FIX: Deduplicate getExplorer and getDbExplorer API calls via Zustand store

[tahierhussain]

What

• Added a new Zustand store frontend/src/store/explorer-store.js that holds the shared responses of both explorerSvc.getExplorer(projectId) and explorerSvc.getDbExplorer(projectId).
• Updated frontend/src/ide/explorer/explorer-component.jsx to become the single writer for the store — it writes on every successful fetch of either API, and clears both slices on project switch through one clearExplorerData() call.
• Updated frontend/src/ide/chat-ai/Body.jsx to stop calling either explorer API directly. It now subscribes to explorerData and dbExplorerData from the store and mirrors them into the existing promptAutoComplete shape used by chat autocomplete.
• Removed the now-unused explorerService import and explorerSvc ref from Body.jsx (chat-ai body no longer touches explorerService at all).

Why

• explorer-component.jsx and Body.jsx were both independently calling the same two explorer endpoints with the same arguments, causing duplicate GET /explorer and GET /db_explorer requests every time the chat drawer was opened.
• The two responses were identical per-endpoint; only the downstream consumption differed — the left-side explorer tree on one hand, and the chat prompt autocomplete (@models, @seeds, DB schema/tables) on the other.
• Deduplicating both fetches reduces unnecessary network traffic and backend load, removes any chance of inconsistency between what the explorer tree shows and what chat autocomplete suggests, and establishes a single source of truth for explorer data in the frontend.

How

• New store (frontend/src/store/explorer-store.js): a minimal Zustand store with two fields — explorerData (= res.data.children from /explorer, array where [0]=models, [1]=seeds) and dbExplorerData (= res.data from /db_explorer, a single DB tree object). Setters setExplorerData / setDbExplorerData, and a single clearExplorerData action that resets both fields (intentional — both slices belong to the same project and always reset together on project switch). Intentionally no persist middleware — the payloads are refetched on every mount, can be large, and persisting only increases the risk of stale cross-session data.
• explorer-component.jsx:
  • Imported useExplorerStore and pulled setExplorerData, setDbExplorerData, clearExplorerData alongside the existing useProjectStore selectors.
  • Inside the existing getExplorer wrapper, added setExplorerData(treeData) right after setTreeData(treeData). A single write site covers every refresh trigger — mount, schema switch, model create/rename/delete, run completion, explorer refresh events, etc.
  • Inside the existing getDbEXplorer wrapper, added setDbExplorerData(treeData) right after setDBExplorer(mappedData). A single write site covers both callers — the [projectId] mount/project-switch effect and the user-triggered refresh debounce.
  • Added a small useEffect([projectId]) that calls clearExplorerData() so the store doesn't briefly serve the previous project's data to Body.jsx during a project switch.
• Body.jsx:
  • Imported useExplorerStore and subscribed to both explorerData and dbExplorerData.
  • Removed the explorerSvc.getExplorer(projectId) and explorerSvc.getDbExplorer(projectId) fetch blocks entirely, including the entire useEffect that housed them (previously gated on isChatDrawerOpen).
  • Added two small mirror effects that sync explorerData[0]/[1] into promptAutoComplete.modelsData/seedsData and dbExplorerData into promptAutoComplete.dbData. The promptAutoComplete shape consumed downstream by NewChat → InputPrompt is unchanged.
  • Removed the now-unused import { explorerService }, the useRef import, and the const explorerSvc = useRef(explorerService()).current line — Body.jsx no longer has any direct dependency on explorerService.

Can this PR break any existing features. If yes, please list possible items. If no, please explain why. (PS: Admins do not merge the PR without this section filled)

• Low risk. Neither the left-side explorer tree path nor its DB explorer tab is changed functionally — setTreeData, transformTree, setDBExplorer, mapIconsToTreeData, and all other callers of the internal wrappers in explorer-component.jsx continue to work exactly as before. The only new behavior there is an additional side-effect of writing the raw response into the shared store.
• Chat autocomplete: the promptAutoComplete shape consumed by NewChat / InputPrompt is preserved. The only behavioral change is that models/seeds/DB autocomplete data appears when the explorer has fetched at least once (as part of the IDE's normal boot), rather than being refetched on chat-drawer open. Per the agreed design, the chat drawer is only expected to be opened after the IDE and explorer have mounted, so this is not a regression in practice.
• Project switch: autocomplete will briefly show empty data (between clearExplorerData firing and the next fetches completing) instead of the previous project's entries. This is intentional and the safer behavior — it prevents cross-project data leakage in the autocomplete list.

Database Migrations

• None

Env Config

• None

Relevant Docs

• None

Related Issues or PRs

• None

Dependencies Versions

• No new dependencies. Uses the already-installed zustand package, consistent with frontend/src/store/project-store.js, frontend/src/store/model-status-store.js, etc.

Notes on Testing

• Deduplication check — open the IDE with DevTools → Network tab filtered to explorer. Reload the page, then open the chat drawer. Expect exactly one GET /explorer and one GET /db_explorer request on boot (previously each fired twice once the chat drawer was opened).
• Chat autocomplete (models/seeds) — open the chat drawer, type @ in the prompt input. Verify that model and seed suggestions still populate identically to before.
• Chat autocomplete (DB) — trigger the DB-specific autocomplete (schema/table suggestions) in the chat prompt. Verify it still populates identically to before.
• Explorer trees still render — both the models/seeds tree and the DB explorer tab in the left panel render identically. Verify that model create / rename / delete / run completion / manual refresh still update the models tree, and that the DB refresh button still refreshes the DB explorer (via onRefreshDebounce → getDbEXplorer(projectId, hardReload)).
• Project switch — switch projects via the project switcher with the chat drawer open. Verify:
  • Explorer tree swaps to the new project's models/seeds; DB explorer swaps to the new project's DB tree.
  • Chat autocomplete reflects the new project's models/seeds/DB data and does not momentarily show entries from the previous project.
• Lint/build — run npm run lint in frontend/ — should pass (the 2 pre-existing import/no-unresolved warnings for ./ExistingChat and ./NewChat in Body.jsx are unrelated to this change). No new unused-import warnings should appear in Body.jsx after the explorerService / useRef / explorerSvc removal.

Screenshots

Note: The explorer APIs are not repeated multiple times

Checklist

I have read and understood the Contribution Guidelines.

[greptile-apps]

Greptile Summary

This PR deduplicates the GET /explorer and GET /db_explorer API calls that were previously fired independently by both explorer-component.jsx and Body.jsx. A new minimal Zustand store (explorer-store.js) acts as the single source of truth: the explorer component writes on every fetch, and Body.jsx passively mirrors the store data into promptAutoComplete. All concerns raised in prior review threads are addressed — the store correctly receives the pre-mutation deep clone (rawTreeDataRef.current) rather than the post-transformTree tree, the clearExplorerData dependency is present in the effect array, and the isChatDrawerOpen removal is documented with a clear inline comment explaining the mount-order assumption.

Confidence Score: 5/5

Safe to merge — all prior P1 concerns are addressed and no new issues were found.

All three previously-raised review concerns (mutated-tree store data, missing clearExplorerData dep, isChatDrawerOpen gate removal) are explicitly resolved in this revision. The schemaMenu null-init change avoids a redundant duplicate mount fetch while the previewTimeTravel effect still fires getExplorer on first render, keeping store population intact. No P0 or P1 findings remain.

No files require special attention.

Important Files Changed

Filename	Overview
frontend/src/store/explorer-store.js	New minimal Zustand store with explorerData, dbExplorerData, setters, and clearExplorerData — correct pattern, no persist middleware by design.
frontend/src/ide/explorer/explorer-component.jsx	Stores pre-mutation deep clone (rawTreeDataRef.current) in explorerData and raw res.data in dbExplorerData; project-switch clear uses prevProjectIdRef pattern; schemaMenu init changed to null to eliminate duplicate mount-time getExplorer call (still fires from previewTimeTravel effect on mount).
frontend/src/ide/chat-ai/Body.jsx	Removed direct explorerService calls; now passively mirrors explorerData/dbExplorerData from store into promptAutoComplete; explorerSvc ref and explorerService import cleaned up; comment documents the mount-order assumption.

Sequence Diagram

sequenceDiagram
    participant EC as explorer-component.jsx
    participant ES as explorer-store.js (Zustand)
    participant B as Body.jsx (chat-ai)
    participant API as Backend API

    EC->>API: GET /explorer (single fetch)
    API-->>EC: res.data.children
    EC->>EC: deep clone → rawTreeDataRef.current
    EC->>ES: setExplorerData(rawTreeDataRef.current)
    EC->>EC: mutate + transformTree (local only)
    EC->>API: GET /db_explorer (single fetch)
    API-->>EC: res.data
    EC->>ES: setDbExplorerData(res.data)

    ES-->>B: explorerData (reactive subscription)
    B->>B: mirror → promptAutoComplete.modelsData/seedsData
    ES-->>B: dbExplorerData (reactive subscription)
    B->>B: mirror → promptAutoComplete.dbData

    Note over EC,ES: On project switch
    EC->>ES: clearExplorerData() (prevProjectIdRef guard)
    ES-->>B: explorerData=null, dbExplorerData=null
    B->>B: promptAutoComplete reset to {}

Loading

_{Reviews (4): Last reviewed commit: "fix: only clear explorer store on actual..." | Re-trigger Greptile}

[wicky-zipstack]

LGTM.
Minor note: consider adding a null fallback in handleAddNewSchema when spreading prevMenu (purely defensive).

[abhizipstack]

Good refactor — the single-writer Zustand store pattern is clean, and the PR description is thorough. Two issues to address before merge (details in inline comments).

[abhizipstack]

LGTM