343 Commits
Author SHA1 Message Date
noahandClaude Opus 4.8 8f7aab3f80 GUI round 11 (live-miner items #5/#6b/#8/#11): control-board scan gaps, VNish partner build, CVITek PIC bus
All root-caused and fixed against the live fleet (read-only scans; no device
state changed). asic-rs changes go through the vendored tree + thirdparty/asic-rs
patch queue (0003-0005) with refreshed checksums, per ADR 0003.

#5 (XIL/CVITek + WhatsMiner blank control board):
- VNish reports the board in short form in /api/v1/info "platform": "cv"
  (CVITek), "xil" (Xilinx), "aml" (AMLogic); AntMinerControlBoard::parse matched
  only the full names, so CVITek/Xilinx boards were blank. Added CV/XIL.
- Old M30S firmware (< 2022.7.29 -> v1 backend) omits summary's "CB Platform"
  and v1's parse mandatorily stripped an "ALLWINNER_" prefix. Added get_version
  /Msg/platform as a fallback source and made the strip optional.
- Verified live: S19k Pro cv->CVITek, S19 xil->Xilinx, M30S+ VE40->H6OS.

#6b/#11 (VNish partner / shared-devfee build):
- The partner build shows as /api/v1/info "build_name" ("WildRose",
  "vnishreturn"). The VNish backend (v1_2_0 + v1_3_0) now reads /build_name and
  combines it with the /fw_version number, so firmware_version reads e.g.
  "WildRose 1.2.7". Verified live on S19j Pro / S19k Pro.

#8 (hashboard PIC detection):
- pic_probe tried i2c-1/i2c-0 for CVITek, but a live VNish CVITek S19k Pro
  exposes the hashboard MCU on i2c-3 (i2cdump 0x48 = period-8 -> PIC). Broadened
  bus_candidates (cvi->[3,4,1,0], aml->[1,3,4,0], bb->[0,1,3,4], else->[1,0,3,4]);
  probe_pic stops at the first bus that classifies. The classify logic was
  already correct + unit-tested; only the bus map was wrong.

The GUI picks up #5/#6b automatically through asic-rs (CPU/Board columns now
populate for VNish CVITek/Xilinx and old M30S; FW Version shows the partner
build); #8 is fixed in the GUI's own pic_probe. The miner_probe diagnostic's
local-only mutation modes are left uncommitted as before.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
2026-09-19 09:07:57 -06:00
noahandClaude Opus 4.8 e593548746 GUI round 11 (#4): drag-reorder + multi-select columns dialog (no up/down buttons)
Replaces the columns dialog's wxRearrangeCtrl (checkboxes + up/down buttons)
with a wxListCtrl-based list (new wxbr_reorder_list / Widget::reorder_list):
- No up/down buttons; rows are reordered by dragging.
- Native multi-select: Shift/Ctrl-click and rubber-band drag.
- Keeps per-row checkboxes for show/hide; same [idx | !idx] order encoding, so
  apply_columns is unchanged.

Drag-reorder is implemented by recording the selected rows' base indices on
wxEVT_LIST_BEGIN_DRAG and moving that group to the row under the cursor on the
following mouse-up (no real DnD loop is started, so the mouse-up still arrives).

Verified: the dialog opens without crashing and renders the checkable, button-
free list (headless click test). The drag and shift/ctrl multi-select
interactions are not headlessly simulatable and need interactive confirmation.

Bridge + Rust mirrored into patches 0002 (widgets.rs) and 0003
(widgets_bridge.cpp) per ADR 0003.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.19.2
2026-09-18 18:03:47 -06:00
noahandClaude Opus 4.8 2fd433c3b0 GUI round 11 (#15): drawn icons on the KPI cards
Each KPI card's value now leads with a small vector icon drawn into a
transparent bitmap via wxGraphicsContext (new wxbr_kpi_icon -> Widget::kpi_icon):
rabbit = Hashing, turtle = Slow, broken wine glass = Not hashing, otter =
Paused, hive = SHA-256 Hashrate, lightning bolt = Est. power; Total has none.
The icon takes the card's value colour (amber on the neutral Est. power card so
the bolt reads).

The icons are simple flat silhouettes so they render at ~22px without external
assets (no emoji/pictographs in source). The rabbit, turtle, wine glass, hive,
and lightning read clearly; the otter is the least distinct at this size.

Bridge + Rust mirrored into patches 0002 (widgets.rs) and 0003
(widgets_bridge.cpp) per ADR 0003. Self-test passes.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.19.1
2026-09-18 17:51:09 -06:00
noahandClaude Opus 4.8 5399f9fc49 GUI round 11: fix headless self-test after Status dropped from default columns
Round 11 #12 removed the Status column from the default view, but the runtime
self-test (CI's "Headless UI self-test", separate from cargo test) read each
miner's status from the displayed Status cell -- now absent -- so three status
spot-checks failed and the self-test binary exited non-zero, failing CI. Status
is a model property, so the checks now read it straight from the row
(status_of(row) = rows[view_idx[row]].status()) instead of a displayed cell.
Self-test passes (exit 0).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.19.0
2026-09-18 17:42:53 -06:00
noahandClaude Opus 4.8 6b4961d7bc GUI round 11 (#2): collapse the two wide KPI cards when the window is too narrow
When the window is narrowed past the width needed to fit all seven KPI cards at
their minimum size, the two wide cards (SHA-256 Hashrate, Est. power) are hidden
so the five miner-count cards fill the row; they return when there is room again.
The five count cards still floor the window's minimum width (unchanged).

Adds a resize event to the wx-rs bridge (WxEV_SIZE -> WxEventKind::Sized, p1 =
client width) and a wxbr_sizer_show helper to hide/restore a sizer child and
reclaim its space. The main panel binds it; App::on_resize toggles the two cards
on a state change only (no per-drag relayout thrash). Bridge + Rust changes
mirrored into patches 0002 (widgets.rs) and 0003 (widgets_bridge.cpp) per
ADR 0003.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
2026-09-18 17:39:56 -06:00
noahandClaude Opus 4.8 fc7c665166 GUI round 11 (#9): option to hide columns with no scanned values
The Columns dialog gains a "Hide columns with no scanned values" checkbox. When
on, columns whose cell is blank for every current row are hidden from the table.
It is non-destructive: the hiding is applied at table-build time via a new
App::table_columns() (used by every table-path caller in place of
view.display_columns()), so clearing the option restores the columns and the
underlying column order/visibility is never mutated. IP is never hidden, and
nothing is hidden before the first scan (no rows -> nothing empty).

Note: while the option is hiding columns, drag-reorder of the header is a no-op
(the reorder validates against the full visible set); turn the option off to
reorder.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
2026-09-18 17:32:20 -06:00
noahandClaude Opus 4.8 ae437c0d24 GUI round 11 (batch 1): default columns, control-board split cleanup, Stock/Env fixes, relative sort, consistent zebra + focus-dimming, hashboard/light dots, access icon
Table/model (Rust):
- #12: default columns are now IP, Model, Errors, Hashrate, Power, Chip temp,
  Firmware, CPU, Boards, Access, in that display order (initial ViewState and the
  built-in "Default" preset both updated).
- #14: removed the Control board column (CPU + Board replace it); the underlying
  control-board string stays on MinerRow for the SSH PIC probe.
- #6 (part 1): firmware family strips the vendor word asic-rs prepends on stock
  firmware ("AntMiner Stock" -> "Stock"), moving it into FW Version.
- #17: air-cooled miners' ambient reading routes to a new "Env temp" column
  instead of "Fluid temp" (asic-rs reports ambient in its fluid field and has no
  cooling-type flag; cooling type is inferred from the model name).
- #16: relative-hashrate (actual/expected) sort -- the Hashrate header cycles
  absolute-asc/desc then relative-asc/desc, with right-click menu entries;
  HashrateRel is a sort-only pseudo-column.

Rendering (wx-rs SegmentRenderer bridge, mirrored to patches/0003 per ADR 0003):
- #3: the per-row zebra background now fills the whole column-wide cell (wx
  shrinks a custom renderer's rect to the content size, so the earlier fill
  covered only the text; Render now fills GetOwner()->GetWidth() and wxGTK clips
  it to the cell).
- #1/#7: uncoloured segments (combined-cell slashes, "Stock" firmware) take the
  tree widget's themed text colour at its live GTK state, so they dim on
  window-backdrop like the native cells.
- #10: the Boards column renders per-hashboard status dots (green hashing / red
  faulted / grey undetected, board 1 rightmost); the Light column renders a
  yellow locate dot while flashing.
- #13: the Access column renders a padlock icon (green open = writable, red
  closed = view-only) instead of "locked"/"unlocked" text.

Shapes draw from control-char sentinels (never emoji/pictographs), font-
independent. Self-test verifies the icon columns structurally. 82 unit tests
pass.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
2026-09-18 17:24:09 -06:00
noahandClaude Opus 4.8 b343746718 GUI round 10 (batch 2): consistent zebra on custom columns + neutral KPIs dim with focus
R10.1 (zebra consistency): wxGTK does not paint a custom renderer's cell
background, so the per-row zebra colour set via GetAttrByRow appeared on the
native columns but not on the SegmentRenderer columns (Hashrate, Chip temp,
Firmware, ...), making the striping look inconsistent. SegmentRenderer::Render
now paints the row's GetAttr() background itself (skipping selected rows, which
the control already highlights) so every column stripes identically.

R10.2 (focus dimming): the Total-miners and Est-power KPI cards and the plain
segments in combined cells forced an explicit foreground colour, so they did
not dim when the window lost focus like the rest of the dashboard. mk_card and
repaint_theme now leave ColorKind::Neutral cards on the default text colour, and
SegmentRenderer draws an uncoloured segment with the DC's current text
foreground (falling back to wxSYS_COLOUR_WINDOWTEXT) instead of a hardcoded
system colour.

The widgets_bridge.cpp edit is mirrored into the wx-rs patch queue
(patches/0003-widget-bridge-cpp.patch) per ADR 0003.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.18.1
2026-09-18 15:44:48 -06:00
noahandClaude Opus 4.8 aa75031112 GUI round 10 (batch 1): firmware family colour + FW Version, CPU/Board split, per-board chip temp, error count
#3 Firmware column now shows the firmware FAMILY, coloured (Stock=default,
VNish=green, Braiins=purple, LuxOS=gold, BixBit=pink, ePIC=light-blue,
Hive=orange, Alkane=navy -- Alkane wired for our future scannable firmware);
the version moves to a new "FW Version" column.

#4 Control board split into a "CPU" column (AML/CVI/BBB/XIL, WhatsMiner
H3/H6/H616) and a "Board" column (C76/C88/CBE/C49/C55, CB2/CB4/CB6). Fixed the
AMLogic 140mm heuristic so S21 Pro -> CBE (was C76). The combined ControlBoard
column stays available (hidden).

#5 "Worker" now shows just the rig name; a new "Pool account" column holds the
account (pool username "account.rig" split on the first dot).

#6 Chip temp shows per-hashboard temps as integers ("77 / 64 / 74"), each
coloured on the blue-to-red scale, falling back to the integer average.

#7 Errors column shows just the count (a number, even for one; no "errors"
word); the hover tooltip lists the messages, now concise -- Info dropped, the
repetitive "[Error]"/"[Warning]" prefixes stripped, and duplicates collapsed.

Default profile updated to show Firmware/FW Version/CPU/Board/Errors. 80 offline
tests pass; headless self-test PASS; screenshot-verified the colours, splits,
per-board temps and error counts in the offline demo.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.18.0
2026-09-18 15:32:38 -06:00
noahandClaude Opus 4.8 8eb9a2caad GUI round 10 (#8): Z15/Z15J Equihash hashrate shows the correct SI prefix (k, not T)
Root-caused and verified against the live fleet. An Antminer Z15 shows its
hashrate as e.g. "288T" when it is really ~288 kSol/s ("288k"). Equihash miners
report a Sol/s rate in the cgminer "GHS" key, but asic-rs AntMiner parse_hashrate
takes the unit from the key name ("GHS" -> GigaHash), scaling the value 1e9x too
large. The value is a raw Sol/s count, so for Equihash use the base Hash unit;
the toolkit`s SI-prefixing then yields "288k".

Matched by algorithm, and by "Z15" model name for the Z15J (which asic-rs leaves
algo Unknown). Fixed in both the v2020 and v2023_07 AntMiner backends, via
asic-rs patch queue 0002. Verified live: Z15 288T -> ~288k, Z15J 138T -> ~110k.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
2026-09-18 15:18:12 -06:00
noahandClaude Opus 4.8 9e2b453eb5 GUI round 9 (#4 v2): WhatsMiner v2 power-limit now reports success on the apply-restart
Follow-up to 3530d67a. Verified on the live fleet that the v2 (firmware
< 2024-11-01) "adjust_power_limit" command DOES apply the new watt limit --
confirmed the set value settles via the read-only :4028 "Power Limit" after the
miner restarts to re-tune. But the restart drops the connection mid-command, so
the read times out; the code then reported that as a failure even though it
applied.

Both v2 and v3 set_power_limit now treat a transient RPC error after the command
(is_expected_write_error -- read timeout / connection reset, the helper the
pause/resume paths already use) as success. A genuine rejection is a
StatusCheckFailed and still errors. v1 miners keep the "not supported" bail.

Verified: v2 M30S+ (fw 20240930) set 3500 -> :4028 settles to 3500, restore 3600
-> settles to 3600 and the miner resumes hashing.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
2026-09-18 15:01:42 -06:00
noahandClaude Opus 4.8 3530d67a5c GUI round 9 (#4): fix WhatsMiner power-limit (v3 auth/param) + asic-rs patch queue
Round-9 #4 -- "the power-limit on the WhatsMiners doesn't apply properly" --
root-caused and verified against the live fleet, then fixed in vendored asic-rs
via a new patch queue (ADR 0003).

Root cause: asic-rs routes a WhatsMiner to its v3 (WhatsMinerOS) backend when
the firmware date is >= 2024-11-01. The v3 privileged power-limit write
(command "set.miner.power_limit" on port 4433) authenticated against account
"super" but hashed the token with the WEB/admin password, and sent the watt
value as a float -- so the miner rejected every write with "invalid json or
password is wrong". asic-rs then swallowed that (Ok(data.is_ok()) -> Ok(false))
so the GUI showed "refused" with no reason. (Older miners, fw < 2024-11-01, use
the v2 "adjust_power_limit" path, which restarts the miner on a change -- see
below.)

Fix (thirdparty/asic-rs/patches/0001):
- v3 set_auth: the "super" account's token password is a SEPARATE v3 credential
  (factory default "super"), NOT the web/admin password carried by `auth`; use
  "super". The web password still drives the v2 (admin) path.
- v3 set_power_limit: send the watts as a bare integer (a float trips the
  firmware's "invalid json" parser).
- v2 + v3: propagate the RPC error with `?` instead of swallowing it to
  Ok(false), so a real failure reaches the round-9 bottom-right error indicator
  with the miner's own message.

Verified on a live M30S+ (fw 20241108, api 3.0.1) via a read-only 4028
capture/restore harness: the fixed request returns {"code":0,"msg":"ok"};
asic-rs set_power_limit now returns Ok(true) and the miner applies the limit
(it restarts to re-tune, ~10 min ramp -- expected WhatsMiner behaviour). Matches
the pyasic and HashSource whatsminer-api-3 references. All test miners were
returned to their prior configured limit.

asic-rs patch-queue mechanism (ADR 0003 updated): asic-rs is cargo-vendored, so
in-place edits are documented as a numbered patch under thirdparty/asic-rs/
plus a checksum-refresh helper and scripts/apply-asic-rs-patches.sh for
re-vendoring. The committed vendor/ tree has the patch applied and its
.cargo-checksum.json refreshed, so builds need nothing extra (CI builds fresh;
a note documents the local `cargo clean -p` cargo-caching gotcha).

Known follow-up: the v2 backend (fw < 2024-11-01) reaches the miner via
"adjust_power_limit" but a changed value restarts the miner and the in-restart
status check reports failure; whether it applies correctly needs its own
verification. The error-propagation change surfaces that honestly meanwhile.

Compile-checked (release + tests); 79 offline tests pass.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.17.0
2026-09-18 14:10:05 -06:00
noahandClaude Opus 4.8 3ac4e104a0 GUI round 9 (#3): mask power-limit on VNish; naming passthrough fix; live-scan probe tool
#3 (power-limit shown where it shouldn't be): asic-rs advertises
supports_set_power_limit = true unconditionally for whole firmware families,
VNish included. Per operator guidance -- VNish Antminers should not expose the
power-limit control, while Braiins keeps it -- mask the capability for VNish
firmware in to_row; every other backend keeps asic-rs's answer. Verified
against the live fleet (all VNish Antminers report true; Antminer stock Z15 /
S21Pro already report false).

#11 follow-up: asic-rs Unknown-model passthroughs carry the raw vendor string
with the brand still in it (seen live: a NoPIC S19 as "ANTMINER S19 (88)").
Drop a redundant leading brand word in clean_antminer so the Model column
never repeats the Brand column: "ANTMINER S19 (88)" -> "S19 (88)".

New read-only diagnostic: crates/alkane-gui/src/bin/miner_probe.rs -- a
standalone asic-rs CLI (`miner_probe scan <cidr>...`) that reuses the GUI's
MinerFactory + saved-credential setup to inventory a range (ip/make/firmware/
model/control-board/watts/supports_power_limit/supports_tuning) without going
through the wx event loop. Used to validate the round-9 naming + capability
work against real hardware; kept as a hardware-check tool. It performs no
mutations (asic-rs exposes no power-limit read-back, so exact restore-to-prior
cannot be guaranteed from it alone).

naming unit tests + headless self-test PASS; 80 offline tests pass.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.16.0
2026-09-18 12:05:37 -06:00
noahandClaude Opus 4.8 3c99934513 GUI round 9 (#11): device-naming cleanup -- Brand/Submodel columns, clean model names, control-board short codes
asic-rs hands us compact Rust-identifier model strings ("S19KPro",
"M30SPlusVE40") and bare control-board family names ("Xilinx", "AMLogic",
"BeagleBoneBlack", WhatsMiner "H3"/"H6"/"H6OS"/"H616"). This adds a pure
display-layer transform (new crate::naming) that turns those into the
human-facing identity the table shows -- no asic-rs edit, so it stays a
build-on-top layer over the vendored crate.

- Model column now shows the cleaned name: brand dropped, family-letter
  suffix lowercased, word suffixes spaced, "+"/"++" for Plus/PlusPlus, and
  (WhatsMiner) the trailing version block moved out. "S19KPro" -> "S19k Pro",
  "S21Pro" -> "S21 Pro", "M30SPlusVE40" -> "M30S+".
- New Brand column ("Antminer"/"WhatsMiner"/...) and Submodel column
  (WhatsMiner version block, e.g. "VE40"; blank for Antminer -- its hashboard
  type is not reported over the network). Both hidden by default; available
  via the Columns menu and filterable.
- Control-board column now shows the Bitmain short code: BeagleBoneBlack ->
  "BBB (C77)", CVITek -> "CVI (C88)", Xilinx -> "XIL (C49)" (S17-era) /
  "XIL (C55)" (S19-era), AMLogic -> "AML (C76)" (120mm) / "AML (CBE)"
  (140mm); WhatsMiner H3 -> "H3 (CB2)", H6 -> "H6 (CB4)", H6OS ->
  "H6os (CB4)", H616 -> "H616 (CB6)". Xilinx/CVITek/AMLogic/BeagleBone are
  already detected by asic-rs, so no crate extension was needed.

Heuristics (asic-rs reports only the board family, never the exact board
number): the AMLogic 120mm-vs-140mm (C76/CBE) and Xilinx S17-vs-S19-era
(C49/C55) splits are inferred from the model family. These are documented in
crate::naming and are the one place a live-hardware correction may be needed.

ALL_COLUMNS is now 41. Fixtures refactored to carry asic-rs-style
make/model_id/board so they exercise the real transform. naming has unit
tests for the model/submodel/brand/control-board rules; the headless
self-test gained 32 checks (brand/model/submodel/control-board per fixture),
all PASS. 79 offline tests pass; screenshot-verified the cleaned Model and
control-board codes in the default view.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.15.0
2026-09-18 11:29:15 -06:00
noahandClaude Opus 4.8 9b5d811b1e GUI round 9 (batch 2): KPI box layout, bottom-right error indicator, algorithm filter + hashrate cycling
Round-9 UI items #5, #6, #7 and the #9 algorithm-cycle follow-on.

#6 KPI boxes: switch the header from a wrap sizer to a horizontal box
sizer with every card at proportion 1 + EXPAND, so the cards are always
equal width and grow to span the full window; a per-card min size
(116x62) floors them at their largest content. Fix the filter-alignment
drift by re-running each card's sizer in refresh_kpis after set_label
(the centring spacers now recompute around the new value width). The
active-filter highlight now paints the card panel and both its texts, so
the whole box tints instead of only the strip behind the text. Bind the
card click on the value and label static texts too (a click on the number
or caption registers the quick-filter, not just the card gap).

#5 error indicator: add a second, colour-tinted status-bar field pinned
to the bottom-right corner (wxbr_status_indicator_init/set in the widget
bridge; a wxStaticText repositioned into the right field on resize). Scan
and apply outcomes route here without overwriting the left scan-status
text: a total apply failure and a scan failure show red, a partial apply
failure amber, full success clears it; a new scan/action clears any stale
message.

#7 algorithm filter: the Filter dialog now offers every column (via a new
ViewState::filter_columns) rather than only the visible ones, so Algorithm
-- and any other hidden metric -- can be filtered without first un-hiding
its column. Filters already match on the plain model cell, independent of
visibility.

#9 hashrate cycling: compute per-algorithm hashrate totals in compute_kpis
(Kpis::hashrate_by_algo, ordered by total). The hashrate KPI card shows
the selected algorithm's total under a "<Algo> Hashrate" caption and
cycles through the running algorithms when clicked (a no-op with a single
algorithm). SHA256 renders as "SHA-256".

Bridge provenance patches 0002/0003 regenerated to match the -patched tree.
Compile-checked (release + tests); 74 offline tests pass; headless
self-test PASS; screenshot-verified equal-width cards, full-box highlight,
click-through-text, and centred values after filtering.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.14.0
2026-09-18 11:09:50 -06:00
noahandClaude Opus 4.8 77399bdd06 GUI round 9 (batch 1): column-reorder persistence, post-action refresh, algorithm column, hashrate KPI label
- #1 Column reorder now persists (incl. after sorting a moved column). On wxGTK,
  GetColumnPosition() does not reflect a native GtkTreeView column drag, so the
  reorder was read as a no-op and the rebuild snapped columns back. dvlc_col_at_pos
  now reads the real display order from the GtkTreeView (gtk_tree_view_get_columns
  + GTKColumnToWX) on __WXGTK__, falling back to GetColumnPosition elsewhere.
- #2 A right-click config action (e.g. Flash lights) now re-probes the miners it
  targeted when it finishes, so the change shows in the table (skipped after a
  flash, since the unit is rebooting). New App.action_targets.
- #7 New Algorithm column (from asic-rs hashrate.algo: SHA256/Scrypt/Equihash/...),
  categorical so it is filterable via the value checklist.
- #9 SHA-256 Hashrate KPI: card relabelled "SHA-256 Hashrate" and the value drops
  the "/s" ("685 TH", "1.2 PH"). (Per-algorithm click-cycling deferred.)

Note on #8 (Z15 shows T not k): our formatting already derives the SI prefix from
the base value (si_mantissa), so a wrong 'T' means asic-rs reports the wrong unit
for that Equihash model -- an asic-rs backend fix that needs a live Z15 to verify.

cargo check --release --tests clean; patch 0003 regenerated + validated.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.13.0
2026-09-18 10:41:21 -06:00
noahandClaude Opus 4.8 408b7d12b4 GUI: real access/lock-state detection via credentialed digest probe
asic-rs has no lock-state signal (a wrong-password miner just silently returns
fewer fields, because the collector drops the 401). This builds the real check
ON TOP of asic-rs:

- access_probe.rs: a NON-mutating, digest-authenticated GET to a privileged
  endpoint (Antminer get_miner_conf.cgi, read-only). The HTTP status is decisive
  -- 200 = credentials authenticate (writable -> unlocked), 401/403 = rejected
  (view-only -> locked); anything else -> indeterminate. Pure classifier +
  unit tests; shells out to curl --digest (no HTTP crate) so the offline build
  is unaffected.
- Right-click "Check access (auth)" on the selected miners runs start_access_probe
  on a background thread using the saved scan (web) credentials (default root/root),
  updating write_locked in the rows as results arrive. Fake-scan mode returns
  deterministic demo results. Guarded against a flash / concurrent probe.

Replaces the previous capability-based inference for the miners the user
explicitly checks. cargo check --release --tests clean.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.12.0
2026-09-18 10:32:44 -06:00
noahandClaude Opus 4.8 50cd463bee GUI fix: guard out-of-range dvlc GetItem (crash selecting rows after refresh)
Round-9 #10: selecting miners could trip
`datavcmn.cpp: assert "row < m_hash.GetCount()" failed in GetItem(): invalid
index` and crash. wxDataViewIndexListModel::GetItem() asserts on an out-of-range
row, and three bridge helpers called it with an index that can go stale when the
row set shrinks (a filter, or a scan that returns fewer rows): wxbr_dvlc_
ensure_visible (the round-8 scroll-hold, unguarded), wxbr_dvlc_select_rows and
wxbr_dvlc_select_row (only lower-bounded). All three now bound the row against
the store's item count before calling GetItem, so a stale/empty index is a
no-op instead of a crash.

Local wx-rs bridge compile clean; patch 0003 regenerated + validated.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.11.1
2026-09-18 10:29:14 -06:00
noahandClaude Opus 4.8 d73299166e GUI: real PIC detection via on-board SSH + i2c probe (right-click action)
asic-rs and the miner's network API cannot report PIC-vs-noPIC (only one legacy
SKU, ANTMINER S19J88NOPIC, carries it in the model string; the authoritative
signal lives on the control board's I2C bus). So this builds the real detector
ON TOP of asic-rs as an opt-in SSH probe, per the project wiki
(30-power-thermal-and-board-identity/hashboard-eeprom-pic.md):

- pic_probe.rs: SSH to the control board and run i2cdump -y <bus> 0x48 b, then
  discriminate by the register period -- a PIC/MCU board exposes an 8-byte
  register file that WRAPS every 8 bytes ("17 00 4b 50 7f 00 00 00 ..."), while
  a noPIC board has a 4-register LM75 temp sensor (4-byte wrap). The classifier
  is pure + unit-tested against the wiki's example dumps and is conservative
  (indeterminate -> None, never a guess). The I2C bus is chosen from the
  control-board family (AMLogic->1, BeagleBone->0, else try both). We shell out
  to ssh/sshpass (no SSH crate) so the offline/vendored build is unaffected.
- Right-click "Detect PIC (SSH)..." on the selected miners: a dialog collects
  SSH login (prefilled from a new 0600 ssh-credentials.txt, default user root),
  then start_pic_probe runs one miner at a time on a background thread (each is a
  slow SSH round trip; never part of the fast network scan) and upserts
  pic_present into the rows as results arrive. Fake-scan mode returns
  deterministic demo results. Guarded against a firmware flash / concurrent probe.

The classifier is unit-tested; the live SSH round-trip needs a reachable bench
miner with SSH + i2c-tools to verify end-to-end. cargo check --release --tests
clean.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.11.0
2026-09-18 10:25:15 -06:00
noahandClaude Opus 4.8 a81fef3c7d GUI round 8: Errors column hover tooltip (full message list)
Item 12 (tooltip half): hovering an Errors cell now shows the full list of the
miner's error/event messages ("[Severity] text (component)", most severe first)
as a tooltip; the cell itself keeps the compact red summary.

wxDataViewCtrl has no per-cell tooltip API, and wx mouse-motion events are not
delivered to wx handlers on the native GTK control, so this uses GtkTreeView's
native "query-tooltip" signal directly (guarded by __WXGTK__): a per-cell
tooltip map (row, model column) -> text on the AttrListStore, populated from
Rust via a new dvlc_set_cell_tooltip bridge fn, and looked up by the signal
handler using dv->GTKColumnToWX(gcol)->GetModelColumn(). The map is populated on
every platform (a no-op where nothing reads it) so the non-GTK platform round
can wire its own tooltip surface later.

Local wx-rs bridge compile + full cargo check clean; patches 0002/0003
regenerated + validated.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.10.2
2026-09-17 17:37:18 -06:00
noahandClaude Opus 4.8 365732b5d4 GUI round 8b.2 fix: rubber-band drag-select uses native GtkTreeView on GTK
The wx-mouse-event rubber-band approach (LEFT_DOWN/MOTION/LEFT_UP) does not work
on the native GTK wxDataViewCtrl: the underlying GtkTreeView grabs the pointer
during a button drag, so wx never delivers the MOTION events, and only the
initial click row was selected (verified against the shipped binary).

Fix: on GTK, enable the GtkTreeView's own native rubber-band selection via
dv->GtkGetTreeView() + gtk_tree_view_set_rubber_banding(TRUE), guarded by
__WXGTK__ (gtk/gtk.h included only there). The wx-mouse-event approach is kept
as the non-GTK fallback for the future Windows/macOS builds, where it will be
verified in the platform round. The column-header right-click menu (native wx
dataview event) already worked and is unchanged.

Local wx-rs bridge compile clean; patch 0003 regenerated + validated.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.10.1
2026-09-17 17:27:16 -06:00
noahandClaude Opus 4.8 98603077e3 GUI round 8: view presets (save/load column arrangement, filters, sort)
Item 11: a "Presets" button (top-right) opens a dialog to save the current
table view -- column arrangement + visibility, sort, and all active filters --
as a named profile, load a saved one, or delete it. The built-in "Default"
profile is always the first preset and shows IP, Model, Hashrate, Power, Chip
temp, Firmware, Control board, PIC and Status; the app now boots with it.

- New presets.rs: a serde-serialized PresetView snapshot of ViewState (Vec-based
  so the JSON has no enum map keys), Preset { name, view }, and load/save to
  ~/.config/alkane-toolkit/presets.json. "Default" is re-seeded on load and not
  written, so it can't be lost or deleted. A loaded preset missing newer columns
  still covers every column (they're appended hidden).
- serde/serde_json added to alkane-gui (already vendored + locked via asic-rs);
  Column/ColFilter/TextOp/NumOp derive Serialize/Deserialize and ViewState is
  now Clone.
- ui.rs: build_presets_dialog + apply_preset_load/save/delete, wired to the new
  Presets button.

Unit tests: default-preset round-trips through ViewState and JSON.
cargo check --release --tests is clean.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.10.0
2026-09-17 17:21:49 -06:00
noahandClaude Opus 4.8 bc9f847689 GUI round 8b.2: column-header menu + rubber-band drag-select
Bridge + UI (local wx-rs bridge compile + full cargo check clean).

- Column-header right-click menu (#5): right-clicking a column header opens an
  Excel-style menu -- Sort ascending / descending, Auto-size this column, Hide
  <column> (not offered for the fixed IP column), and Columns... (the full
  show/hide picker). New wx event WxEV_DVLC_HEADER_RIGHT_CLICK bound from the
  native wxEVT_DATAVIEW_COLUMN_HEADER_RIGHT_CLICK, plus a wxbr_dvlc_autosize_col
  bridge fn (wxCOL_WIDTH_AUTOSIZE). (Sticky/frozen columns are the separate 8c
  dual-control piece.)
- Rubber-band (click-drag) range selection (#10): wxDataViewCtrl has no native
  drag-to-extend selection (only ctrl/shift-click), so it is tracked in the
  bridge -- LEFT_DOWN records the anchor row (and lets the native handler select
  it), dragging with the button down selects the inclusive row range under the
  cursor, LEFT_UP clears the anchor; modifier-clicks are left to the native
  multi-select. Reuses the existing Select/UnselectAll primitives and emits
  SELECTION_CHANGED so the model's selected-IP set stays in sync.
- KPI cards widened to 150px so "685 TH/s" / "21.3 kW" are no longer clipped.

wx-rs patches 0002/0003 regenerated + round-trip validated.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.9.2
2026-09-17 17:13:27 -06:00
noahandClaude Opus 4.8 71cdf2d988 GUI round 8b.1: scroll-hold, zebra striping + vertical gridlines, default-colour text
Bridge + rendering rework (validated with a local wx-rs bridge compile + full
cargo check; CI is the build oracle).

- Scroll position hold (#2): the table no longer jumps to the top as new rows
  stream in during a scan. fill_rows captures the top visible row before the
  clear+refill and restores it after (new wx-rs bridge fns dvlc_get_top_row /
  dvlc_ensure_visible, wrapping wxDataViewCtrl::GetTopItem / EnsureVisible).
- Default-colour text (#3b): rows are no longer tinted wholesale by status. Every
  cell is now the default text colour (black in light mode, white in dark -- the
  SegmentRenderer falls back to wxSYS_COLOUR_WINDOWTEXT for any run with no
  explicit colour), so IP addresses, slashes, expected/target values and all
  non-conditional cells read in the default colour. Only the conditionally-
  formatted columns carry colour: Hashrate/Boards/Chips actual (ratio scale),
  Chip temp (blue-red), fans (grey-blue), Access (red/green), Errors (red), and
  the Status column keeps its own status colour on its own cell.
- Zebra striping (#8): odd rows get a subtle theme-aware background via the
  AttrListStore (a per-row BACKGROUND attr, so it does not disturb the
  default-foreground fallback and the native selection highlight still shows
  through). This replaces the status row-tint as the row separator. (The earlier
  abandoned attempt used SetAlternateRowColour / GTK CSS against the status
  foreground tint; removing that tint is what makes the background approach
  clean now.)
- Vertical gridlines (#8): wxDV_VERT_RULES added to the table style (row lines
  were already on), so columns have borders.
- KPI cards: wxWrapSizer now drops wxEXTEND_LAST_ON_EACH_LINE, so the last card
  on a line is no longer stretched -- the uniform-size KPI boxes stay uniform.

wx-rs patches 0002/0003 regenerated and round-trip-validated against the source.
cargo check --release --tests is clean.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.9.1
2026-09-17 16:55:27 -06:00
noahandClaude Opus 4.8 4fea3becf6 GUI round 8a.2: PIC, Access (locked), and Errors columns
Adds the round-8 columns that asic-rs either already exposes or that we infer.
Control board type (#9) and identify-light state (#4b) were already scannable
columns; this adds the three new ones:

- PIC (#9): asic-rs does not report PIC-vs-noPIC over the network (it needs an
  on-board SSH/I2C probe -- a separate feature), so this is a best-effort
  inference from the model string ("NoPIC" variants), showing "?" when unknown.
  New MinerRow.pic_present.
- Access (#13): locked (view-only) vs unlocked (writable), inferred from whether
  the miner exposes any writable control capability (asic-rs has no direct
  lock-state field). Red = locked, green = unlocked. New MinerRow.write_locked
  and MinerCaps::any_control().
- Errors (#12): wires asic-rs's structured MinerData.messages (previously
  dropped) into MinerRow.errors as "[Severity] text (component)" lines, most
  severe first. The column shows a summary (blank / the single message / "N
  errors") in red when non-empty; the full list becomes a hover tooltip once the
  tooltip bridge support lands in 8b.

The identify-light column header is relabelled "Light" (from "Fault light") to
match the round-8 terminology. Fixtures populated with demo values so all new
columns render offline. Unit tests for the three cell renderers + any_control.

cargo check --release --tests is clean.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.9.0
2026-09-17 16:37:21 -06:00
noahandClaude Opus 4.8 7781ff71d3 GUI round 8a: no phantom-offline rows, hashrate/temp conditional formatting, multi-range parse, uniform KPI cards
First slice of the round-8 batch (Linux; Windows/macOS builds deferred to a
dedicated platform round per the design discussion). Feature/format changes
with no C++ bridge dependency; the bridge-dependent items (scroll-hold, zebra +
default-colour rework, header menu, rubber-band select, tooltips, presets,
frozen columns, plus the new asic-rs columns) follow in 8b/8c.

- Offline recognition (reverses the round-6 interpretation): the scanner no
  longer fabricates a row for an address with no miner. Only miners that
  actually respond become rows; a responding non-mining miner is "not hashing".
  Removed the MinerRow.offline field, MinerRow::offline(), the "offline" status,
  and the synthetic fixture/fake-scan offline rows. The bench fleet is now 9
  rows (was 10); KPI not-hashing is 1.
- Hashrate conditional format: the scale now bottoms out at 50% (full red at
  <=50% of expected, amber at 75%, green at 100%+), matching "green = 100%,
  red = 50% or lower".
- Hashrate cell format: prefix-only unit, no space, no spaces around the slash
  (e.g. "60.0T/200T", "80.5T/95.0T") instead of "60.0 TH/s / 200 TH/s". The
  combined-cell encoder now builds from formatted parts rather than re-splitting
  a joined string on " / ", so hashrate ("/") and count columns (" / ") each
  keep their own separator.
- Chip temperature: new blue-to-red conditional scale (cool blue at/below 40 C
  to hot red at/above 90 C); the Chip temp column is now a coloured cell.
- Multi-range scan input: each octet may be a range, so "10.1-9.1.1-255" expands
  to the 9 /24s (cartesian per-octet parse). CIDR / full-range / short-range
  forms are preserved.
- KPI header: every card is now a uniform size with its value + label centred.
- Right-click menu: "Fault light on/off" relabelled to "Flash lights on/off"
  (the underlying asic-rs set_fault_light call is unchanged).

Tests updated for the new formats/counts; new unit tests for the 50%-100% ratio
scale, the blue-to-red temp scale, the parts-based combined encoder, and the
per-octet range parser. cargo check --release --tests is clean; CI runs the
release build + tests + headless self-test.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.8.0
2026-09-17 16:26:19 -06:00
noahandClaude Opus 4.8 86e4761316 GUI round 7: code-review remediation (flash safety, unknown-actual colour, credentials, coverage)
Fixes the Critical/High findings from the round-6 review (6 parallel dimension
reviewers + adversarial verify), plus the cheap high-value non-blocking ones.
Full write-up in code-review-report.md.

Firmware-flash safety (the one irreversible, hardware-destroying action):
- Critical: re-triggering an action no longer races the previous one. A flash
  now takes an exclusive lock (new App.flashing): it will not start while any
  scan / refresh / config action is running, and none of those (nor another
  flash) start while it runs. begin_scan, start_action, start_refresh and
  apply_flash_confirm all gate on it, and the context menu offers only the
  read-only "Open in browser" while flashing. This removes the two overlapping-
  network-operation paths that could interrupt/corrupt an upload and brick a
  miner (old.stop() never joined the in-flight upload).
- High: the flash confirm dialog now lists every target IP + model and warns
  loudly when the batch spans more than one model (a single image is rarely
  valid for all), instead of showing only a bare count.

Correctness:
- High: combined columns no longer paint the "-" placeholder solid red when the
  actual is unknown but the expected is known (an idle/paused/booting miner of a
  recognised model looked like a 0% catastrophic failure). fill_rows now gates
  the ratio on the actual being known; unknown-actual falls to the plain,
  uncoloured segment. Was masked by the all-zero-expected fixtures.
- compute_kpis buckets by status() instead of re-deriving the precedence from
  raw flags (KPI header can no longer disagree with a row's status; offline
  folds into not_hashing).

Security:
- credentials file is created at 0600 up front (OpenOptions::mode) and
  re-tightened while still empty, so the plaintext password is never on disk
  under a looser mode (was create-0644 then chmod-0600).
- the "Set scan credentials" dialog no longer pre-fills the saved password in
  plaintext; a blank Password field keeps the stored one.

Robustness / API:
- the fake-scan branches of the config-action / refresh / flash background
  threads now honour the cancel flag (were ignored, so stop() was a no-op).
- MinerRow.online -> offline (default false = online), removing the footgun
  where a struct-update literal that forgot `online: true` read back offline.

Tests (the review's primary focus -- these paths had zero coverage):
- scan.rs fake_mode_background_actions: start_scan offline-row emission,
  start_refresh (Row/Failed/Done), start_firmware_flash_action, and that a
  cancelled run still terminates with Done rather than hanging.
- model.rs: flash_confirmed (exact/case/prefix/whitespace), compute_kpis vs
  status() consistency, Default-is-online.
- new ui::tests: ratio_color/fan_color/tint_toward stops+clamping+blends, and
  the combined-cell unknown-actual and cooling-cell encodings.
- credentials.rs: configured_factory smoke (both branches) + pre-existing-0644
  re-tighten.
- self-test: offline row (.199) spot-check, KPI quick-filter toggle on/off, and
  a numeric (not lexical) IP-sort assertion via model::ip_to_u32.

Deferred (non-blocking, tracked in code-review-report.md): start_* boilerplate
extraction; per-model firmware image-compatibility check; scoping saved
credentials to a chosen backend; a masked password field (needs a wx-rs
TE_PASSWORD binding); flash-specific status text; a segment-encode microbench.

Verified: cargo check --release --tests is clean; unit tests + benches + the
headless self-test run in CI (local release link is OOM-blocked on this
desktop).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.7.0
2026-09-17 14:06:03 -06:00
noahandClaude Opus 4.8 c1234f6df7 GUI round 6 self-test fix: decode combined-column cells in spot-checks
The round-6 combined columns (Hashrate/Hashboards/TotalChips/Cooling) now
store colour-segment-encoded strings in the data view, so the headless
self-test's per-IP spot-checks read raw encoded bytes instead of plain text.
Two checks failed in CI as a result:
- S21 Pro (.149) hashrate read "B95E18<US>60.0 TH/s / 200 TH/s" (colour hex
  prefix + encoded combined text) rather than the plain value.
- S17 (.112) blank-hashrate read an encoded empty cell (invisible control
  chars), so is_empty() was false.

Fix: the spot-check `cell` closure now runs dvlc_get_cell output through
decode_segments_plain (a no-op for plain cells), and the S21 hashrate check
expects the combined "60.0 TH/s / 200 TH/s" form. The "every cell matches the
model" loop already decoded, so it was unaffected.

Verified: cargo check --release --tests is clean; the self-test itself is a
CI gate (local release link is OOM-blocked on this desktop).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.6.0
2026-09-17 13:27:54 -06:00
noahandClaude Opus 4.8 b6afe08a5f GUI round 6: offline detection, combined columns, per-cell colours, paused status, credentials, refresh, firmware flash
Large round implemented via a research workflow (6 parallel readers) then a
file-partitioned implementation workflow (model/scan/bridge/ui/fixtures), all to
one shared contract.

Table / model:
- Offline recognition fix: the scan loop now emits a row for EVERY probed IP -
  non-responders become an "offline" row (new MinerRow.online flag +
  MinerRow::offline()), so unreachable miners show up (offline, red) instead of
  vanishing. Previously only responders got a row.
- Not-hashing rows are now RED (were grey); offline also red.
- New "paused" status (operating_state contains "pause" while not mining):
  its own blue row colour, a blue "Paused" KPI card, and a Kpis.paused count.
- Combined actual/expected columns: Hashrate, Hashboards and TotalChips fold
  their separate Expected* columns into one "actual / expected" cell; the actual
  is drawn on a smooth red->amber->green scale by ratio (100% = green), the
  expected in plain text, the slash black. num_value is carved out so a numeric
  filter never matches a row that only has an expected value.
- Fan column shows every fan's RPM as "rpm/rpm/rpm" (max 3 sig digits, SI-style),
  each fan coloured on its own grey->blue speed scale, slashes black.
- These four columns render via a new wxDataViewCustomRenderer (SegmentRenderer)
  that draws colour runs from a \x1f/\x1e-encoded value; every other column keeps
  the stock renderer and per-row status colour.

Interaction:
- KPI quick-filter: clicking the Hashing / Slow / Not hashing / Paused card
  filters the table to that status (composed with any column filters) and
  highlights the active card; clicking again clears it.
- Removed "Select all" / "Clear selection" from the right-click menu.
- Right-click "Refresh": re-polls the selected miners (asic-rs scan_miner) on a
  background thread and upserts the rows, without a full network scan.
- "Set scan credentials..." button (top-right): stores a username/password
  (~/.config/alkane-toolkit/scan-credentials.txt, 0600) that is fanned out to
  every asic-rs firmware backend via with_firmware_discovery_auth, so scans and
  all right-click actions authenticate. Applied through a new
  credentials::configured_factory() used everywhere MinerFactory::new() was.
- "Flash firmware..." (right-click): asic-rs UpgradeFirmware for the backends
  that support it (stock Antminer + EPIC; every other backend declines and is
  reported as unsupported). File picker (new wxFileDialog binding) + a typed
  "FLASH" confirmation; one miner at a time; cancel only between miners.
- Responsive layout: the KPI and controls rows use wxWrapSizer so they reflow
  when narrow; the min window size drops from 1200x720 to 640x480.

Bridge (wx-rs): SegmentRenderer + dvlc_append_colored_col, wrap_sizer,
bind_click (clickable KPI cards), file_dialog; dropped a vestigial gtk.h include.
Patches 0002/0003 regenerated + round-trip validated.

Portability note: every widget is a stock, natively-backed wxWidgets control
(ports to wxMSW/wxOSX); wxDataViewCtrl is native on Linux/macOS and wx's own
generic (still-correct) drawing on Windows.

Verified: cargo check --release --tests is clean; the release build + tests +
headless self-test run in CI. (Local release link is OOM-blocked on this
desktop; CI is the authoritative build.)

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
2026-09-17 13:09:16 -06:00
noahandClaude Opus 4.8 eb08a9ef97 GUI: per-miner capabilities + live config write path (right-click actions)
Completes the right-click config menu (round-2 item 11): the menu now offers
real actions computed from the intersection of what every selected miner
supports, and applies them to the fleet over the network.

- Capability detection: to_row probes each miner's asic-rs supports_* methods
  (fault light, restart, pause, resume, power limit, tuning) at scan time and
  stores them in MinerRow.caps -- no extra network round-trip.
- Intersection: MinerCaps::intersect_all yields the actions common to the whole
  selection; the context menu offers only those (a mixed mining + idle selection
  shows just fault light + restart; an all-mining selection shows everything).
- Write path: scan::start_config_action runs on a background thread -- it
  re-discovers each selected miner with asic-rs scan_miner(ip), applies the
  setter, and reports per-IP results that drain into the status bar (running
  progress while applying, a pass/fail summary when done). Cancellable, and a
  new action cancels any in-flight one so two never overlap. In fake-scan mode
  it simulates success so the flow is exercisable offline.
- Actions: fault light on/off, pause, resume, restart apply immediately; set
  power limit and set tuning % open a small value-prompt dialog first. Factory
  reset, change password, and firmware upgrade are deliberately not exposed in a
  casual right-click.

Adds the `measurements` dependency (already in the lock, transitive via asic-rs)
for Power::from_watts.

Review fixes (one review pass):
- Restart / pause / resume now require a confirmation dialog (naming the count)
  so a misclick can't stop a producing fleet; fault light stays instant, and
  power/tuning are confirmed by their value entry.
- The action result is held on the status bar for a few ticks so a fleet
  action's outcome (including failures) isn't clobbered by the next timer tick.
- The action loop uses a current-thread tokio runtime (it applies to one miner
  at a time; a worker pool was wasted).
- The value prompt validates before closing: a bad value keeps the dialog and
  the captured selection so it can be corrected.
- The menu's "(N)" counts use the rows-backed selection, matching what is
  actually acted on.

Verified off-screen: capability intersection (mixed vs all-mining selections),
the value-prompt and confirmation nested modals (menu -> dialog) with no crash,
the end-to-end apply flow, 46 unit tests (incl. a caps-intersection test), and
the headless self-test.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.5.0
2026-09-17 11:05:15 -06:00
noahandClaude Opus 4.8 297e21a519 GUI round 3: Excel-style filters, status row colours, icon controls, context menu, live theme
Addresses the round-3 feedback batch:

- Right-click now works: a context menu (open selected in browser, select all,
  clear) pops up from the table. It was never wired before. (The live-config
  intersection over the selection remains a follow-up; no dead menu items.)
- Selection is readable again in both themes: the alt-row/zebra CSS hack (which
  painted row backgrounds over the native highlight, leaving white-on-white) is
  gone; the native selection highlight shows through.
- Row text is coloured by status (green = hashing, amber = slow, grey = not
  hashing), which also separates rows without fragile GTK zebra striping.
- Dark mode: KPI cards and rows follow the system theme (no more white boxes in
  dark mode); a system-colour-changed event repaints them live if the OS toggles
  light/dark while the app is open.
- Play / Pause / Stop are coloured media icons (green triangle / amber bars /
  red square) instead of text buttons.
- Filter panel redesigned Excel-style: a column dropdown at the top selects the
  column; controls adapt to its type. Text columns: a value checklist ("is any
  of") or a predicate (contains / does not contain / begins with / ends with /
  equals). Numeric columns: between / = / != / > / >= / < / <=. Filters AND
  across columns; edits held in a working copy, committed on Apply.
- Auto-rescan label shortened to "Rescan (s):" so the value field has room.

Table: replaced wxDataViewListCtrl with a raw wxDataViewCtrl + a custom
AttrListStore (per-row foreground via GetAttrByRow), which keeps in-table column
drag while enabling status row colours. New wx-rs bridge primitives: wxChoice
dropdown, coloured media buttons (drawn on a transparent bitmap via
wxGraphicsContext), a wxEVT_SYS_COLOUR_CHANGED event, destroy_children, and
call_after (deferred rebuilds). Patches 0002/0003 regenerated + round-trip
validated.

Model: ColFilter { Values, Text{op}, Num{op} } with TextOp/NumOp and
passes_filters -> ColFilter::matches; Column::num_value for numeric filtering.

Review fixes (one review pass):
- Deferred the two rebuilds that destroyed a control from inside its own event
  (filter op-dropdown change; column drag-reorder) via call_after, avoiding a
  use-after-free.
- num_value returns None for a blank-displaying value-type cell (Hashrate,
  Power, ChipTemp, ...), so "ChipTemp <= 50" no longer matches a miner with no
  temperature reading.
- Value checklist maps a checked entry back to its real value by index, so a
  literal "(blank)" value is not confused with an empty cell.

Verified off-screen: 45 unit tests, headless self-test (14 checks), and
interaction screenshots of selection, right-click, the text/numeric filter, the
comparator swap, and dark mode.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.4.0
2026-09-17 09:19:38 -06:00
noahandClaude Opus 4.8 7fac0ce726 CI: make the headless UI self-test a hard gate
The self-test was continue-on-error while the container display path was
unproven. Run 529 confirmed it launches and passes (SELFTEST PASS) once xauth
was installed, so remove continue-on-error: a future regression that breaks GUI
init now fails the build instead of releasing silently.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.3.0
2026-09-16 19:30:17 -06:00
noahandClaude Opus 4.8 8e9c5e7105 GUI + CI: code-review-harness fixes, expanded tests, version-logic tests
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
v0.2.0
2026-09-16 19:22:49 -06:00
noahandClaude Opus 4.8 e5fb63ad52 CI: install libssl-dev + pkg-config for openssl-sys
Run 2 got past the wx build (the bash fix worked) but cargo build --release
failed in openssl-sys 0.9.117: the debian:trixie-slim CI image has no OpenSSL
dev headers, which a transitive asic-rs dependency links. Add an install step
(libssl-dev, pkg-config, ca-certificates) before the build. actions/cache only
saves the wx prefix on job success, so this is the minimal change to get a
green run and persist the ~10-min wx build for subsequent iterations.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
v0.1.0
2026-09-16 19:00:55 -06:00
noahandClaude Opus 4.8 8e65162b07 CI: invoke build-wx via bash; mark shell scripts executable
Run 1 of the release pipeline failed at the "Build wxWidgets" step with
"scripts/build-wx-linux.sh: Permission denied" - the checked-out scripts
carry mode 100644 (no exec bit), so a bare shebang invocation fails. The
compute-version step worked only because it was already called as
"bash ci/compute-version.sh".

- Call the wx build as "bash scripts/build-wx-linux.sh" (interpreter
  explicit; no dependency on the exec bit).
- Set the exec bit in git on all ci/ and scripts/ shell scripts so they
  also run directly for humans and future steps.
- Make the headless UI self-test non-fatal for now (continue-on-error)
  and install dbus-x11 + fonts-dejavu-core alongside xvfb, so a container
  display quirk cannot discard an otherwise-good 10-minute build. The
  gate is re-tightened once it is confirmed green in-container.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
2026-09-16 18:41:31 -06:00
noahandClaude Opus 4.8 a1de684635 CI: build-test-release pipeline on merge to main
Gitea Actions workflow (.gitea/workflows/release.yml) triggered on push to
main: build the GUI against the cached vendored wx, run the 22 unit tests and
two microbenchmarks (view pipeline + cell render, with perf-regression floors),
run the headless UI self-test under xvfb, then package the binary and cut a
release when the change warrants one.

Auto-versioning X.Y.Z (ci/compute-version.sh):
  X - manual (ci/version-x.txt); a bump resets Y.Z
  Y - backend-logic changes (core/discovery/access/flash/artifacts/cli, the
      GUI crate's scan/model/main, build + CI config); resets Z
  Z - UI changes (GUI view, fixtures, the wx-rs widget bridge)
Backend beats UI when both change; docs-only changes cut no release.

Runner: a dedicated act_runner (label alkane-ci) on bfe-mpc-0640 (the Gitea
host) runs jobs in a Debian-trixie image with Rust + GTK3/pcre2 + node.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
2026-09-16 18:27:48 -06:00
noahandClaude Opus 4.8 6955b1e3e3 GUI round 2: selection, scan loop, column management, dark-mode fix
- Fixed "Scan ranges" button label (no longer grows with the range list).
- Dark-mode KPI cards: derive colours from the live system theme luminance
  instead of is_dark(), which mis-detected some GTK dark themes.
- Removed the checkbox column; the table is multi-select (wxDV_MULTIPLE,
  ctrl/shift-click), with the selection tracked by IP so it survives refreshes.
- Replaced Rescan with Play / Pause / Stop; added an editable auto-rescan
  interval (scan-ranges dialog; ~/.config/alkane-toolkit/auto-rescan-secs.txt)
  with a timer-driven countdown scheduler.
- Status bar shows the last-probed IP during a scan and a countdown to the next
  scan when idle (dropped the "scan complete" text).
- Default sort is IP-ascending with a header chevron; IP is column 0, sortable.
- Choose-columns dialog is a wxRearrangeCtrl (check to show; up/down or drag to
  reorder) exposing every metric, including Pool 1/2/3 URL + worker.
- Column drag-reorder moves the whole column with its data on drop.
- Row separator lines (wxDV_ROW_LINES). GTK3 native has no zebra striping; a
  GTK4-forward zebra-CSS hook and pkg-config gtk cflags are in place.

Scan engine: parse each configured pool's (url, worker); Progress carries the
last-probed IP. Model: Column::Sel removed; Pool1/2/3 columns; default IP sort.

wx-rs bridge: multi-select + selection/context-menu events + get_selections,
alternating-row + zebra-CSS hooks, in-place column title, rearrangeable list,
and popup-menu primitives (for the upcoming right-click config menu). Patches
0001-0003 regenerated and round-trip validated.

Review fixes: scan loop no longer sticks "scanning" if a thread ends without
Done (channel-disconnect handling + Stop force-reset); Play no longer starts a
second scan; stale selections pruned on rescan; column apply order/visibility
applied atomically.

Verified: 22 unit tests, a 14-check headless UI self-test, and interaction
checks (multi-select, column rearrange/drag, play/pause/stop, dark mode).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
2026-09-16 17:27:20 -06:00
noahandClaude Opus 4.8 d88df603ba Fix GUI layout/crash bugs; add operator features, tests, and dev loop
Round 1 (bug fixes + test harness):
- Fix table rendering in the top-left corner: the dvlc was never added to the
  sizer; add it (proportion 1) and re-run layout; drop the space-eating stretch.
- Fix abort on dialog buttons: the dialog root sizers passed ALIGN_CENTRE_VERTICAL
  to a vertical sizer, tripping a wx consistency assert whose handler pumps a
  nested event loop, re-entering dispatch while it held the App RefCell borrow.
  Remove the invalid flags; split dispatch into two phases with try_borrow_mut.
- Fix header-click sort mapping when columns are hidden (use display_columns);
  drag-reorder splices back into the full order without dropping hidden columns.
- Add offline fixtures + ALKANE_FAKE_SCAN and ALKANE_SELFTEST modes; scripts
  ui-selftest.sh and ui-screenshot.sh; AGENTS.md documents the dev loop.

Round 2 (operator requests):
- Redesigned color-coded KPI cards; light/dark mode follows the system theme.
- SI-formatted hashrate/power on the KPIs and columns (685 TH/s, 21.3 kW).
- Scan ranges accept CIDR, single IP, and A.B.C.D-E.F.G.H / A.B.C.D-N ranges;
  oversized ranges (> 1,000,000 addresses) are rejected.
- Scan progress bar over the table + status-bar text; out-of-range miners are
  dropped when a scan starts.
- Columns panel exposes every scannable MinerData metric (All / None / Defaults).
- Excel-style filter panel: per-category value checklists, per-column All/None,
  Apply / Reset / Cancel, blank handling.
- Sort direction marker in the header (updated in place); double-click a row
  opens the miner in the default browser.

wx-rs: extend the widget bridge (gauge, in-place column title, sort indicator,
browser launch, system appearance/colours, widget colours, item-activated);
patches 0002/0003 regenerated and round-trip validated.

Verified: 22 unit tests + a 13-check headless UI self-test, plus a live scan
against the bench fleet.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y8h447BPyJR4fioYZvQ9bh
2026-09-16 16:24:05 -06:00
noah fcc70c04f6 M1 primary view: wx-rs patched+extended (Linux/GTK3 build, widget bridge), vendored wx 3.3.3 + asic-rs in thirdparty/, UI-SPEC table+KPIs+dialogs, build scripts 2026-09-15 13:59:55 -06:00
noah a2b65bb47f Vendor asic-rs 0.8.1 (280 crates, offline builds) + wxWidgets 3.3.3 (pristine + patch queue); UI spec + ADR-0003; M1 on asic-rs 2026-09-15 10:41:00 -06:00
noah 6dc77e8eb0 Plan: bench-validated pathway matrix, artifact registry seed (14 sha256s), inventory integration 2026-09-15 09:16:55 -06:00
noah 4fdc49aaca ADRs: recipes-as-data, GUI binding decision; cli as bin target 2026-09-15 09:08:36 -06:00
noah 302c8b6e49 Scaffold: workspace crates + implementation plan draft 2026-09-15 09:06:29 -06:00