dangernoodle-io / TaipanMiner / 26733673553

90%

push

github

fix(mining): reverify HW-SHA shares to drop DPORT-corrupt reads; add stats-reset endpoint (#475)

On classic ESP32 (D0) the DPORT cross-bus erratum can return a corrupt
HW-SHA result. The existing guard only dropped all-zero corrupt reads; a
partially-corrupt read (a few leading-zero bytes, not all-zero) passed the
target compare, was recorded as a garbage record best_diff (~1e10), and
because it exceeded the DGB network target it fired a phantom block — then
persisted to NVS. The ASIC path already re-verifies in software; this brings
the HW-SHA dongle path to parity.

- share_validate: add share_reverify() — recompute SHA256d in software for
  the winning nonce and compare to the HW hash; mismatch means a corrupt read
- mining: drop the share (no best_diff, no block, no submit) when reverify
  fails, right after share_validate returns SHARE_VALID
- mining_pool_stats: add mining_pool_stats_reset() (zero pool slots + lifetime
  counters under mutex, persist via mining_pool_stats_save)
- mining: add mining_stats_session_reset() (zero session under mutex)
- webui: POST /api/stats/reset zeroes session + pool/lifetime stats, returns 204
- webui: System page gains a ConfirmDialog-guarded "Reset stats" action +
  resetStats() api helper
- webui embed: drop the now-unemitted index3.js chunk from the asset list +
  routes table (the System change shifted vite's shared-chunk graph; TA-408
  tracks generating the embed list to stop this recurring)
- tests: host coverage for share_reverify (good hash kept, byte-flipped
  dropped) and mining_pool_stats_reset; frontend tests for resetStats + the
  System reset action

Validated on esp32-wroom32: scrubbed the corrupt persisted best_diff/block to
0, mining continues at ~360 kH/s, a fresh real share is accepted with a sane
best_diff, ota_validated=true, heap healthy.

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

2765 of 3487 branches covered (79.29%)

Branch coverage included in aggregate %.

40 of 40 new or added lines in 4 files covered. (100.0%)

5409 of 5591 relevant lines covered (96.74%)

147734.24 hits per line