blob-scan.txt

tools/blob-scan.py: a passing run, a run without the vendor checkouts, and the failure modes (tasks 1.2, 1.4)
========================================================================================================
Run 2026-09-22 on solomon, in the worktree for every-blob-is-built-from-source-or-named,
with the vendor checkouts reachable at .build/k230_linux_sdk (kendryte/k230_linux_sdk @ 1104236)
and repo/ (Xinyuan-LilyGo/T-Display-K230_canmv_rt). Output is pasted from the runs, not typed.

1. The tree as committed, plus both vendor checkouts
----------------------------------------------------
$ ./tools/blob-scan.py; echo rc=$?
blob-scan: /mnt/MediaVolume/home/jadams/src/gitlab.daringbit.com/josh/k230-every-blob-is-built-from-source-or-named
  walked vendor checkout (sdk) at .build/k230_linux_sdk/: 9171 files
  walked vendor checkout (lilygo) at repo/: 36055 files
  walked 45422 files: 3597 binary (17 by file row, 3580 by group row); 54 inventory rows verified on disk, 7 not present here; 3 nix pins checked
blob-scan: ok -- every binary is accounted for
rc=0

2. The committed tree only -- what CI sees, where no vendor checkout exists
---------------------------------------------------------------------------
$ ./tools/blob-scan.py --no-vendor; echo rc=$?
blob-scan: /mnt/MediaVolume/home/jadams/src/gitlab.daringbit.com/josh/k230-every-blob-is-built-from-source-or-named (committed tree only)
  walked 196 files: 9 binary (0 by file row, 9 by group row); 19 inventory rows verified on disk, 42 not present here; 3 nix pins checked
blob-scan: ok -- every binary is accounted for
rc=0

3. Every binary in the committed tree, and the row that covers it
------------------------------------------------------------------
$ ./tools/blob-scan.py --no-vendor -v | grep -E '^  (binary|pin|absent)'
  binary  docs/evidence/panel-photos/01-urandom-noise.jpg  <- group row 629 (DATA) docs/evidence/panel-photos/*.{jpg,png}
  binary  docs/evidence/panel-photos/02-all-black.jpg  <- group row 629 (DATA) docs/evidence/panel-photos/*.{jpg,png}
  binary  docs/evidence/panel-photos/03-all-white.jpg  <- group row 629 (DATA) docs/evidence/panel-photos/*.{jpg,png}
  binary  docs/evidence/panel-photos/04-normal-mode-fullscreen-noise.jpg  <- group row 629 (DATA) docs/evidence/panel-photos/*.{jpg,png}
  binary  docs/evidence/panel-photos/05-console-on-panel.jpg  <- group row 629 (DATA) docs/evidence/panel-photos/*.{jpg,png}
  binary  docs/evidence/panel-photos/06-sheared-console-31.68mhz.jpg  <- group row 629 (DATA) docs/evidence/panel-photos/*.{jpg,png}
  binary  docs/evidence/panel-photos/07-rectified-panel-space.png  <- group row 629 (DATA) docs/evidence/panel-photos/*.{jpg,png}
  binary  docs/evidence/panel-photos/08-four-fiducials-oblique.png  <- group row 629 (DATA) docs/evidence/panel-photos/*.{jpg,png}
  binary  site/src/assets/t-display-k230-board.png  <- group row 630 (DATA) site/src/assets/*.png
  absent  firmware/stage1/fn_u-boot-spl.bin  (row 622; not on this machine)
  absent  firmware/stage1/fn_ug_u-boot.bin  (row 623; not on this machine)
  absent  firmware/stage1/env.env  (row 624; not on this machine)
  absent  firmware/stage1/fw_jump.bin  (row 625; not on this machine)
  absent  firmware/stage1/fw_jump_add_uboot_head.bin  (row 626; not on this machine)
  pin     nix/k230-sdk-src.nix: sha256-P3XkeyJPkpe/h0oHAHCiz8a0VXofw1zsQ3YSa6UcG8w=  <- row 638 (SRC) src:kendryte/k230_linux_sdk@1104236,sparse
  pin     nix/uboot-k230.nix: sha256-ULRIKlBbwoG6hHDDmaPCbhReKbI1ALw1xQ3r1/pGvfg=  <- row 636 (SRC) src:u-boot-2022.10.tar.bz2
  pin     nix/opensbi-k230.nix: sha256-T8ZeAzjM9aeTXitjE7s+m+jjGGtDo2jK1qO5EuKiVLU=  <- row 637 (SRC) src:riscv-software-src/opensbi@v1.4

4. The failure modes, each planted in a scratch copy and each reported
-----------------------------------------------------------------------
$ ./tools/blob-scan.py --self-test; echo rc=$?
--- baseline: the inventory and the nix files alone
    scan exit status: 0  (passed, as expected)
    blob-scan: /tmp/blob-scan-selftest-igwoqwgc (committed tree only)
    blob-scan: ok -- every binary is accounted for
--- a binary with no inventory row is planted at firmware/planted.bin
    scan exit status: 1  (failed, as expected)
    blob-scan: /tmp/blob-scan-selftest-igwoqwgc (committed tree only)
    error: binary with no inventory row: firmware/planted.bin
    blob-scan: FAIL (1 problem(s))
--- nix/stage1.nix pins a release hash the inventory does not list
    scan exit status: 1  (failed, as expected)
    blob-scan: /tmp/blob-scan-selftest-igwoqwgc (committed tree only)
    error: nix/stage1.nix pins sha256-0EBVCBOWyeHBJ8LHZLw3PnwuSFYlDgiFCwb32Qm+i+o=, which the inventory does not list. Add a `src:` row if it is source, a `release:` or `dl:` row if it is a binary
    blob-scan: FAIL (1 problem(s))
--- a listed blob (firmware/stage1/env.env) has different bytes than its row
    scan exit status: 1  (failed, as expected)
    blob-scan: /tmp/blob-scan-selftest-igwoqwgc (committed tree only)
    error: firmware/stage1/env.env: sha256 is d5c06a6a6248ac25733244fb3c2ad566bb05da800dedd86753fed304bd0a5a36, inventory row 624 records cf108755a3cf3a8070a2f0ded84af672108e2301dfd58e5c2aa6a7555dbbd352
    blob-scan: FAIL (1 problem(s))

self-test: ok -- every planted fault was reported
rc=0

The self-test's own exit status is 0 when every planted fault made the scan
exit non-zero and name the fault; the three 'scan exit status: 1' lines above
are the scanner refusing an unlisted binary, an unlisted release hash, and a
listed blob whose bytes moved.

5. What the first run found, before the inventory was corrected
----------------------------------------------------------------
The scanner's first run over the vendor checkouts, on 2026-09-22, reported 3 080
binaries with no row and four group rows whose counts were wrong. The inventory
as first committed had counted the LilyGO clone's opencv archives as 25 (there
are 15), its kmodel directory as 87 files (401, most of them sample images),
the examples glob one directory too shallow, and had never given the C8
k230_priv_gzip copies MANIFEST rows. Two of its own stage-1 rows were stale:
fn_ug_u-boot.bin at 0f8feb74... (the build before the CM-byte fix; the file
the last image carried is c0fb8d95...) and env.env at f522ba13... (the SDK
default; the file the last image carried is cf108755...). All of it is
corrected in the MANIFEST the runs above were made against, and the
correction is the argument for the scanner.