declare_logger_component! binds a component to the extern "C" getter that DECLARE/DEFINE_LOGGER_COMPONENT now emit on the C++ side, and bsw_debug!/info!/warn!/error!/critical! format into a stack buffer and forward to bsw_cpp_logger_log. The component index is fetched from C++ on every call rather than cached, so there is no Rust-side init step and no dependency on logger::init() ordering (the symbols hold COMPONENT_NONE until applyMapping() runs). Messages routed through the `log` facade (e.g. third-party crates) fall back to a single component set via set_default_component, with the crate name prepended; they are dropped if no default is configured. bsw_cpp_logger_log passes the already-formatted message through a literal "%s" so stray '%' characters are not interpreted as printf specifiers against an empty va_list. Trace maps to DEBUG, since a trace level does not exist in the BSW logger. Use in referenceApp: rustHelloWorld logs under the DEMO component and app.cpp installs the logger and default component under BUILD_RUST. Document the Rust API in the logger module docs. Change-Id: Iab2756f03f322b5e5354c3ee94e41c8f4f00eac9

This will exclude 3rdparty code from the GitHub language stats bars. Change-Id: Ied6c8add46a2a7ec819ed4762c3966242a9593dd

Vendor the workspace's crate dependencies into libs/3rdparty/cargo-vendor alongside the other third-party imports. Remove cbindgen from the build so the vendored tree stays small. cbindgen only generates the logger's C header (include/BswLogger.h, checked in), and made cargo vendor 88 MB large across 71 crates by dragging in unneeded windows deps. Dropping it leaves just log and paste (576 KB, 2 crates). The header is now a checked-in artifact regenerated outside of the build with the cbindgen CLI. .ci/cbindgen.py regenerates it and fails on drift via git diff --exit-code, mirroring .ci/format.py, and a matching cbindgen.yml workflow runs it in the development container. Bump the cbindgen CLI already installed in that image from 0.27.0 to 0.29.2 so it matches the version that produced the committed header. Remove the logger crate's build.rs, which existed only to run cbindgen Change-Id: I2de261e9b4e4e5cc1c3504051874578b701087f7

Also adapt the .treefmt.toml script and Dockerfile because on the CI rustfmt wasn't available for the new Rust version. Change-Id: Ie7b52473792e1d5f0a74dd595952d568368c8394

ThoFrankACN

LGTM

ThoFrankACN reviewed Jun 12, 2026

View reviewed changes

felixm-acc force-pushed the cr-624756 branch from 23f07e1 to 228aed5 Compare June 18, 2026 09:53

felixm-acc added 3 commits June 18, 2026 11:57

Add a .gitattributes file

3d57556

This will exclude 3rdparty code from the GitHub language stats bars. Change-Id: Ied6c8add46a2a7ec819ed4762c3966242a9593dd

felixm-acc force-pushed the cr-624756 branch from 228aed5 to 1cc21d3 Compare June 18, 2026 11:58

felixm-acc changed the title ~~bump default Rust version to 1.96.0~~ Integrate Rust -> BSW logger bridge Jun 18, 2026

felixm-acc added 2 commits June 18, 2026 16:58

bump default Rust version to 1.96.0

592c285

Also adapt the .treefmt.toml script and Dockerfile because on the CI rustfmt wasn't available for the new Rust version. Change-Id: Ie7b52473792e1d5f0a74dd595952d568368c8394

felixm-acc force-pushed the cr-624756 branch from 1cc21d3 to 592c285 Compare June 18, 2026 14:58

ThoFrankACN approved these changes Jun 18, 2026

View reviewed changes

christian-schilling approved these changes Jun 30, 2026

View reviewed changes

rolandreichweinbmw merged commit 8bcfb3d into eclipse-openbsw:main Jun 30, 2026
131 checks passed

	gated	naive (always format)	delta
disabled component	~6.5 ns/call	~160 ns/call	saving ~155 ns (~24x)
enabled component	~160 ns/call	~160 ns/call	overhead ~ 0

Uh oh!

Conversation

felixm-acc commented Jun 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ThoFrankACN Jun 12, 2026

Choose a reason for hiding this comment

Uh oh!

felixm-acc Jun 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ThoFrankACN left a comment

Choose a reason for hiding this comment

Uh oh!

ThoFrankACN Jun 12, 2026

Choose a reason for hiding this comment

Uh oh!

felixm-acc Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

felixm-acc Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

felixm-acc Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ThoFrankACN Jun 12, 2026

Choose a reason for hiding this comment

Uh oh!

felixm-acc Jun 18, 2026

Choose a reason for hiding this comment

Uh oh!

ThoFrankACN Jun 12, 2026

Choose a reason for hiding this comment

Uh oh!

felixm-acc Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ThoFrankACN left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

felixm-acc commented Jun 12, 2026 •

edited

Loading

felixm-acc Jun 18, 2026 •

edited

Loading

felixm-acc Jun 17, 2026 •

edited

Loading

felixm-acc Jun 17, 2026 •

edited

Loading

felixm-acc Jun 17, 2026 •

edited

Loading

felixm-acc Jun 17, 2026 •

edited

Loading