AGENTS.md, HANDOFF.md, code-review-report.md, the implementation plan and the
aftermarket parity roadmap described the internal bench, its addresses and
in-house firmware; they now live in the private miner-firmware workspace and
are git-ignored here. The remaining docs and the fixture header no longer
point at them or at local infrastructure.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Ran the code-review-harness (four parallel reviewers: correctness, reliability,
performance, test-coverage) over the GUI crate and fixed every Critical/High
finding, plus grew the test + microbenchmark suite that gates the pipeline.
Correctness / reliability fixes:
- IP column sorted lexically, not numerically. The default view is IP-ascending
but Column::Ip fell through to the generic lowercase-string sort key, so
10.255.1.10 sorted before 10.255.1.2 on every launch. Add a numeric IP sort
key (u32, exact in f64) and switch view() to decorate-sort-undecorate so a
sort key is computed once per row, not twice per comparison.
- Scans had no cancellation. Stop/Apply/Play only dropped the receiver, leaving
the background thread probing dead hosts (each waits out an asic-rs timeout),
and Play could spawn a second concurrent scan + tokio runtime. Add a
ScanHandle with an Arc<AtomicBool> cancel flag the loop checks each iteration,
a begin_scan() helper that stops the in-flight scan before starting a new one,
and Stop now signals cancellation.
- Native column drag never rebuilt the control: column_reordered mutated
view.order but skipped rebuild_table(), so the store's creation order diverged
from display_columns() and cells rendered under the wrong headers on the next
scan tick. Rebuild after applying the reorder.
- format_si rounding rolled a mantissa up to "1000" without advancing the SI
prefix (999.9 W -> "1000 W", a 4-digit result). Round first, then roll over;
[99.95,100) drops its decimal.
- start_scan used .expect() on thread spawn, which would panic across the wx FFI
boundary; it now sends ScanMsg::Failed and returns the handle.
Tests / benchmarks (the pipeline's regression gate):
- Unit tests 24 -> 41: format_si/format_uptime boundaries, numeric IP sort,
upsert-new-IP, Light three-state + pool-slot-out-of-range cells, filter
AND/OR, reorder non-permutation guard; scan.rs parse_cidr/prefix_mask,
single-host /32, empty/whitespace, short-range single/reversed, parse_targets
aggregate cap + error propagation, validate_ranges, targets_to_cidrs
non-contiguous, auto-rescan-secs round-trip.
- Microbenchmarks 2 -> 9: view default-sort, sort-by-each-column, filtered view,
upsert cold-start, parse_line near MAX_ADDRS, parse_targets many ranges,
targets_to_cidrs large range. Floors are 5-10x above local numbers.
- New ci/test-compute-version.sh: 9 cases for the X.Y.Z logic (first release,
pre-bumped X, backend/UI bumps, docs-only no-op, major-wins-over-UI,
backend-wins-over-UI, nested ci/ and thirdparty/wx-rs paths). Wired in as a
fast-fail CI step.
CI: install libssl-dev + pkg-config (openssl-sys); add xauth so the headless
self-test actually launches under xvfb (was aborting on "xauth not found").
Deferred (documented in code-review-report.md, guarded by benches): Cow cells,
an ip->index map for upsert. Local: 41 tests + 9 benches + 9 shell tests green.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh