Many hyperlinks are disabled.
Use anonymous login
to enable hyperlinks.
| Comment: | Keep the client's hostname in its own folder
Intent: the maintainer asked why the hostname could not be part of the root folder. It can hold the name. Only the kernel holds the hostname itself, and the machine's cannot be changed from inside, so the name was being typed into a command outside the folder for no good reason, and a start command without it runs the client under the server's settings, as the first start on the host did: serve linked server programs and stopped. Expression: the client's build writes x86-1 to /etc/hostname, and its start command becomes bq2 sysroot-namespace, then unshare --uts inside the folder, then hostname -F /etc/hostname before serve. Nothing outside the folder names it, and no outer unshare is needed. The README and the test start the client this way. Demonstrates: in an unpacked seed entered by bq2 1.0.0, plain hostname -F fails with "Operation not permitted", and with unshare --uts both hostname and a Crystal program's System.hostname read x86-1 while the machine's stays as it was. ocx-rootfs/test-rootfs.cr compiles with this start but has not run, because the built folders have moved out of reach, and neither the folders nor serve itself have run this way. ci.cr is the project check, run on this tree. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TUH8KMezTpKtTjPzg196Hf |
|---|---|
| Downloads: | Tarball | ZIP archive |
| Timelines: | family | ancestors | descendants | both | rootfs-from-seed |
| Files: | files | file ages | folders |
| SHA3-256: |
b297c80b37602935ed52383372ad8e7a |
| User & Date: | ollama-session 2026-11-01 23:43:13.715 |
|
2026-11-02
| ||
| 12:07 |
Merge the cleared-environment gate and the two root folders' build onto trunk
Intent: the maintainer, 2026-11-02: "Do merge into trunk that which has been or can be verified." clean-env-gate (3e3b1d83e2) runs the specs with no environment but PATH, a HOME of each part's own and TMPDIR, so a spec cannot pass because its host set a variable; it waited for the maintainer's deploy from trunk. rootfs-from-seed (461be54bf7, 2dbf94a4d5, b297c80b37) is how the server and the client now running were built, so trunk should hold it and describe the two machines. Expression: the two branches merged as they were, without conflicts. ocx-rootfs/README.md's list of what the build leaves out drops "OCX pins 1.18.2", since trunk now pins the seed's 1.19.1 (9c32130f12), and names the gap that remains in its place: the seed's Fossil opens no .fslckout for `fossil sql --no-repository`, so OCX's wait in a checkout waits on nothing (ticket fa798aee06). The README's other open item, the git export hook's --nocgi, is unchanged and on hold, as the maintainer said. Demonstrates: `crystal run ci.cr` on the client, Crystal 1.19.1, specs in the cleared environment: green, 597 examples, 0 failures, 2 pending, coverage 100.0%, 2 executables build. The build scripts under ocx-rootfs/ were not run again for this merge; their own evidence is in their check-ins. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TUH8KLoKjuoNhWk3Pa8Aoo check-in: a0c09042cd user: x86-1-session tags: trunk | |
|
2026-11-01
| ||
| 23:43 |
Keep the client's hostname in its own folder
Intent: the maintainer asked why the hostname could not be part of the root folder. It can hold the name. Only the kernel holds the hostname itself, and the machine's cannot be changed from inside, so the name was being typed into a command outside the folder for no good reason, and a start command without it runs the client under the server's settings, as the first start on the host did: serve linked server programs and stopped. Expression: the client's build writes x86-1 to /etc/hostname, and its start command becomes bq2 sysroot-namespace, then unshare --uts inside the folder, then hostname -F /etc/hostname before serve. Nothing outside the folder names it, and no outer unshare is needed. The README and the test start the client this way. Demonstrates: in an unpacked seed entered by bq2 1.0.0, plain hostname -F fails with "Operation not permitted", and with unshare --uts both hostname and a Crystal program's System.hostname read x86-1 while the machine's stays as it was. ocx-rootfs/test-rootfs.cr compiles with this start but has not run, because the built folders have moved out of reach, and neither the folders nor serve itself have run this way. ci.cr is the project check, run on this tree. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TUH8KMezTpKtTjPzg196Hf Leaf check-in: b297c80b37 user: ollama-session tags: rootfs-from-seed | |
| 23:18 |
Write the .bq2-rootfs marker in both root folders
Intent: the maintainer asked whether the marker had been forgotten, and it had. The seed leaves `/.bq2-rootfs` out on purpose (SEED_EXCLUDES in bq2's sysroot_workspace.cr) and bq2's own builder writes it, so a root folder built from a seed has to write it too, or bq2's workspace commands inside the folder refuse to run. Expression: `inside` in ocx-rootfs/build-rootfs.cr writes the file with the content bq2's builder gives it. The two folders already built got the same file by hand; neither was rebuilt. Demonstrates: with bq2 1.0.0 entering the client folder, `bq2 sysroot-runner --dry-run` failed with "Missing BQ2 rootfs marker" before the file existed. With it, the same command reports namespace=BQ2 and stops at "Unknown build phase all", since these folders hold no build plan, and `bq2 sysroot-status` answers. Only the marker and its parent folders' times changed. `crystal tool format --check` passes on the script; the build itself and ci.cr were not rerun. Committed without syncing because the server is stopped for the switch-over. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TUH8KMezTpKtTjPzg196Hf check-in: 2dbf94a4d5 user: ollama-session tags: rootfs-from-seed | |