feat: admin model info CRUD (provider/price/context/vision) (v0.6.4) - #97
Merged
Merged
Conversation
…in UI tab (v0.6.4) Rant 2026-08-19T20:40:29: admins manage model info (provider/model/ price/context/vision) with full CRUD. - db: SCHEMA_VERSION 7; models + context_length/max_output/vision/ cache_hit_input_per_m (ensure_column, idempotent); seed_models writes them from models.example.json (context_window stays aligned) - api: GET|POST /api/admin/models + PATCH|DELETE /api/admin/models/:id (admin-only 403; (provider,model) uniqueness 409; validation 400; direct delete → 0 billing per rant) - market: GET /api/models now includes context_length/max_output/ vision/cache_hit_input_per_m (real values for expandable rows) - ui: admin '模型管理' tab — search / inline add-edit form / inline delete confirm; zh+en i18n; zero-mock load-error retry - tests: admin_models_crud_full_cycle (403/409/400/CRUD) + market_models_include_new_fields; 115/115 pass + clippy + fmt + node + live smoke (list/create/patch/delete on 18080)
This was referenced Sep 11, 2026
argszero
added a commit
that referenced
this pull request
Sep 13, 2026
## Summary Two hand-written rosters in `ui/js/app.js` enumerated the "live" containers by id, and **both omitted the admin models table `#model-body`** — which was added later (PR #97) while the rosters date from PR #69/#79: 1. **`bindLiveRetry("<id>", fn)`** (13 entries, bound once at `DOMContentLoaded`) attached the load-error 重试 button of each container. `#model-body` was never registered, so `renderAdminModels()` rendered a 重试 button that **did nothing when clicked**. The smoking gun is the renderer itself: `loadErrorHtml(emptyLabel, retryFn, retryLabel)` declared a `retryFn` parameter and **never used it** — the renderer was always meant to hand its loader over, and the roster was the only thing wiring it up. 2. **`KBD_TABLE_IDS`** listed the tables reachable by ↑/↓/Enter. `#model-body` was absent, so its rows could not be activated (no `.row-active`) while all seven sibling tables could. The fix **removes both rosters instead of extending them** — a roster that must be maintained by hand is the defect, not the one missing entry. Both containers are now derived from the DOM, so no future table can be forgotten: | | Before | After | |---|---|---| | retry wiring | `bindLiveRetry(id, fn)` ×13 + per-container `addEventListener` | `setLiveError(container, html, loader)` — the renderer hands over its loader; one-time container delegation (`WeakMap`/`WeakSet`) dispatches `[data-live-retry]` | | kbd container | `KBD_TABLE_IDS.indexOf(tb.id) >= 0` | `kbdTbodyOf(t) = t.closest("tbody")`, via a single **document-level** click delegation | Runtime evidence (a jsdom probe booting the real `index.html` + the four scripts, only `fetch` stubbed so the admin endpoints 500): | check | pre-change | post-change | |---|---|---| | `#dept-body` retry re-requests (control) | 5 | 5 | | `#model-body` retry re-requests (test) | **0** | 5 | | `#model-body` second retry click | **0** | 5 | | `#dept-body` row click activates (control) | true | true | | `#model-body` row click activates (test) | **false** | true | | raise-requests rows activate (declared side effect) | **false** | true | | a run-time-injected table's rows activate (no roster can cover it) | **false** | true | | **verdict** | **DEFECT — 5/7 failed** | **OK — 7/7** | ## Related Issue None — the repository carries no open issues. Found by reconciling the two enumerated rosters against the tables that actually exist. ## Changes - [x] `ui/js/app.js` - `setLiveError(container, html, loader)` replaces the deleted `bindLiveRetry`; the dead `retryFn` parameter is dropped from `loadErrorHtml`. All **12** `loadErrorHtml`/`loadErrorRow` call sites now pass their loader through it. - `KBD_TABLE_IDS` deleted; the keyboard container is derived from the event target (`kbdTbodyOf` = `closest("tbody")`) and the row/table click delegations are merged into a single document-level listener. - `kbdRows`/`kbdContainerFrom` treat a container that has been rebuilt (`isConnected === false`) as "no active table" instead of highlighting rows that are gone. - [x] `ui/index.html` — cache-bust `js/app.js?v=20260913-2` → `-3`. - [x] `ui/README.md` — the keyboard-nav and degradation-pattern sections named both deleted rosters; they now describe the DOM-derived contract. - [ ] 涉及配置/数据结构的改动已同步示例文件 — **N/A**: no config or data-structure change. **Declared side effect** (a consequence of "every data table is navigable"): the JS-built raise-requests table now has keyboard row navigation. It was outside the old roster *and* passed `null` for the retry callback, so its degraded state had no 重试 button at all; it now reloads through `loadAdmin()`. ## Tests - [x] `cargo test` 全部通过 — **190 passed / 0 failed**, i.e. **unchanged** from `main` (this change is JS + docs only; the count staying put is the evidence that no Rust behaviour moved). - [x] `cargo fmt --check` 通过 - [x] `cargo clippy --all-targets -- -D warnings` 通过 - [ ] 新增/更新了单元测试 — **N/A**: `ui/` carries no test harness in-repo (the probes live outside it), and the change's assertion is the jsdom A/B above: with the pre-change `app.js` the probe is **red (5/7)**, with this branch it is **green (7/7)**, and the `#dept-body` control passes in both legs. ## Checklist - [x] 分支命名符合约定 — `fix/ui-live-container-rosters` - [x] Commit message 使用 Conventional Commits 格式 — `fix(ui): derive the live containers from the DOM, not from rosters` - [x] 单一职责，改动最小化 — one defect class (enumerated live containers), no opportunistic refactor; `node --check ui/js/app.js` clean.
7 tasks
argszero
added a commit
that referenced
this pull request
Sep 13, 2026
ui/ writes each table's column count by hand in several places — the <thead>, the empty-state emptyRow(N, …) colspan, the load-failure loadErrorRow(N, …) colspan, and the number of <td> in the row template. Nothing fails at runtime when they disagree: a short colspan just paints the row narrow, a long one is silently clipped, and the empty/error row is only visible when the list is empty or the request failed. That is how #97 happened (adding an 8th column to the admin models table left the empty state at colspan=7); #162 only fixed the symptom by hand. src/table_gate.rs is a test-only gate in the same shape as catalog_gate.rs and i18n_pack.rs (#[cfg(test)] in main.rs, zero new dependencies, no regex, no JS execution, no browser). It reads ui/index.html and ui/js/app.js as text and asserts, for every <tbody> table: - positive control: exactly the seven registered tables are found, each with its true <thead> column count (a scanner bug that returns an empty set fails here instead of letting the set assertions pass vacuously); - emptyRow(N) colspan == the table's column count; - loadErrorRow(N) colspan == the table's column count; - the row template has exactly one <td> per column, with whole-row colspan cells (the marketplace detail row) excluded; - the only literal colspan is that detail row, and index.html contains none; - the transactions table derives its colspan from columns.length instead of hard-coding a number; - <tbody> containers receive <tr>-shaped rows (loadErrorRow) and never a bare <div> — the browser hoists a bare <div> out of the table and the retry button then loses setLiveError's container-level delegation. The assertions are split per dimension so a single failure names the dimension that drifted. Site-to-container attribution is positional (the nearest preceding $("#id").innerHTML assignment / setLiveError($("#id"), …)), not a JS parse. Tests: 9 new; cargo test 226 -> 235, fmt and clippy clean. A/B: six mutations, each applied alone and restored byte-identically (md5 verified), produce five distinct red sets — empty colspan, thead column count, row-template cell count, error-row colspan, and the literal/dynamic colspan pair. An 8th thead column and an unregistered new table both trip the positive control (same assertion, different inputs — recorded honestly). The unmutated tree is green on all six legs, as a live control.
argszero
added a commit
that referenced
this pull request
Sep 13, 2026
…ow number (#227) Three tables rebuild their list with .filter() and then tag every row button with the index *inside the searched list*, while the handlers resolve that index against the full cached array: #api-keys copyKey / renameKey / deleteKey -> Live.apiKeys[i] #model-body editModelRow / deleteModel -> Live.adminModels[i] #dept-body openDeptForm / deleteDept -> Live.departments[i] So as soon as the search box hides one earlier row, a button acts on a different record than the row it sits in: the delete removed the wrong API key / model / department, copy handed out another key's secret, and the edit form opened with another record's values. Both correct idioms already exist in the same file — #emp-body tags rows with users.indexOf(u) and #ops-body with u.id — these three tables were simply never updated when their search boxes were added (#79, #80, #97). - api-keys: carry the cached index through the mapping, add data-key-row and resolve the row by that locator instead of nth-child - model-body / dept-body: derive the locator with indexOf, mirroring #emp-body - bump the app.js cache-bust
8 tasks
argszero
added a commit
that referenced
this pull request
Sep 13, 2026
`#model-confirm` is wired through a click listener that already applies the busy
state -- `addEventListener("click", (e) => withLoading(e.currentTarget, confirmModel))`
-- and `confirmModel` then wrapped its own request in a *second* `withLoading`.
`withLoading` opens with `if (!btn || btn.dataset.loading) return;`, and the outer
wrapper has already set `btn.dataset.loading = "1"` when the click was dispatched,
so the inner call returns immediately: the request was never sent. Adding and
editing a model were both dead, and the button silently went back to its idle
label after ~320ms, which is why it looked like nothing happened.
Both wrappers and the listener come from 98a7a9b (#97), the commit that introduced
the admin models tab, so this has never worked.
Fix: drop the inner wrapper and keep the request inline, mirroring its sibling
`confirmDept` (same admin pane, same listener pattern) -- the loading state still
comes from the listener, so double-click protection and the spinner are unchanged.
The other handlers were checked: `confirmTopup` / `confirmRaise` / `confirmDept`
call the API directly, while `commitNewKey` owns its `withLoading` because its
listener does not wrap it. `confirmModel` was the only one with both.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Admins can now manage model information (provider, model, price, context length, max output, vision, cache-hit price) with full CRUD (host rant 2026-08-19T20:40:29, project: aitokenpool).
Changes:
ensure_column):modelsgainscontext_length,max_output,vision,cache_hit_input_per_m;seed_modelsnow writes them fromdata/models.example.json(existingcontext_windowkept, aligned on write)GET|POST /api/admin/models+PATCH|DELETE /api/admin/models/:id— validation (non-empty provider/model, non-negative prices, currency ∈ {USD,CNY}, vision ∈ {0,1}), (provider, model) uniqueness → 409, direct delete (per rant: calls for a deleted model bill at 0)GET /api/modelsnow returnscontext_length/max_output/vision/cache_hit_input_per_m(real values for expandable market rows)Related Issue
Rant: 2026-08-19T20:40:29.994555+08:00 (管理员模型信息 CRUD — 厂商/模型/价格/上下文/读图等字段)
Tests
cargo test— 115/115 pass (2 new:admin_models_crud_full_cycle,market_models_include_new_fields)cargo fmt --checkcleancargo clippy --all-targets -- -D warningscleannode --checkapp.js / i18n.js