fix(diagnostics): bound formatting and release ingress before callbacks

Bound UTF-8 formatting and release diagnostic ingress before invoking
callbacks. Check error-ID collisions at Rust/C++ release final link.

Validation: diagnostic tests, strict Clippy, Miri and final-link cases.

docs/src/errors.md

@@ -23,8 +23,15 @@ requires a nonzero explicit ID on every variant and retains no type, variant,
 or module string. `Display` renders the decimal ID. Renaming, moving, or
 reordering a Rust variant therefore does not change its receiver contract.
 Missing and same-enum duplicate IDs are rejected by the derive. Reusing an ID
-across error enums generates the same reserved linker symbol and fails the
+across error enums generates the same reserved symbol. Collision detection
-final firmware build.
+requires both defining objects to participate in compilation or final linking.
+It does not inspect unextracted archive members or prove catalog-wide uniqueness.
+Stable ID allocation still requires review.
+
+The [final-link fixture](../../firmware/libs/kaptan-err-reportable-macros/tests/final_link.md)
+checks Rust 1.90.0 release fat-LTO and native Linux static-library links.
+Run it with `make -C firmware native-test`. Each run retains its results and
+Rust compiler identity in `results.txt`.
 
 `#[kaptan_err_reportable::propagate]` turns an annotated function into a
 reporting point. If a reportable `Err` leaves that function, the attribute makes
@@ -76,16 +83,14 @@ different reportable error is already pending.
   `ERROR`, subject to bounded transport admission. The payload is
   `{ error_id, raised_at[96] }`. `raised_at` is a zero-padded UTF-8
   `file:line` prefix for the reporting point.
-  The message kind itself identifies an error; no level is duplicated in the
+  Dynamic human-readable context belongs in `LOG`, while
-  payload. Dynamic human-readable context belongs in `LOG`, while
   structured protocol failures remain in their owning response message.
 - Control, navigation, and safety error types are derive-ready but
   their live owners are not yet attached to a MAVLink worker.
 
-F722 still uses `panic-halt`; it does not claim a live panic frame or
+F722 uses `panic-halt`, without live panic reporting or retained-record replay.
-retained-record replay.
 
 No diagnostic message may synchronously transmit from a panic handler, hard
 fault, ISR, DMA callback, or motor-inhibit path. A later target integration
 must inhibit outputs first, persist a validated crash record, then report it
-only after the next boot’s static board telemetry is available.
+only after MAVLink is available on the next boot.

docs/src/logs.md

@@ -1,6 +1,6 @@
 # Logs
 
-Kaptan code uses the familiar logging macros on both host and flight targets:
+Logging macros work on host and flight targets:
 
 ```rust,ignore
 kaptan_log::debug!("estimator waiting for IMU");
@@ -11,16 +11,22 @@ kaptan_log::trace!("prediction step complete");
 ```
 
 All five macros emit `LOG`; their level never selects a different wire
-message. When a reportable firmware error escapes a reviewed
+message. When a reportable firmware error escapes a
 `#[kaptan_err_reportable::propagate]` function, it makes one bounded admission attempt
 to the separate `ERROR` ingress.
 
-`kaptan-log` formats each event once into a fixed bounded buffer and never
+`kaptan-log` formats each event once into fixed 160-byte UTF-8 storage. Its core
-sends MAVLink, allocates, blocks, changes flight state, or touches DMA. The
+storage and admission paths use no allocation or blocking, and do not send
-composition fans the event into each enabled sink; the sole MAVLink worker
+MAVLink, change flight state or touch DMA. Formatting invokes caller-supplied
-drains its bounded sink queue. Ordinary log text uses one 160-byte bounded
+`Display` implementations, which may allocate, block, ignore `fmt::Error` or do
-record. Reportable-error records contain the explicit ID and a static reference
+unbounded work. Overflow stops formatters that propagate the error and latches
-to the reporting point until the MAVLink adapter copies its
+truncation even if a formatter ignores it. Bounded storage is not a universal
+worst-case execution-time guarantee. Producers must choose formatters suitable
+for their execution context.
+
+The composition fans the event into each enabled sink. The MAVLink worker drains
+its bounded sink queue. Reportable-error records contain the explicit ID and a
+static reference to the reporting point until the MAVLink adapter copies its
 `file:line` into the 96-byte `raised_at` field.
 Logs and reportable errors use independent fixed atomic
 ingresses; interrupt and worker producers make one nonblocking admission
@@ -47,8 +53,6 @@ or the complete field when no zero is present.
 (2), Info (3), Warn (4), or Error (5). Off (0) is a filter-only sentinel and
 is never emitted as a `LOG` event. `LOG.flags` currently owns only the
 Truncated bit (1).
-The `ERROR` message kind already means a reportable firmware error, so it has
-no redundant level or flags field.
 
 `error_id` is the stable nonzero catalog number explicitly assigned to the
 Rust error variant. Renaming or reordering a variant does not change it.
@@ -73,8 +77,8 @@ runtime log levels. F722 uses the validated filter recovered at boot, falling
 back to its board factory when the journal is empty or invalid.
 
 Events retain the compact typed `Level` and sink configuration retains the
-compact typed `LevelFilter`. Generic schema discovery exposes the filter's
+compact typed `LevelFilter`. Schema metadata supplies the filter choices to the
-named choices to the configurator, so the UI carries no hardcoded field list.
+configurator.
 Only the MAVLink adapter converts the five event levels to the typed Kaptan
 wire enum.
 

firmware/Makefile

@@ -1,5 +1,6 @@
 override SHELL := /bin/sh
 override MAKEFLAGS += --no-print-directory
+export CARGO_BUILD_JOBS ?= 2
 
 .DEFAULT_GOAL := help
 .DELETE_ON_ERROR:
@@ -29,7 +30,7 @@ ifeq ($(PRIMARY_GOAL),run)
 
 else
 
-.PHONY: help run fmt clippy test docs pre-commit check clean
+.PHONY: help run fmt clippy test native-test docs pre-commit check clean
 
 help:
 	@printf '%s\n' 'Usage: make firmware -- <cargo arguments...>' '' \
@@ -38,6 +39,7 @@ help:
 		'  fmt     Check Rust formatting' \
 		'  clippy  Run host and embedded Clippy checks' \
 		'  test    Run Rust and firmware script tests' \
+		'  native-test  Check cross-crate error IDs at Rust/C++ release final link' \
 		'  docs    Build Rust API documentation with warnings denied' \
 		'  pre-commit  Run fmt, clippy, test, and docs' \
 		'  check   Alias for pre-commit' \
@@ -58,6 +60,11 @@ test:
 	$(CARGO) test --workspace $(CARGO_FLAGS)
 	$(PYTHON) -m unittest \
 		scripts/test-analyze-fw.py scripts/test-summarize-test-coverage.py
+	+@$(MAKE) native-test
+
+native-test:
+	python3 libs/kaptan-err-reportable-macros/tests/final_link.py \
+		--work-dir "$(or $(CARGO_TARGET_DIR),target)/native-tests/final-link"
 
 docs:
 	RUSTDOCFLAGS='-D warnings' $(CARGO) doc --workspace --no-deps \

firmware/libs/kaptan-err-reportable-macros/src/lib.rs

@@ -1,4 +1,4 @@
-//! Attribute expansion for `kaptan-err-reportable`.
+//! Error derive and reporting-attribute expansion for `kaptan-err-reportable`.
 
 use proc_macro::TokenStream;
 use quote::{format_ident, quote};
@@ -12,8 +12,12 @@ use syn::{
 /// [`core::error::Error`] for an explicitly numbered error enum.
 ///
 /// Every variant must declare a stable nonzero
-/// `#[kaptan_error(id = ...)]`. Duplicate IDs fail compilation within an enum
+/// `#[kaptan_error(id = ...)]`. The derive rejects duplicates within an enum.
-/// or crate and fail final linking when they originate in different crates.
+/// Across enums, duplicate IDs emit the same reserved symbol. Collision detection
+/// requires both defining objects to participate in compilation or final linking.
+/// Unextracted archive members are not checked, so catalog allocation still needs
+/// review. The fixture guide in `tests/final_link.md` describes Rust release-LTO
+/// and native Linux checks. Embedded linker combinations need their own checks.
 #[proc_macro_derive(KaptanError, attributes(kaptan_error))]
 pub fn derive_kaptan_error(input: TokenStream) -> TokenStream {
     let input = parse_macro_input!(input as DeriveInput);

firmware/libs/kaptan-err-reportable-macros/tests/final_link.md

@@ -0,0 +1,28 @@
+# Cross-crate error-ID final-link checks
+
+From the repository root, run `nix develop -c make -C firmware native-test`.
+For a separate artifact directory:
+
+```sh
+nix develop -c env CARGO_BUILD_JOBS=2 python3 firmware/libs/kaptan-err-reportable-macros/tests/final_link.py \
+  --work-dir /tmp/kaptan-final-link
+```
+
+The fixture builds two crates using the production `KaptanError` derive as
+rlibs and staticlibs. Rust and C++ executables cover unique and duplicate IDs,
+used and unused error values, and an unused second crate. Successful executables
+must run. Expected duplicate failures must name `__kaptan_error_id_65401`.
+
+The helper seeds its fixture lock from `firmware/Cargo.lock`, verifies registry
+versions, sources and checksums, then builds with Rust 1.90.0 and `--locked
+--offline`. It retains source, manifests, locks, binaries, linker output and
+`results.txt` with the Rust compiler identity and all twelve outcomes. Native linker flags
+are Linux-specific.
+
+The exported `#[used]` marker detects collisions when both defining objects
+participate in the link. It cannot force extraction of an unreferenced staticlib
+member. The native case that calls only the first crate can therefore link despite
+the duplicate ID. Rust's release-LTO fixture explicitly names the second crate
+and rejects that collision. These cases do not establish catalog-wide uniqueness
+across every supplied archive or every embedded linker/LTO combination. Stable
+error-ID allocation still requires review.

firmware/libs/kaptan-err-reportable-macros/tests/final_link.py

@@ -0,0 +1,229 @@
+#!/usr/bin/env python3
+"""Exercise the actual derive across crates at the host release-LTO final link.
+
+Run with Python 3.11+. Build output is retained under the supplied directory.
+This is a link experiment, not a trybuild or cargo-check substitute.
+"""
+
+import argparse
+import hashlib
+import json
+from pathlib import Path
+import subprocess
+import tomllib
+
+
+def run(arguments, cwd):
+    return subprocess.run(arguments, cwd=cwd, text=True, capture_output=True)
+
+
+def require_success(result):
+    if result.returncode:
+        raise RuntimeError(result.stdout + result.stderr)
+
+
+def registry_packages(lock):
+    return {
+        (package["source"], package["name"], package["version"], package["checksum"])
+        for package in tomllib.loads(lock.decode())["package"]
+        if package.get("source", "").startswith("registry+")
+    }
+
+
+def verify_registry_packages(firmware_lock, fixture_lock):
+    packages = registry_packages(fixture_lock)
+    drift = packages - registry_packages(firmware_lock)
+    if drift:
+        raise RuntimeError(
+            f"fixture registry dependencies differ from firmware lock: {sorted(drift)!r}"
+        )
+    return sorted(packages)
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--work-dir", type=Path, required=True)
+    args = parser.parse_args()
+    work = args.work_dir.resolve()
+    work.mkdir(parents=True, exist_ok=True)
+    runtime = Path(__file__).resolve().parents[2] / "kaptan-err-reportable"
+    version = run(["rustc", "+1.90.0", "-vV"], work)
+    require_success(version)
+    host = next(
+        line.removeprefix("host: ")
+        for line in version.stdout.splitlines()
+        if line.startswith("host: ")
+    )
+    print(version.stdout, end="")
+    for name in ["a", "b", "app"]:
+        (work / name / "src").mkdir(parents=True, exist_ok=True)
+    (work / "Cargo.toml").write_text("""
+[workspace]
+members = ["a", "b", "app"]
+resolver = "2"
+[profile.release]
+opt-level = 3
+codegen-units = 1
+lto = "fat"
+panic = "abort"
+strip = "symbols"
+""")
+    for name in ["a", "b"]:
+        (work / name / "Cargo.toml").write_text(f"""
+[package]
+name = "error-id-{name}"
+version = "0.0.0"
+edition = "2024"
+[lib]
+crate-type = ["rlib", "staticlib"]
+[features]
+duplicate = []
+unused = []
+[dependencies]
+kaptan-err-reportable = {{ path = {json.dumps(str(runtime))} }}
+""")
+        unique = 65401 if name == "a" else 65402
+        (work / name / "src/lib.rs").write_text(f"""
+#[derive(Debug, kaptan_err_reportable::KaptanError)]
+pub enum Error {{
+    #[cfg_attr(feature = "duplicate", kaptan_error(id = 65401))]
+    #[cfg_attr(not(feature = "duplicate"), kaptan_error(id = {unique}))]
+    Fault,
+}}
+
+#[unsafe(no_mangle)]
+pub extern "C" fn fixture_{name}() -> u16 {{
+    #[cfg(not(feature = "unused"))]
+    {{ kaptan_err_reportable::ReportableError::error_id(&Error::Fault).get() }}
+    #[cfg(feature = "unused")]
+    {{ {unique} }}
+}}
+""")
+    (work / "app/Cargo.toml").write_text("""
+[package]
+name = "error-id-app"
+version = "0.0.0"
+edition = "2024"
+[dependencies]
+error-id-a = { path = "../a" }
+error-id-b = { path = "../b" }
+""")
+    (work / "app/src/main.rs").write_text("fn main() {}\n")
+    firmware_lock = (runtime.parents[1] / "Cargo.lock").read_bytes()
+    (work / "firmware.Cargo.lock").write_bytes(firmware_lock)
+    fixture_lock_path = work / "Cargo.lock"
+    fixture_lock_path.write_bytes(firmware_lock)
+    # Resolve the fixture's local workspace entries while retaining locked
+    # registry versions. generate-lockfile would instead select newer versions.
+    require_success(
+        run(
+            [
+                "cargo",
+                "+1.90.0",
+                "metadata",
+                "--offline",
+                "--all-features",
+                "--format-version",
+                "1",
+            ],
+            work,
+        )
+    )
+    fixture_lock = fixture_lock_path.read_bytes()
+    packages = verify_registry_packages(firmware_lock, fixture_lock)
+    provenance = [
+        f"firmware Cargo.lock SHA-256: {hashlib.sha256(firmware_lock).hexdigest()}",
+        f"fixture Cargo.lock SHA-256: {hashlib.sha256(fixture_lock).hexdigest()}",
+        "Fixture registry dependencies verified as an exact subset of the firmware lock:",
+        *[
+            f"  {name} {version} source={source} checksum={checksum}"
+            for source, name, version, checksum in packages
+        ],
+    ]
+    print("\n".join(provenance))
+    results = []
+    for mode in ["used", "unused", "unextracted"]:
+        use_b = mode != "unextracted"
+        rust_call = (
+            "assert!(std::hint::black_box(error_id_b::fixture_b()) >= 65401);"
+            if use_b
+            else ""
+        )
+        native_call = "|| fixture_b() < 65401" if use_b else ""
+        (work / "app/src/main.rs").write_text(f"""
+extern crate error_id_b;
+fn main() {{
+    assert_eq!(std::hint::black_box(error_id_a::fixture_a()), 65401);
+    {rust_call}
+}}
+""")
+        (work / "main.cpp").write_text(f"""
+#include <cstdint>
+extern "C" uint16_t fixture_a();
+extern "C" uint16_t fixture_b();
+int main() {{ return fixture_a() != 65401 {native_call}; }}
+""")
+        for duplicate in [False, True]:
+            label = f"{mode}-{'duplicate' if duplicate else 'unique'}"
+            features = ["error-id-b/duplicate"] if duplicate else []
+            if mode != "used":
+                features += ["error-id-a/unused", "error-id-b/unused"]
+            target = work / "target" / label
+            cargo = [
+                "cargo",
+                "+1.90.0",
+                "build",
+                "--locked",
+                "--offline",
+                "--release",
+                "--target",
+                host,
+                "--target-dir",
+                str(target),
+            ]
+            if features:
+                cargo += ["--features", ",".join(features)]
+            # Each archive must build independently before linking them together.
+            archives = run(cargo + ["-p", "error-id-a", "-p", "error-id-b"], work)
+            require_success(archives)
+            release = target / host / "release"
+            native = run(
+                [
+                    "c++",
+                    str(work / "main.cpp"),
+                    str(release / "liberror_id_a.a"),
+                    str(release / "liberror_id_b.a"),
+                    "-ldl",
+                    "-lpthread",
+                    "-lm",
+                    "-o",
+                    str(release / "native-app"),
+                ],
+                work,
+            )
+            rust = run(cargo + ["-p", "error-id-app"], work)
+            for kind, result, binary in [
+                ("staticlib", native, release / "native-app"),
+                ("rlib", rust, release / "error-id-app"),
+            ]:
+                (work / f"{label}-{kind}.log").write_text(result.stdout + result.stderr)
+                if not duplicate:
+                    require_success(result)
+                elif (use_b or kind == "rlib") and result.returncode == 0:
+                    raise RuntimeError(f"{label} {kind}: duplicate unexpectedly linked")
+                if result.returncode == 0:
+                    require_success(run([str(binary)], work))
+                    outcome = "linked and ran"
+                else:
+                    if "__kaptan_error_id_65401" not in result.stderr:
+                        raise RuntimeError(result.stderr)
+                    outcome = "duplicate ID rejected"
+                results.append(f"{label} {kind}: {outcome}")
+    print("\n".join(results))
+    (work / "results.txt").write_text(
+        version.stdout + "\n".join(provenance + results) + "\n"
+    )
+
+
+if __name__ == "__main__":
+    main()

firmware/libs/kaptan-err-reportable/src/ingress.rs

@@ -78,6 +78,54 @@ mod tests {
     use super::ErrorIngress;
     use crate::ErrorId;
 
+    #[test]
+    fn concurrent_publication_and_take_preserve_complete_unique_records() {
+        let ingress = ErrorIngress::new();
+        let start = std::sync::Barrier::new(3);
+        let mut records = std::thread::scope(|scope| {
+            let ingress = &ingress;
+            let start = &start;
+            let producers: [_; 2] = core::array::from_fn(|producer| {
+                scope.spawn(move || {
+                    start.wait();
+                    for sequence in 0..8 {
+                        ingress.publish(
+                            ErrorId::new(1000 + (producer * 8 + sequence) as u16),
+                            ["left", "right"][producer],
+                        );
+                        std::thread::yield_now();
+                    }
+                })
+            });
+            let consumer = scope.spawn(move || {
+                start.wait();
+                let mut records = std::vec::Vec::new();
+                for _ in 0..64 {
+                    records.extend(ingress.take());
+                    std::thread::yield_now();
+                }
+                records
+            });
+            for producer in producers {
+                producer.join().unwrap();
+            }
+            consumer.join().unwrap()
+        });
+        records.extend(ingress.take());
+        assert!(ingress.take().is_none());
+        assert!(!records.is_empty());
+
+        // Contention may discard publications. Every retained record must be
+        // one complete publication, and it may be consumed only once.
+        let mut seen = [false; 16];
+        for record in records {
+            let index = usize::from(record.id().checked_sub(1000).unwrap());
+            assert!(index < seen.len());
+            assert_eq!(record.raised_at(), ["left", "right"][index / 8]);
+            assert!(!core::mem::replace(&mut seen[index], true));
+        }
+    }
+
     #[test]
     fn retains_the_first_id_and_location_until_consumed() {
         let ingress = ErrorIngress::new();

firmware/libs/kaptan-err-reportable/src/lib.rs

@@ -1,8 +1,5 @@
 //! Stable reportable firmware errors with bounded diagnostic ingress.
 //!
-//! This crate owns error identity and ingress. It contains no logging levels,
-//! filtering, MAVLink, board, filesystem, or clock dependency.
-//!
 //! `#[propagate]` turns an annotated function into a reporting point. When a
 //! reportable error leaves that function, the macro makes one reporting
 //! attempt. Nested annotated callers would therefore attempt to report the

firmware/libs/kaptan-log/src/facade.rs

@@ -71,18 +71,18 @@ impl<const N: usize> LogIngress<N> {
     }
 
     fn drain_one<R>(&self, consume: impl FnOnce(&LogRecord) -> R) -> Option<R> {
-        self.with_state(|entries, head, len| {
+        let record = self
-            if *len == 0 {
+            .with_state(|entries, head, len| {
-                return None;
+                if *len == 0 {
-            }
+                    return None;
-            let record = entries[*head].as_ref()?;
+                }
-            let result = consume(record);
+                let record = entries[*head].take();
-            entries[*head] = None;
+                *head = (*head + 1) % N;
-            *head = (*head + 1) % N;
+                *len -= 1;
-            *len -= 1;
+                record
-            Some(result)
+            })
-        })
+            .flatten()?;
-        .flatten()
+        Some(consume(&record))
     }
 
     fn with_state<R>(
@@ -116,8 +116,9 @@ pub fn publish(level: Level, arguments: fmt::Arguments<'_>) {
 
 /// Visits and removes the oldest pending log event.
 ///
-/// The callback runs during the ingress's one nonblocking ownership attempt;
+/// The record is removed and exclusive ingress access is released before the
-/// it should perform only bounded composition-level sink fanout.
+/// callback runs. Reentrant logging can use the freed slot. If the callback
+/// unwinds, this record is consumed and other pending records remain available.
 #[inline]
 pub fn drain_one<R>(consume: impl FnOnce(&LogRecord) -> R) -> Option<R> {
     GLOBAL_INGRESS.drain_one(consume)
@@ -138,6 +139,70 @@ mod tests {
         core::str::from_utf8(&text.bytes()[..usize::from(text.byte_length())]).unwrap()
     }
 
+    #[test]
+    fn concurrent_publication_and_drain_preserve_complete_unique_records() {
+        let ingress = LogIngress::<2>::new();
+        // A critical record cannot be evicted. This guarantees one observation
+        // even if contention rejects every concurrent publication.
+        ingress.publish(Level::Error, core::format_args!("initial"));
+        let start = std::sync::Barrier::new(3);
+        let mut records = std::thread::scope(|scope| {
+            let ingress = &ingress;
+            let start = &start;
+            let producers: [_; 2] = core::array::from_fn(|producer| {
+                scope.spawn(move || {
+                    start.wait();
+                    for sequence in 0..8 {
+                        ingress.publish(
+                            [Level::Info, Level::Error][producer],
+                            core::format_args!("{producer}:{sequence}"),
+                        );
+                        std::thread::yield_now();
+                    }
+                })
+            });
+            let consumer = scope.spawn(move || {
+                start.wait();
+                let mut records = std::vec::Vec::new();
+                for _ in 0..64 {
+                    records.extend(ingress.drain_one(|record| *record));
+                    std::thread::yield_now();
+                }
+                records
+            });
+            for producer in producers {
+                producer.join().unwrap();
+            }
+            consumer.join().unwrap()
+        });
+        for _ in 0..2 {
+            records.extend(ingress.drain_one(|record| *record));
+        }
+        assert!(ingress.drain_one(|_| ()).is_none());
+
+        // Contention and critical replacement may discard publications. A
+        // delivered record must retain its own level and text without repeats.
+        let mut seen = [false; 16];
+        let mut initial_seen = false;
+        for record in records {
+            if message(&record) == "initial" {
+                assert_eq!(record.level(), Level::Error);
+                assert!(!core::mem::replace(&mut initial_seen, true));
+                continue;
+            }
+            let (producer, sequence) = message(&record).split_once(':').unwrap();
+            let producer: usize = producer.parse().unwrap();
+            let sequence: usize = sequence.parse().unwrap();
+            assert!(producer < 2 && sequence < 8);
+            assert_eq!(record.level(), [Level::Info, Level::Error][producer]);
+            assert!(!core::mem::replace(
+                &mut seen[producer * 8 + sequence],
+                true
+            ));
+        }
+        assert!(initial_seen);
+    }
+
     #[test]
     fn every_macro_produces_the_same_log_record_kind() {
         let _guard = GLOBAL_INGRESS_TEST
@@ -213,4 +278,48 @@ mod tests {
             Some(())
         );
     }
+
+    #[test]
+    fn callback_can_publish_and_drain_reentrantly() {
+        let ingress = LogIngress::<1>::new();
+        ingress.publish(Level::Info, core::format_args!("outer"));
+        assert_eq!(
+            ingress.drain_one(|record| {
+                assert!(!ingress.busy.load(Ordering::Relaxed));
+                ingress.publish(Level::Info, core::format_args!("inner"));
+                assert_eq!(
+                    ingress.drain_one(|inner| assert_eq!(message(inner), "inner")),
+                    Some(())
+                );
+                assert_eq!(message(record), "outer");
+            }),
+            Some(())
+        );
+        assert!(ingress.drain_one(|_| ()).is_none());
+    }
+
+    #[test]
+    fn callback_unwind_leaves_ingress_free_and_other_records_pending() {
+        let ingress = LogIngress::<2>::new();
+        ingress.publish(Level::Info, core::format_args!("consumed"));
+        ingress.publish(Level::Error, core::format_args!("pending"));
+        let result = std::panic::catch_unwind(std::panic::AssertUnwindSafe(|| {
+            ingress.drain_one(|record| {
+                assert_eq!(message(record), "consumed");
+                panic!("consumer failed");
+            });
+        }));
+        assert!(result.is_err());
+        assert!(!ingress.busy.load(Ordering::Relaxed));
+        ingress.publish(Level::Info, core::format_args!("new"));
+        assert_eq!(
+            ingress.drain_one(|record| assert_eq!(message(record), "pending")),
+            Some(())
+        );
+        assert_eq!(
+            ingress.drain_one(|record| assert_eq!(message(record), "new")),
+            Some(())
+        );
+        assert!(ingress.drain_one(|_| ()).is_none());
+    }
 }

firmware/libs/kaptan-log/src/lib.rs

@@ -1,12 +1,11 @@
 //! Bounded, transport-neutral runtime logging for Kaptan.
 //!
-//! This crate deliberately contains no MAVLink, board, filesystem, or clock
+//! Each sink filters and queues events for a transport adapter. Logging must
-//! dependency. Its sole configuration coupling is compile-time enum metadata
+//! not affect flight-control decisions or actuator authority.
-//! for [`LevelFilter`]. Flight code reports bounded events to
+//!
-//! independently filtered sink queues; a platform-owned adapter may later
+//! Text overflow stops formatting that propagates `core::fmt::Error`. Custom
-//! encode one queue for a link.
+//! `Display` implementations may ignore that error or perform arbitrary work,
-//! Reporting is observational and must never influence flight control or
+//! so fixed storage and queue bounds do not establish a universal time bound.
-//! actuator authority.
 
 #![no_std]
 

firmware/libs/kaptan-log/src/text.rs

@@ -27,6 +27,9 @@ impl LogText {
     }
 
     /// Formats one message without allocation.
+    ///
+    /// Overflow stops formatters that propagate `fmt::Error`. A custom `Display`
+    /// implementation can ignore that error or do unbounded work of its own.
     pub(crate) fn from_arguments(arguments: fmt::Arguments<'_>) -> Self {
         let mut text = Self::new();
         let _ = fmt::write(&mut text, arguments);
@@ -58,6 +61,9 @@ impl LogText {
 
 impl fmt::Write for LogText {
     fn write_str(&mut self, message: &str) -> fmt::Result {
+        if self.is_truncated() {
+            return Err(fmt::Error);
+        }
         let remaining = MAXIMUM_LOG_TEXT_BYTES - usize::from(self.length);
         let prefix = if message.len() <= remaining {
             message.len()
@@ -73,13 +79,18 @@ impl fmt::Write for LogText {
         let used = usize::from(self.length);
         self.bytes[used..used + prefix].copy_from_slice(&message.as_bytes()[..prefix]);
         self.length = (used + prefix) as u8;
-        Ok(())
+        if self.is_truncated() {
+            Err(fmt::Error)
+        } else {
+            Ok(())
+        }
     }
 }
 
 #[cfg(test)]
 mod tests {
     use super::{LogText, MAXIMUM_LOG_TEXT_BYTES};
+    use core::fmt::{self, Write};
 
     #[test]
     fn truncates_only_at_utf8_boundaries() {
@@ -95,4 +106,61 @@ mod tests {
             MAXIMUM_LOG_TEXT_BYTES - 1
         );
     }
+
+    #[test]
+    fn exact_multibyte_fit_is_not_truncated() {
+        let input = "x".repeat(MAXIMUM_LOG_TEXT_BYTES - 2) + "é";
+        let mut text = LogText::new();
+        assert_eq!(text.write_str(&input), Ok(()));
+        assert_eq!(text.write_str(""), Ok(()));
+        assert!(!text.is_truncated());
+        assert_eq!(text.bytes().as_slice(), input.as_bytes());
+        assert_eq!(text.write_str("a"), Err(fmt::Error));
+        assert!(text.is_truncated());
+        assert_eq!(text.bytes().as_slice(), input.as_bytes());
+    }
+
+    #[test]
+    fn overflow_latches_even_if_a_formatter_ignores_the_error() {
+        struct Fragments;
+        impl fmt::Display for Fragments {
+            fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
+                formatter.write_str(&"x".repeat(MAXIMUM_LOG_TEXT_BYTES - 3))?;
+                formatter.write_str("é")?;
+                let _ = formatter.write_str("🙂");
+                let _ = formatter.write_str("a");
+                Ok(())
+            }
+        }
+        let text = LogText::from_arguments(format_args!("{Fragments}"));
+        let expected = "x".repeat(MAXIMUM_LOG_TEXT_BYTES - 3) + "é";
+        assert!(text.is_truncated());
+        assert_eq!(
+            &text.bytes()[..usize::from(text.byte_length())],
+            expected.as_bytes()
+        );
+        assert_eq!(text.bytes()[MAXIMUM_LOG_TEXT_BYTES - 1], 0);
+    }
+
+    #[test]
+    fn large_standard_padding_stops_at_the_first_overflow() {
+        struct CountWrites {
+            text: LogText,
+            writes: usize,
+        }
+        impl fmt::Write for CountWrites {
+            fn write_str(&mut self, message: &str) -> fmt::Result {
+                self.writes += 1;
+                self.text.write_str(message)
+            }
+        }
+        let mut output = CountWrites {
+            text: LogText::new(),
+            writes: 0,
+        };
+        assert_eq!(write!(output, "{:>65535}", "end"), Err(fmt::Error));
+        assert!(output.writes <= MAXIMUM_LOG_TEXT_BYTES + 1);
+        assert!(output.text.is_truncated());
+        assert_eq!(output.text.bytes(), &[b' '; MAXIMUM_LOG_TEXT_BYTES]);
+    }
 }

flake.nix

@@ -22,6 +22,7 @@
           default = pkgs.mkShell {
             packages = [
               pkgs.rustup
+              pkgs.gcc
               pkgs.dfu-util
               pkgs.mdbook
               pkgs.ripgrep