igneum-network/igneum
1
0
Fork 0
Code Issues Pull requests Releases Activity

release-0.3.20 plan: section 34, every artefact on c4459193 with shas, the workers with the Arc fix, the hive package and its smoke, the Windows inputs and run, the staging plan

Browse source 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
igneum-labs 2026-10-07 13:00:53 +00:00
parent ec06347123
commit 650cc59669
1 changed files with 14 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

14
docs/plans/release-0.3.20.md
View file

@ -449,3 +449,17 @@ Speculative builds on the candidate from the ship worktree, box only, sequential
**c4459193 seed pair held (the build-server lane, 13:48 BST):** zigbuild 2.35, 374 s, rc 0; igneumd 56,304,592 B sha256 4a2d8a8db0911a9236264ba7ef9ca8891e8a9df53f0892a695e176323cfaf415 (string c4459193, the N13 line, needs GLIBC_2.34 against the seed ceiling 2.35); igneum-miner 10,235,664 B sha256 18099e976b844653d2ccc7a2d37cef785d7537cbc3df042b073c7fcd8dd31fd0. Held at scratch bs0320/d-seed/x86_64-unknown-linux-gnu/release/. The c4459193 pair set is complete: hands a80ed39c / 70a5180f, seed 4a2d8a8d / 18099e97. The Windows cross of c4459193 from 13:48 BST, then 6a3432a3's, then the PC 2 gate job.
**Slip (the fleet, 12:5xZ; 13:5x BST):** the cases rerun's target starts from pool-1's kept copy 22,000 blocks behind the tip and this line walks the headers first (the morning's poison pod: 50 minutes on the same copy), so c20-1 syncs about 14:50 BST and CASES END lands about 16:10 BST, forty minutes past the checkpoint; no faster path. The rest holds: c19-1's wipe on the class (synced about 15:17 BST), p12-vast synced about 14:45 with the 12 GB line about 15:00 to 15:10. Put to main: (a) hold the pin to CASES END and publish about 16:15 BST, or (b) cut about 15:35 on the wipe line, the kept reads, the restart and the 12 GB line with the cases as a confirmation before the sweep's first box (the diff argument: the class byte is the only touch; the old hub took 235 byte-5 headers in the void run). The shipper's read: (a). The pods run on unchanged.
## 34. The 0.3.20 artefacts on the pin c4459193 (14:0x BST), staged for a one-step publish
**Main (13:5x BST):** (a) for miner-reliability: 0.3.21 takes both halves; 0.3.20's trees stay closed; its app half rebases onto the published 0.3.20 tree right after the publish. The report states what a 0.3.20 user still meets from MF-1 to MF-10 and what the pin removes (N13, N12 and the shutdown poll, the template latency memo).
**The shipper's box builds (scratch r0320, the vendor worktree at c4459193, app 4c89372a, igneum-pow 8c728ca3's tree):** fleet-native node (2.39) igneumd a8d08da5 57,628,128 B / igneum-miner 474273ce 10,214,648 B; hive class (2.31) igneumd b1c4e841 56,305,232 B / igneum-miner 4050c255 10,236,448 B; every igneumd carries c4459193 twice, no stub string, the N13 line; Windows cross igneumd.exe 49502cc7 52,569,088 B / igneum-miner.exe b4871b5d 11,219,968 B; the app's Windows exes igneum-app.exe b73120f1 3,956,224 B, igneum-ota-sign.exe 5355a5b6, igneum-prove-verify.exe 15998eab; the app's Linux binary igneum-app dd22a4ae 3,389,776 B. igneum-pow tests on the box with the amended packs: 61 + 7 + 4 + 19 + 2 + 7, 0 failed. The Mac, under the lock: igneumd b306baba 48,072,304 B and igneum-miner deb4d263 (c4459193 twice, the N13 line), the DMG Igneum-Miner-0.3.20.dmg 44,467,804 B sha256 73796c5febc20506 (build 202610071249, hdiutil VALID, the prover pair in).
**The workers with the Arc fix (built on the box from this tree's proto-opencl at 9088293a's content):** Linux 2.31 for HiveOS: igneum-worker-cuda c08e4694 6,761,464 B, igneum-worker-opencl 56cbe32f 312,304 B; Linux 2.35: cuda ed3dfe58, opencl 8ef81d8d; Windows igneum-worker-opencl.exe 618a2b10 530,432 B (mingw on the box, static, imports KERNEL32 and msvcrt only, the rewrite text present), in place of the 01:33 build (479,232 B, kept as igneum-worker-opencl-0133-old.exe in scratch); igneum-worker-cuda.exe unchanged since 0.3.19 (its sources untouched). Asked of the Intel lane: how bench d's exe (3c62470f) was built and whether it is the same source; if its exe differs the installer is rebuilt on it.
**HiveOS package:** igneum-hive-0.3.20.tar.gz 27 MB sha256 d9dd12dfe900bf9f (the 2.31 node pair, the 2.31 workers, version.txt naming c4459193 and 8c728ca3; no AppleDouble entries). Smoke in ubuntu:20.04 on the box (glibc 2.31): igneumd answers "igneumd 2.1.0" with c4459193 in the binary, igneum-miner its usage, both workers load and answer.
**Windows inputs and installer:** payload-inputs.zip published 13:59 BST (igneumd.exe 49502cc7, igneum-miner.exe b4871b5d, the three mingw runtime DLLs, igneum-worker-cuda.exe with nvrtc64_120_0 and nvrtc-builtins64_128, igneum-worker-opencl.exe 618a2b10, the 0.3.17 Linux prover pair 71bc2438 / 263bf4ce, which stands because the pre-push prover-pair-check read c4459193's evm-types equal to the exec pin's; IGNEUM_NODE_SRC = the vendor worktree at c4459193); windows.yml run 37625030022 dispatched 12:59:29Z on a255b095, the installer fetched with fetch-ci-artifacts.sh (no --deploy) when it lands.
**Staging plan at the publish (one step):** a scratch copy of the downloads folder, `IGNEUM_DLSITE=<copy> publish-manifest.sh --no-deploy --version 0.3.20 --mac <dmg> --win <installer> --override <the live sixteen-field file, digest eada4bda> --notes ...`, the hive tar through publish-public.sh, the floor-moved file beside the release with its digest and the one-line diff (not published); then the deploy step on CASES END, PC 1 first, the lock lines carrying "sweep complete before 13 October 09:00 UK", hands and seed last by the build-server lane.