gitoria

7.5 KB

1. # Report 002 (w082): gitoria code order, no behaviour change
3. **Where:** commits on main in loreana:/media/STORAGE/projects/gitoria.worldapi.org:
4. - `110c279`: file moves (1/4)
5. - `090a20c`: topics, map and thin wrappers (2/4)
6. - `cc7bf7b`: `let` (3/4)
7. - the docs + tests commit (4/4), which also holds this report and the mission brief
9. Not pushed, not deployed. No `.env` read. Only ports 8750–8759 were used. Every server and sshd container was stopped by its PID or name; nothing of mine is still running.
11. ## Counts (before → after; `python3 tests/letcount.py .`)
13. | | before (fdfb4b1) | after |
14. |---|---|---|
15. | `.hl` files in the root | 14 | **1** (project.hl) |
16. | project.hl lines | 251 | **126** (index comment, PWA config, routes, audience, server) |
17. | `let` total | 530 | 213 |
18. | `let` never reassigned | **296** | **0** |
20. The 213 `let`s left are all needed:
21. - 125 are reassigned;
22. - 88 are re-bound in a loop body. Hybriel refuses a plain declaration there on its 2nd pass (tickets lesson).
24. ## Layout
26. The full file map is in README "Files". Imports only go one way: util / git-helpers / markdown ← users ← repos-helpers ← git ← repos ← tickets-helpers ← homepage / code / pulls / releases ← tickets ← tokens / sshkeys ← transport ← api-helpers ← api ← project.hl.
28. - **`lib/git.hl`** was 732 lines covering five topics. It is now split:
29. - `git.hl`: the git calls, branches, `initRepo`, `isEmptyNow`, `tmpDir` (moved here from transport.hl);
30. - `homepage.hl`, `code.hl`, `pulls.hl`, `releases.hl`: one per topic;
31. - `git-helpers.hl`: path, name and id checks, tab fields, sizes, and the `|||PR` / `|||RL` subject parsers.
32. - **`lib/repos.hl`**: the table and every write to it.
33. - `mayPush` is new here. It is the single "only the owner pushes" rule; before, transport.hl and sshgate.hl each had their own copy.
34. - `repos-helpers.hl`: slug rules, the row a page shows, `withChange`, `pageTitle`, the Markdown view.
35. - **`lib/users.hl`**: users and ident.
36. - `userOfLoginCode` is the one exchange + `ensureUser` step. Before, the login route and the selector face each had their own copy.
37. - `tagOf` moved here out of project.hl.
38. - **`lib/tickets.hl`** gained `finishConnect`, the logic of `/settings/connected` (taken out of project.hl).
39. - `tickets-helpers.hl`: the tickets URL, the answer shapes, `#12` references.
40. - **`lib/transport.hl`** holds `gitTransport` and `callGit`.
41. - `transport-helpers.hl`: the Basic password, plain answers, `Git-Protocol`, pkt-line.
42. - **`lib/sshkeys.hl`** gained `keyLine`, the authorized_keys line (taken out of sshgate.hl).
43. - **`lib/api.hl`**: every function route, as thin wrappers:
44. - `/api/repos…`;
45. - `/login/callback`;
46. - `/settings/connect…`;
47. - `/__git/keys` and `/__git/access` (this was sshgate.hl).
48. - **`lib/api-helpers.hl`**: JSON / Markdown / text answers, `goTo`, `sshGuard`, `repoOfRequest`, `safeNext`.
49. - **`lib/util.hl`**:
50. - was localtime.hl;
51. - also holds env, `storageDir`, the addresses (`publicUrl`, `slugOfHost`, `domainFor`), lists, text checks;
52. - `newestFirst` replaces three copies of the same insertion sort.
53. - **Faces made thin:**
54. - "Make main" → `code.hl makeMain`;
55. - the Merge answer → `pulls.hl pullsView`;
56. - the selector login → `userOfLoginCode`.
57. - The session writes stay in the faces, because a face's `session` parameter is the framework's own.
58. - **Removed:**
59. - `jsoncheck.hl` was imported nowhere;
60. - `git.hl parseView` was called nowhere.
61. - **Moved:** `styles.hl` → `components/styles.hl`. `tools/migrate-short-ids.hl` now imports from `lib/`.
63. ## Tests
65. | test | result |
66. |---|---|
67. | gate `tests/browser.mjs` (8750–8759) | **200 passed, 0 failed** before and after each step |
68. | `tests/push.mjs` | **46/0** before and after each step |
69. | `tests/ssh.mjs` | **44/0** before and after each step |
70. | live-data reads + writes | identical, except **1 intended difference** (below) |
72. The gate commands are in README "Test". The browser gate runs with `GITORIA_GATE_TICKETS_DIR` set to a `git archive HEAD` copy of tickets. The reason: the tickets working tree is currently being changed by another worker, and with it the gate crashed at the start ("tickets name failed").
74. **Live-data comparison** (new `tests/realdata-baseline.mjs` and `tests/realdata-compare.py`; how to run: README "Test" → "Same output"):
75. - **Data:** a copy of Byrodin's `storage/mpackdb` + `storage/git`, taken 2026-10-03. Each run gets a fresh copy.
76. - **Reads** (678), signed in as az5b2 through a session file:
77. - every page of all 13 repos, signed in and out: `/` including the 3 MB tracker README, code, files, folders, branches, commits, pulls, releases, tickets, settings;
78. - unknown addresses;
79. - the 20 browser modules;
80. - `/api/repos…` as JSON and as Markdown;
81. - every refs advertisement, v0 and v2;
82. - the refusals;
83. - the login, connect and `/__git` routes.
84. - **Writes** (80):
85. - faces: create, token, ssh key, Make main, Merge, open ticket, disconnect, remove, login, logout;
86. - over HTTPS: clone, then push with a token (also refused without one or with a bad one);
87. - over SSH, through the real sshd container: clone tracker, push a branch to email, refused after the key is removed;
88. - the pages that follow each write.
89. - **Tickets:** the connected repo's tickets are a snapshot, served on port 8759, so nothing reaches live tickets. The live tickets API changed its links during this run, which is why a snapshot is needed.
90. - **Masks:** two runs of the OLD code compare clean (758/758).
91. - **Browser modules:** they differ only in the tables of imported statics and in source positions. The script checks the component code and the kind of every imported name it uses. Step 3 also changes `var` → `const`.
93. **The 1 difference (intended, a fix that came with the cleanup):**
94. - **What changed:** a failed `/login/callback` in a browser that already has a session now shows its reason on `/login/failed`. For example: "login is not set up …".
95. - **Before:** it always showed "the login did not finish — try again".
96. - **Cause:** the old `failed(req, why)` was a second call. It got a COPY of `req` and so a copy of the session, and its write was lost.
97. - **Fix:** project.hl now hands `&req` and `&server.sessions` to the session-writing routes, the same fix as in tickets mission 010.
98. - **To undo it:** pass `req` by value to `loginFailed` in lib/api.hl.
100. **Other checks:**
101. - `tools/migrate-short-ids.hl` loads and refuses exactly as before (no storage, no key, ident down).
102. - `tests/short-id-switch.mjs` was **not run**: it is historical and uses fixed ports 8724/8725, which are outside my range.
104. ## Found (in STATUS "Lessons")
106. 1. **A session-writing route needs `&req` and `&server.sessions`.** gitoria lost the login-failure reason through a by-value `req`.
107. 2. **The browser module carries the statics of every file a component imports.** Moving a static changes the module's tables and the `?v=` hashes, but not the page. Compare the component code instead.
108. 3. **The tickets working tree can be mid-change.** Gate against a `git archive HEAD` copy.
109. 4. **A crashed gate leaves its servers running.** Check `ss -ltnp | grep :875` and kill by PID.
111. ## Not done / open
113. - **Not deployed** (that is the architect's job). Use the normal `deploy.sh`. There is no data change.
114. - deploy.sh's rsync does not delete, so the old root `.hl` files stay on Byrodin. They are unused and can be removed by hand.
115. - **Scratch** (gitignored) in `.scratch/w082/`:
116. - `old/` (the old tree, code only);
117. - `tickets-head/` (code only);
118. - `gates.sh` (the three gates on 8750–8759);
119. - `unlet.py`.
120. - **Deleted after the run:** the live-data copy and every baseline output (they hold live data).