Build Triage

Build triage

1. Fix the first error, not the last. Everything after the first error is usually cascade. cargo build 2>&1 | grep -m1 -B1 -A6 "^error".
2. Exit code 0 can still be a failure. The exit code of a pipeline belongs to its last stage: cargo build 2>&1 | tail -3 reports tail's success. Trust the output text (error[, FAILED, panicked) over the code.
3. Bound the output. | tail -20 for verdicts (test runners put the verdict last), | head -30 for first errors, --stat for diffs. Never dump a full build log.
4. Never escalate a timeout. A command that timed out will time out again with a bigger budget. Kill leftovers (pkill -f cargo), then narrow: one package, one test, one file. cargo check -p one-crate beats a 10 minute workspace build.
5. Label pipeline stages so partial output says where it died: echo "=== build ===" && cargo check 2>&1 | tail -3 && echo "=== test ===" && cargo test 2>&1 | tail -5
6. Filter noise to signal. A hundred identical warnings hide one error: 2>&1 | grep -E "^(error|FAILED|thread)" | head -10.