Changelog

All notable changes to etch341. The format follows Keep a Changelog; versions follow SemVer with 0.x.y meaning the public API

• on-disk formats may change between minor releases.

Unreleased

Added

• UEFI BIOS Setup explorer (read-only) — browse a BIOS dump’s Setup options as a searchable, human-readable list: label, current value, and the choices behind each variable byte (e.g. “VT-d — Enabled; Disabled/Enabled”). Walks firmware volumes → FFS files → sections (LZMA + EFI/Tiano decompression), parses the IFR Setup-form bytecode and HII string packages, and joins them against the AMI NVRAM (NVAR) store for live values. Options the firmware may hide or lock at runtime are flagged. CLI: etch341 bios settings -i <dump> [--find <substr>]; GUI: the “BIOS explorer” pane (file picker + label filter + list with per-row help/choices tooltips). Landed on both surfaces together. New src/uefi/ module, kept free of internal imports for a planned standalone crate; adds the mu_uefi_decompress (BSD-2-Clause-Patent) and lzma-rs dependencies. Validated end-to-end against a real AMI Aptio (Kaby Lake) image. The Setup write side remains out of scope.

[0.6.0] — 2026-06-04

Stable release of the 0.6.0 line. Folds in everything from 0.6.0-beta.1 below, plus:

Changed

• Hex viewer restyled with byte-category colours and an 8/8 split. Both the hex and ASCII panels now colour each byte by category (cyan = printable, green = whitespace / control, gold = non-ASCII, dim for the blank 0x00 / 0xFF bytes), and each row is split into two 8-byte groups by a dotted divider — making structure and a byte’s offset easier to scan. A ”?” toggle next to the footer expands a colour legend explaining the scheme.

[0.6.0-beta.1] — 2026-05-31

Added

• Side-by-side diff view for a failed verify (GUI). When a SPI or I²C verify finds mismatches, a “View diff in Hex” button opens a side-by-side comparison in the Hex pane: the file you verified against on the left (red), the chip’s read-back on the right (green), showing only the differing regions plus a couple of context lines rather than the whole image. Differing bytes are tinted on both sides; Prev/Next chevrons (and Cmd/Ctrl+G) jump between regions, and bytes on either side are selectable and copyable like the main Hex viewer. The red/green are fixed standard diff colours, independent of the chosen accent.
• diff and verify --diff on the CLI. The same byte-level comparison now has a command-line surface. etch341 diff A B compares two files offline (no hardware), printing only the differing regions as a side-by-side hex dump — red for the left file, green for the right — and exits 1 when they differ / 0 when identical, so it slots into scripts like diff(1)/cmp(1). etch341 verify --diff adds that same region view to a hardware verify (file vs chip read-back) instead of only a count and first differing address. Colour follows the terminal: on for a TTY, off when piped or with NO_COLOR set. The GUI view, diff, and verify --diff share one region-grouping core, so all three highlight identically.
• Compare two files in the GUI. The Hex pane gains a “Compare two files…” button: pick two files and their byte-level diff opens in the same side-by-side view used for a failed verify — the GUI counterpart to the CLI’s diff. Red marks the first file, green the second, with their names shown in the header; Close returns to the hex viewer.
• 46 more SPI NOR chips — the database grows from 70 to 116 entries, adding whole vendor families that were missing: Spansion / Cypress / Infineon S25FL (a networking / industrial / automotive staple), Micron / Numonyx N25Q + MT25Q (FPGA config flash), Microchip SST26VF, ISSI IS25WP (1.8V) plus more IS25LP, XTX XT25F (ubiquitous in ESP / WiFi modules), Zbit ZB25VQ, Boya BY25Q, Macronix MX25R (wide-Vcc low-power) and the 512 Mbit / 1 Gbit MX25L51245G / MX66L1G45G, and EON EN25Q. JEDEC IDs were cross-referenced against published vendor ID tables and are flagged not-yet-silicon-tested in the notes. The JEDEC→voltage map learned the IS25WP (9D70) and MT25QU (20BB) 1.8V families so the GUI Voltage column stays correct.

Changed

• Hex viewer file picker matches the other panes now — the Hex pane’s file selection is wrapped in a titled “File to inspect” GroupBox with the same bordered field the Verify / Write / I²C / OTP panes use, instead of a bare path + Browse row.

Fixed

• Pop-out windows actually re-dock on Wayland now. The 0.5.0 fix hooked on_window_should_close, but with client-side decorations a pop-out’s own title-bar close button calls remove_window() directly and never routes through that callback — so on Wayland the inline log still didn’t come back. The re-dock is now also wired to the title bar’s on_close_window handler (the path the Linux close button actually takes); on_window_should_close stays for the macOS / Windows / compositor-initiated close paths, so both decoration modes are covered. Applies to the activity-log pop-out and the chip-database browser.

0.5.0 — 2026-05-30

First stable release of the 0.5.0 line — now picked up by the Homebrew tap, Scoop bucket, and the in-app update check (the betas weren’t). Folds in everything from 0.5.0-beta.1 and 0.5.0-beta.2 below, plus:

Added

• I²C EEPROMs in the GUI — the full 24Cxx workflow now has a GUI, closing the last CLI↔GUI parity gap. A bus toggle at the top of the sidebar swaps the whole tool set between SPI and I²C; the I²C side gets a chip dropdown (I²C has no JEDEC autodetect, so the chip is picked explicitly) plus Scan / Read / Write / Verify / Erase / Blank check panes that mirror the SPI workflow — same stepper rail, same two-stage arm/confirm on the destructive ops (Write / Erase), same shared Hex viewer + Settings. I²C clock is held at 100 kHz. Scan lists the ACKing 7-bit addresses; the chip choice persists as you move between panes.
• In-pane result lines for every operation (both buses) — each op now reports its outcome as a coloured line in the pane itself, not only in the activity log: a green ✓ on success (e.g. “Read 256 bytes from 24C02 → …”, “Wrote … (verified)”, “Chip matches the file”, ”… is blank — all 0xFF”) and a red ✗ on failure — hardware errors, verify mismatches, and blocked clicks (“Pick a chip first”, “Pick an input file first”). Detect / Status / Security-register reads keep their result card on success and surface the red ✗ line on error. The line hugs its text, wraps a long path, and clears on navigation so it stays scoped to the pane that produced it.

Changed

• Programmer dispatch layer over the CH341 backend (internal). The high-level SPI / I²C ops were already generic over the SpiTransport / I2cTransport traits; the concrete Ch341::open* call sites now go through a Programmer enum that owns the device and forwards the transport calls. No behaviour change — it’s the seam for adding a second USB bridge (a new enum variant backed by its own module implementing the two traits) without touching ops / spi / i2c.

Fixed

• Pop-out activity log now re-docks on Wayland. Closing the detached log window didn’t restore the inline log on Wayland (Ubuntu’s default session). The re-dock was tied to the log window’s entity teardown — which gpui runs synchronously on macOS / Windows but defers to a later event-loop turn on Wayland, so on an idle app it never fired and the inline log stayed gone. (The earlier 0.5.0-beta.2 fix tested on X11, where the window is torn down promptly, so it looked complete.) The re-dock now hooks the window’s close request (on_window_should_close), which fires synchronously on every backend — X11 on WM_DELETE_WINDOW, Wayland on xdg_toplevel::Close — so it no longer depends on entity-drop timing. The same fix is applied to the chip-database browser window, which shared the bug (its stale handle would otherwise block reopening the browser on Wayland). (#1)

[0.5.0-beta.2] — 2026-05-29

Added

• Chip-database browser in the GUI — a new window (Detect → “Browse chip database”) that lists every bundled SPI + I²C chip with a vendor dropdown, a live name/JEDEC search, and a colour-coded Voltage column. SPI parts show their single rail (green 3.3V, amber 2.3V, red 1.8V), derived from the JEDEC manufacturer/family byte so it can’t drift from the id that determines it; I²C 24Cxx show the family’s wide 1.8–5.5V range — they’re the 5V-capable parts, commonly run at 3.3V or 5V on the CH341A. The CLI chips table gains a matching VOLT column on both buses, keeping the two surfaces in parity.
• AT25DN512C in the chip database (Adesto/Atmel 512 Kbit, 2.3V, dual-read, JEDEC 1F6501). This part exposes no SFDP table, so it needs an explicit database entry to be recognized by name — silicon-confirmed against a real chip.
• Eight more Adesto / Atmel AT25 parts rounding out the family: AT25SF321 (4 Mbit); the AT25DF series AT25DF041A / 081A / 161 / 321A / 641 (4–64 Mbit, 3.3V); and the 1.8V AT25SL321 / AT25SL128A (32 / 128 Mbit). JEDEC IDs were cross-referenced across independent published vendor ID tables (datasheet-sourced, not yet silicon-tested — detect flags them as such in the notes). The 1.8V SL parts need a 1.8V-capable programmer. SPI NOR database is now 70 entries.

Changed

• GUI read dumps get human-readable filenames — etch341-read-<date>_<time>.bin in local time (e.g. etch341-read-2026-05-29_14-03-07.bin) instead of the raw seconds-since-epoch suffix. Sorts chronologically, uses hyphens (not colons) so it’s legal on Windows and tidy in Finder.

Fixed

• I²C writes no longer always time out. Every I²C write / erase aborted with Error: Timeout because the post-page “wait for write cycle to finish” step polled the chip through a probe that infers presence from a data byte — and on a blank chip (or any page whose next byte is 0xFF) that read 0xFF and concluded “never ready”. The CH341 never exposes the I²C ACK bit, so there’s nothing to poll; writes now wait out the worst-case datasheet write-cycle time instead.
• I²C reads no longer corrupt past ~30 bytes. A multi-byte read was clocked with a single IN | n substream, which makes the CH341 ACK every byte including the last — so the read was never terminated with a NACK and the chip stayed mid-transfer, shifting the bytes of the next chunk. Reads now clock one IN | 1 per byte (master ACKs each) and a final bare IN (master NACKs the last), matching how working CH341 I²C drivers do it; the per-chunk size dropped from 31 to 20 so the longer request stream still fits one 32-byte CH341 packet. With both this and the write-cycle fix, the full I²C read / write / erase / verify / blank-check cycle is byte-exact on real silicon (AT24C02, 100 kHz and 20 kHz) — the first end-to-end silicon validation of the I²C path.
• i2c scan now explains the blank-chip blind spot. A blank EEPROM (all 0xFF) is indistinguishable from an empty bus on the CH341 (no ACK-bit readback), so scan can’t list it; the empty-result message now says so and points at --chip.
• Pop-out activity log now re-docks on Linux/X11 when its window is closed. The re-dock was wired through on_window_should_close, which gpui doesn’t route from the X11 window manager’s close button; it’s now tied to the log window’s entity teardown, which fires however the window closed. (#1)
• Activity log stays pinned to the newest line when the log pane is resized. Dragging the splitter changed the viewport height without re-requesting the paint-time scroll-to-bottom, so the view drifted off the latest entry. (#2)

[0.5.0-beta.1] — 2026-05-28

First beta of the 0.5.0 line — a big GUI pass plus OTP support. Pre-release: not picked up by the Homebrew tap / Scoop bucket (they track stable) or the in-app update check. Grab it from the Releases page if you want to try it.

Added

• OTP / security register access (etch341 otp read / otp erase / otp write + a Security-regs pane in the GUI). Reads, erases, and programs the three 256-byte security registers carried by the Winbond W25Q / GigaDevice GD25Q families (opcodes 0x48 / 0x44 / 0x42). These commonly hold serial numbers, MAC addresses, or vendor keys.
  • Read dumps all three as offset / hex / ASCII; blank (all-0xFF) registers collapse to a one-line note. The GUI card’s Copy button yields the same text as the CLI.
  • Erase / write target one register. Both are read-back verified, and write does not erase first (programs only clear bits — erase the register first for a clean write). The CLI gates them behind --yes; the GUI uses the same two-stage arm/confirm as the Erase / Write panes.
  • Erase + reprogram stay repeatable: etch341 never sets the one-time lock bits (LB in SR2), so locking a register closed is a deliberate non-goal.
  • GUI parity note: the GUI writes from offset 0 of the selected register; use the CLI’s otp write --start for a partial write at an offset.
  • W25Q / GD25Q 0x48 convention only — Macronix’s single security register (opcode 0x2B) isn’t covered.
• Selectable accent color (Settings → Appearance). Pick from eight curated swatches; the whole UI recolors live — buttons, selections, toggles, radio dots, sidebar — and the choice persists. Button labels switch between dark and light per accent so they stay legible.
• New-version check — on launch etch341 checks the project’s GitHub releases and, if a newer stable version exists, paints a dot on the ⚙ Settings sidebar item. Settings → Updates has the status, a “View release” link, a “Check now” button, and an on/off toggle. Detection only — it never downloads or installs.
• Pop-out activity log — a ⧉ chip detaches the log into its own window that stays live (and follows new lines); a × chip clears it.
• Activity-log timezone toggle (Settings → Log timestamps) — render timestamps in local time or UTC; storage stays UTC.
• Hex viewer font sizing — Cmd/Ctrl + = / - / 0 zoom the hex and strings views independently, with size controls in Settings.
• The installed version now shows next to the Settings heading.

Changed

• GUI restyle — Settings, the OTP pane, and the Write / Verify panes now group their controls in outlined cards with bordered, input-style file fields and capped content width. CTA buttons are a notch smaller. Consistent visual language across the app.

[0.4.1] — 2026-05-27

Emergency point release. The 750 kHz -s default in 0.4.0 (and every prior version) is ~2× the spec’d max for every chip in the 24Cxx family and was confirmed to brick an M24C02-R mid-write during bring-up. Upgrade if you use any i2c * subcommand.

Changed

• I²C clock defaults to 100 kHz (was 750 kHz, which is out-of-spec for every chip in the 24Cxx family). Explicit -s 750 is now rejected for I²C ops with a message pointing at the 400 kHz datasheet ceiling. SPI ops are unaffected — the -s default for SPI is still 750 kHz. Background: an M24C02-R was bricked during 2026-05 bring-up because the global 750 kHz default exceeded its 400 kHz max mid-write and the chip never returned to ready.

Documented

• I²C clock + over-clock failure mode in Usage → I²C. Adds a Clock speed section explaining the default + ceiling, and an Error: Timeout mid-write, chip goes silent on retry entry under Troubleshooting. Plus an in-circuit write warning covering the bus-contention failure mode seen on a Cisco C921 during the same bring-up.

[0.4.0] — 2026-05-27

Added

• SFDP support (etch341 sfdp + folded into the Detect pane). Reads JESD216 Serial Flash Discoverable Parameters from chips that carry it (most SPI NOR since ~2011) and decodes the JEDEC Basic Flash Parameter Table: total size, page size, address width, 4K erase opcode, and up to four erase types. Detect now always reads SFDP after JEDEC and shows both a chip-info card (JEDEC + chip name + source + size) and an SFDP card (raw decoded table) inline in the pane.
• SFDP fallback in resolve_chip + GUI ops — when JEDEC isn’t in chips.toml, the read / write / erase / verify / blank-check / Detect paths now synthesize a Chip from SFDP (name like "C22011 (SFDP)", derived size / page / sector) instead of hard-failing with ChipNotRecognized. Explicit --chip <NAME> still wins over both lookups. Silicon-validated against an uncatalogued MX25L1006E: SFDP-synthesized parameters produced the byte-identical read as the curated DB entry.
• Settings → Read save location picker. Read pane dumps used to land in $HOME unconditionally; now configurable via a folder picker in Settings with a free-form display of the current value. Persisted in prefs.toml.
• Status registers Copy button and SFDP Copy button — each card has a small “Copy” pill that puts the plain-text decoded view on the clipboard for paste-into-bug-report / share workflows. GPUI doesn’t support cursor-based text selection in rendered text, so copy-all is the practical substitute.
• Op-pane scrolling — every pane’s content now scrolls within the resizable pane area instead of clipping at the bottom. The Settings pane needed this for the WINDOW + READ SAVE LOCATION
  • PREFERENCES FILE sections to stay reachable on shorter windows; the SFDP card likewise can extend past one viewport.
• Output cards in Status registers and Detect / SFDP panes set “operation result” apart from the pane’s heading + body + button stack with a subtle glass background + hairline border.
• Screenshot of the GUI in docs/public/etch341.png, embedded in usage/gui.md and the project README so visitors landing on either can see what the running app looks like before reaching for the installer.

Changed

• Progress indicator in the session header now renders as an accent-blue pill (background tint + accent-blue text) while an op is running instead of the previous near-invisible tertiary-gray text. Easy to glance and tell whether a Read / Write is in progress; falls back to the quiet “idle” treatment when nothing’s running.
• Em-dashes removed from every user-visible string across the GUI + CLI (panel bodies, armed warnings, log lines, error messages, SPI speed labels, Find input placeholder). Replaced with :, ., or , based on the role each em-dash was playing. Code comments left alone.
• Detect pane body rewritten in plainer language and now clarifies that Detect is optional (every op auto-detects on its own). “Other steps” matches the stepper sidebar’s vocabulary.
• Read pane body trimmed of “runs in the background, watch the log” filler. References Settings → Read save location.

Fixed

• SFDP parser: parse_bfpt was reading “DWORD 11” (page size
  • program/erase timings) regardless of the BFPT’s actual length. Older pre-JESD216A chips like the MX25L1006E ship a 9-DWORD BFPT; reading past it returned garbage that decoded page_size as 32768 on this chip. Now gated on length_dwords >= 11 with a sanity cap to [256, 4096]. New nine_dword_bfpt_defaults_page_to_256 regression test exercises the exact MX25L1006E pattern.
• Activity log typed-text-invisible bug was fixed in v0.3.1; this release adds the matching find-input-placeholder visibility fix that slipped through then.

Documented

• Detect / SFDP consolidation — the standalone “SFDP” sidebar pane is gone; the Detect pane now does both jobs (JEDEC + DB lookup + SFDP read + decoded table). Updated usage/gui.md to describe the combined layout.

[0.3.1] — 2026-05-26

Fixed

• Windows GUI launched a console window alongside the GUI on every double-click. Rust’s default console subsystem allocates a fresh console for any binary launched from Explorer; release gui builds on Windows now set windows_subsystem = "windows" so no console appears. CLI subcommands invoked from cmd / PowerShell still surface stdout via an AttachConsole(ATTACH_PARENT_PROCESS) fallback (output prints after the next prompt redraws, which is usable but ugly — a dedicated etch341-cli.exe console-subsystem sibling is the cleaner fix and remains deferred until Windows CLI usage justifies it).
• Hex pane rendered “ghostly” on Windows + Linux because every fixed-width region called font_family("Menlo") directly. Menlo is macOS-only; the Windows / Linux fallback was a thin substitute that read poorly against the dark theme. New theme::MONO_FONT constant picks Menlo on macOS, Consolas on Windows, monospace on Linux; every callsite in log.rs + panes.rs reads from it.
• Find input text was invisible (typed characters didn’t appear) — the wrapper div carried text_color(theme::bench_black()) from when gpui-component’s Input had a white background, but after we forced Theme::change(ThemeMode::Dark) in v0.3.0 the Input went dark-on-dark. Drop the override; let the theme drive both sides.
• Ctrl+C / Ctrl+F / Ctrl+G silently no-op’d on Windows + Linux. GPUI treats cmd- and ctrl- as distinct chords; the existing bindings were cmd- only. Both forms now bind, each gated to its native platform — cmd-* on macOS only, ctrl-* on Windows + Linux only — so neither OS shadows the other’s convention.

Documentation

• Docs header gains a ”← packetThrower” pill linking back to https://packetthrower.github.io/ for visitors who want to bounce out to the rest of the project index. Sits next to the existing “Docs” pill in the header.
• chipdb.rs module-level comment said the chip database is read at runtime; it’s actually compiled in via include_str! at build time. Comment-only fix.

[0.3.0] — 2026-05-26

Added

• Cross-platform window titlebar — gpui_component::TitleBar widget renders min/max/close buttons + title text on Windows and Linux (where appears_transparent hides the native chrome and there was previously nothing drawing the controls). On macOS the widget paints under the native traffic-light overlay. WindowDecorations::Client pairs with it to keep KDE Plasma’s KWin from stacking a server-side titlebar on top.
• Windows .exe icon — build.rs + resources/icon.rc embed resources/icons/icon.ico into the PE resource section so Explorer / taskbar / Alt-Tab / Start menu show the etch341 icon on the binary itself.
• Settings → “Open folder” button next to the prefs path opens the prefs.toml directory in the OS file manager (open/ explorer/xdg-open).
• Settings → “Restore window position on startup” toggle. Off by default; when on, the window’s last position + size are saved on close and restored on next launch.
• Stepper-style sidebar for the canonical workflow (Detect → Read → Erase → Write → Verify) — diagnostic tools (Blank check, Hex viewer) and Settings sit below a thin divider as flat rows. Free-jump preserved: click any step at any time.
• Macronix MX25L1006E / 2006E / 4006E (1/2/4 Mbit) chip-DB entries. The MX25L1006E (JEDEC C22011) is silicon-validated against a Dell-OEM AMD Samoa GPU BIOS.
• W25Q80DV silicon validation — clean detect + double-read SHA match across two physical chips (Dell Precision 3520 EC firmware).
• W25Q128JV silicon validation — clean detect against a consolidated Intel BIOS + ME + GbE flash on a Dell laptop motherboard.
• Status register dump — etch341 sr (and a matching GUI “Status regs” pane in the sidebar’s tools section) reads SR1 / SR2 / SR3 and decodes the standard bit fields (WIP / WEL / BP[2:0] / TB / SEC-or-BP3 / SRP0 / SRP1 / QE / LB / CMP / SUS / ADP / WPS / DRV / HOLD-RST). Diagnoses the two most common silent-failure modes: block-protect bits set (writes silently fail) and QE clear (quad opcodes NACK). SR1 is universal across SPI NOR vendors; SR2 / SR3 follow the W25Q-family convention and show “didn’t respond” on chips that don’t implement them. Silicon-validated on a Macronix MX25U4033E.
• read -o - dumps the chip to stdout instead of a file — enables etch341 read -o - | sha256sum and similar pipe idioms. Success summary lines suppressed in stdout mode so they don’t interleave with the binary data; errors still surface on stderr.
• Linux .deb / .rpm / .pkg.tar.zst postinstall hook — reloads udev rules and re-triggers attached USB devices immediately on install so the bundled 99-ch341a.rules takes effect without an unplug-replug or manual udevadm control --reload. The single packaging/linux/etch341-postinstall.sh script is wired into all three formats (dpkg-deb -R/-b injection for .deb; fpm’s --after-install for .rpm + .pacman). Chroot / container installs no-op cleanly when /run/udev is absent.

Changed

• Settings pane no longer shows the activity log; the log was designed to surface op progress and steals real estate when you’re configuring rather than operating.
• Activity log fills the full width of its resizable panel (was hugging the longest line and leaving a margin on the right); auto-scrolls reliably on every new line (ScrollHandle::scroll_to_bottom runs at paint time, so the new line is in the height by the time the bottom is computed); min-height raised from 80px to 120px so the splitter can’t shrink the pane to barely-readable.
• Op-pane buttons tightened from px_4 / py_2 / default font to px_3 / py_1p5 / 13px / min_w 96 — the bulky CTAs read as out of proportion against the rest of the chrome. Done in one helper (styled_button) so future tweaks land in a single place.
• Detect button label “Refresh” → “Detect chip” — matches the pane title and reads as the intended action without the user needing to read the body paragraph first.
• Sidebar narrower (220 → 180 px) and content padded to match the right-pane gutter; the workflow’s short labels left ~80 px of slack at the wider size.
• prefs.toml path on Windows now resolves to %APPDATA%\etch341\prefs.toml (was hardcoded to $HOME, which Windows doesn’t set — prefs never loaded or saved on Windows). macOS / Linux still use $HOME/.config/etch341/prefs.toml as before; no existing prefs migration needed.
• Op-pane shell factored into a shared op_pane(heading, body) helper; Erase + Write share armed_warning() and armable_button() (was duplicated). Six panes refactored, net −16 lines, future “what an op pane looks like” tweaks land in one place.

Fixed

• I²C page-write race that garbled writes during silicon bring-up against an AT24C02N. wait_ready was racing the chip’s internal write cycle; the fix sleeps through the datasheet tWR first (silicon is guaranteed busy during that window), then ACK-polls with a 50 ms tail for cycle stretching at voltage/temperature corners. Protocol matches the canonical embedded-hal 24Cxx drivers.
• GNOME Wayland resize handles were a 1-pixel diagonal sliver because the compositor refuses xdg-decoration and the gpui default client inset is 0. window.set_client_inset(px(10)) widens the hit-test margin. No-op on macOS / Windows / X11.

Documented

• Stuck-high MISO on large SPI reads filed in TODO after a W25Q128JVSQ test where read 1 of a fresh detect → read sequence returned ~3241 stale-0xFF bytes scattered across a 4 MB range; reads 2/3/4 (including one at 750 kHz) all matched. Failure rate scales with USB-packet count, so smaller chips passed first-try. Workaround today: dump twice and cmp.

0.2.0 — 2026-05-26

Added

• Native installers for every supported platform (replaces v0.1.0’s bare-binary tarballs):
  • macOS: .dmg + .app.zip
  • Windows: NSIS -setup.exe + MSI (via cargo-wix, stable tags only) + portable bare-.exe zip
  • Linux: .deb + .AppImage + .rpm + .pkg.tar.zst
• App icon — wireframe SOIC-8 in the Baudrun + PortFinder family palette (silver outline, gold gull-wing leads, silver pin-1 dot on a navy squircle).
• CLI inspect commands: chips list/find, strings -i <file>, search <pattern> -i <file>.
• --dry-run now actually works (was advertised in v0.1.0 but silently ignored). Validates chip name + range + input file without opening the CH341A.
• Chip database expanded from 24 to 58 entries: W25X legacy line, W25Q*JW 1.8V, MX25U sizes, GD25LQ 1.8V, GD25Q256, EON EN25QH, PUYA P25Q, ISSI IS25LP.
• Docs site at https://packetthrower.github.io/etch341/.

Fixed

• macOS double-click on the v0.1.0 binary opened a Terminal window instead of launching the GUI. v0.2.0 ships a real .app bundle so Finder routes it through LaunchServices, no terminal.
• README’s “No kernel drivers required” lede was wrong on Windows; WinUSB is a kernel driver that needs a one-time Zadig bind. Corrected to spell out the per-platform reality.
• CI clippy run failed on rustc 1.95 (newer lints than the local toolchain). All 14 lints resolved.

Documented

• Minimum OS versions per platform: macOS 11+, Windows 10 21H2+, Ubuntu 22.04+, Debian 12+.
• I²C path is implemented and unit-tested but not yet validated against silicon — explicit warning in both the feature table and the I²C usage section.

0.1.0 — 2026-05-25

Initial release.

• Cross-platform CLI/GUI flash programmer for the CH341A USB SPI/I²C interface, packaged as a single binary (no subcommand → GUI; subcommand → CLI).
• Full SPI workflow: detect, read, erase (full + range), write (with erase + verify), verify, blank-check. 4-byte addressing for chips > 16 MB.
• I²C scaffolding: protocol layer, CH341A transport, CLI subcommands, 24Cxx chip database. Mock-tested but awaits real-EEPROM validation.
• GPUI-based desktop GUI with hex viewer, strings extraction, find / jump-to-offset / byte selection, live progress, persistent prefs.
• Chip database with 24 entries across W25Q, MX25L, MX25U, GD25Q, SST25VF, AT25SF.
• CI/CD across Linux (amd64 + arm64), macOS (arm64 + cross-compiled amd64), Windows (amd64 + arm64) via GitHub Actions.
• Hardware-validated against Macronix MX25U4033E (1.8V) on a GTX 1060 VBIOS chip — full erase → write → verify cycle, byte-identical SHA-256 match.