Preserve persisted file opener source routing - #2083
Merged
ymichael merged 1 commit intoAug 20, 2026
Merged
Conversation
ymichael
force-pushed
the
bb/post-merge-fix-preserve-file-opener-source-host-thr_evbnkb2vv3
branch
from
August 20, 2026 22:20
2681b5b to
f2ee10c
Compare
ymichael
deleted the
bb/post-merge-fix-preserve-file-opener-source-host-thr_evbnkb2vv3
branch
August 20, 2026 22:30
ymichael
added a commit
that referenced
this pull request
Aug 20, 2026
## What was wrong The Docs file opener rebuilt the SDK's `PluginFileOpenerSource` for its private RPC contract but dropped `experimental_hostId`. Its backend schema could not accept that field, and host-file resolution hardcoded `hostId: null`, so opening, previewing, or autosaving a file selected on remote host A could read or overwrite the same absolute path on the primary host. ## What changed The Docs app now carries the existing optional `experimental_hostId` through its opener source for both `openFile` and `saveOpenedFile`. The Docs RPC schema and parser preserve that value, and host-file resolution returns it to the existing Files SDK read, preview, and write calls. When the field is omitted, resolution still returns `null`, intentionally preserving primary-host behavior. This is the smallest complete fix because it extends the direct data flow for the existing SDK source identity; it does not add a parallel host field or refactor unrelated routing. PR #2083 is now merged, and this branch is rebased on top of it; that PR changes persisted app-tab source routing, while this change fixes the separate Docs app/server RPC hop. There is no host-daemon wire change: the plugin RPC is server-owned and the Files SDK already supports its host selector, so `HOST_DAEMON_PROTOCOL_VERSION` is unchanged. There are no CLI, guide, SDK, or user-facing configuration changes. ## How you verified The regressions failed before the implementation with the exact routing break: the app test received an opener source without `experimental_hostId`, while the server test rejected it as an unrecognized schema key (2 failed, 56 passed): - `pnpm exec turbo run test --filter=bb-plugin-simple-notes -- app.test.tsx server.test.ts` After the fix: - The same focused command passed 58 tests. The app regression verifies read and autosave RPC identities, and the two-host server regression gives the primary and remote hosts different content, preview URLs, and write results, then proves read, preview, and save all selected `host_remote`. - `pnpm exec turbo run test typecheck build --filter=bb-plugin-simple-notes --force` passed after the final rebase: 65 tests passed; the Docs typecheck and build passed; 7 Turbo tasks succeeded. - `git diff --check origin/main...HEAD` passed. Related to #2005 and #2083. > AGENT GENERATED: by GPT-5.6-Sol
ymichael
added a commit
that referenced
this pull request
Aug 20, 2026
## What was wrong The split right panel rendered the same layout twice: a shared tab-header tree and a separate pane-body tree. The header copy always laid its children out horizontally, so stacked splits detached each tab strip from its pane, and the paired resize handlers could mutate the mirrored tree on the wrong axis. Tab rendering was also split across parallel `fileTabs` and `tabModels` representations, while root-compose and plugin fixed tabs used separate adapters, making pane-local selection and content inconsistent. ## What changed - Render each pane's tab strip, controls, divider, and body as one leaf in a single split tree. Vertical and horizontal splits now resize only their own axis. - Preserve #2073’s divider performance fix: resize cursor state stays container-local, the iframe drag guard is the final container child, and pointer cancel/unmount restores the temporary flex values without writing inherited styles to `document.body`. - Preserve #2083’s file-opener routing contract in every unified renderer: persisted params own path and route identity, while the owner record supplies native preview presentation state. - Remove the mirrored header tree and its resize-peer synchronization, plus unused sidebar pane maximize/swap/close state. - Route thread tabs, root/new-thread tabs, plugin nav fixed tabs, Browser, and Terminal through the same split container and pane-local renderer. - Replace the parallel `fileTabs`/`tabModels` join with one renderable tab descriptor that owns its model, chrome, actions, content renderer, and layout behavior. - Document that `navPanel.experimental_fixedTabs` can mount once per active visible split pane; the updated contract remains covered by the unpublished plugin SDK `0.4.10` already reserved on main. There are no host-daemon protocol or other wire changes. ## How you verified - Reproduced the detached header and cross-axis resize behavior with dev-browser before the fix, then manually verified thread-panel, root-compose, plugin fixed-tab, Browser, Terminal, horizontal, and vertical split combinations after it. - Added regression coverage for pane-local tab rendering, split orientation, resizing, persisted-layout reconciliation, plugin fixed tabs, and root-compose content. - `pnpm exec turbo run typecheck --filter=@bb/app --filter=@get-bb/plugin-sdk` - `pnpm exec turbo run lint --filter=@bb/app --filter=@get-bb/plugin-sdk` (0 errors; existing warnings only) - `node packages/plugin-sdk/scripts/check-npm-version-guard.mjs` - Post-rebase divider tests: 2 files, 22 tests passed. - Post-#2083 focused app tests: 8 files, 159 tests passed. - Plugin SDK tests: 13 files, 105 tests passed. - Full app suite: 403 files, 3,097 tests passed, 3 skipped. ### Before and after #### Thread panel split | Before | After | | --- | --- | |  |  | #### Root/new-thread split | Before | After | | --- | --- | |  |  | Fixes the side-panel split regression introduced by [`4e13e3f`](4e13e3f). > AGENT GENERATED: by GPT-5.6
ymichael
added a commit
that referenced
this pull request
Aug 20, 2026
## What was wrong PR #2088 correctly preserved `PluginFileOpenerSource.experimental_hostId` through the Docs frontend and private RPC, but `resolveOpenerFile` consumed it only for `source.kind === "host"`. Root Compose opens a project-source Markdown file as `{ kind: "workspace", environmentId: null, projectId, experimental_hostId }`, so Docs accepted the source identity and then skipped its only workspace branch, throwing “Docs can open workspace, host, and thread-storage files only.” The same project-backed file already works in the PDF opener through the core project content route. ## What changed Docs now resolves a workspace source with no environment and a project ID through the project's authoritative local-path source. An explicit `experimental_hostId` selects that host; omission resolves through `system.config().primaryHostId`, matching the public opener contract and core project-workspace routing. The resolved source root and host feed the existing confined Files SDK read, preview, and optimistic-write calls. Missing, duplicate, or non-absolute project-source identities fail closed instead of selecting another machine. This is the smallest complete fix: one resolver branch plus one server regression. Environment-backed workspace, absolute host-file, and thread-storage behavior is unchanged. There is no SDK, CLI, guide, configuration, or host-daemon wire change, so `HOST_DAEMON_PROTOCOL_VERSION` is unchanged. ## How you verified The pre-fix focused regression failed both project-backed cases with the reported error (2 failed, 31 passed): - `pnpm exec turbo run test --filter=bb-plugin-simple-notes -- server.test.ts` After the fix and the final rebase onto `deb723091` / merged #2059: - The focused server suite passed 32 tests. The regression proves selected-host read, preview, and optimistic save routing; omitted-host primary routing even when a different project source is marked default; and rejection of an unknown selected host before file I/O. - `pnpm exec turbo run test typecheck build --filter=bb-plugin-simple-notes --force` passed: 66 tests, Docs build and typecheck, 7 successful Turbo tasks. - `git diff --check origin/main...HEAD` passed. Related to #2088, #2083, and #2059. > AGENT GENERATED: by GPT-5.6-Sol
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.
What was wrong
PR #1979 correctly persisted the selected host for a project-backed file in the file-opener params, and the PDF plugin read that source. Root Compose did not: it reconstructed
experimental_Original, copy-path resolution, and Open in editor from the currently selected environment/host plusfileOpenerOwner. Because Root Compose intentionally preserves file tabs across context changes, a PDF opened on host A could keep rendering from host A while those native surfaces routed to host or environment B.What changed
useThreadFileTabsnow parses the persisted opener params once and treats them as the routed file identity.fileOpenerOwnerremains responsible only for native presentation state that is not part of the plugin source, such as line ranges and workspace status.Root Compose gives a persisted project-backed opener source precedence over the live compose selection. The same resolved routing now drives the host-rendered project preview, source root and absolute copy path, and editor-open context. An explicit persisted host stays explicit; an omitted host continues to mean primary-host routing. Native file tabs still follow the existing live compose selection.
This is direct data flow from the existing persisted source, so it does not add a second host field or change the tab/server/daemon contract.
HOST_DAEMON_PROTOCOL_VERSIONis unchanged. There are no CLI, guide, SDK, or documentation changes because the documented plugin contract already requires this behavior.How you verified
Regression coverage demonstrates the old behavior and the fix:
pnpm exec turbo run test --filter=@bb/app -- src/views/RootComposeView.test.tsfailed the two new routing cases: explicit host A resolved as selected environment/host B, and omitted-primary routing resolved as host B (67 passed, 2 failed).pnpm exec turbo run test --filter=@bb/app -- src/components/secondary-panel/useThreadFileTabs.test.ts src/views/RootComposeView.test.ts— 88 passed. This includes persisted-tab reload under a different project/environment, explicit-host retention, primary-host omission, and unchanged native-tab routing.pnpm exec turbo run test --filter=@bb/app --force— 404 files passed; 3,126 tests passed and 3 skipped.pnpm exec turbo run typecheck build --filter=@bb/app— passed.pnpm exec turbo run lint --filter=@bb/app— passed with 0 errors and 156 existing warnings outside the changed lines.After rebasing onto
origin/mainata77a391a48003f364fd4d26bd8cde95e051eee3b, the focused opener/Root Compose/shared-panel set passed (3 files, 101 tests), and the Turbo app typecheck and build passed.Related to #1979.