mirror of
https://github.com/zed-industries/zed.git
synced 2026-11-19 06:59:22 +01:00
5377f49c26a5026286febe24a00a5fa6dfa95c0b
40720 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
|
|
5377f49c26 | v1.28.x stable for @JosephTLyons v1.28.0 | ||
|
|
aeb8cba9a8 |
Fix document color inlays multiplying when switching render modes (#66732) (cherry-pick to preview) (#66776)
Cherry-pick of #66732 to preview ---- Closes https://github.com/zed-industries/zed/issues/57600 https://github.com/user-attachments/assets/bd11e708-0e6b-487a-9501-81de918866b3 Release Notes: - Fixed document color inlays multiplying when switching render modes Co-authored-by: Kirill Bulatov <kirill@zed.dev> |
||
|
|
1e31ceb4a6 |
tasks_ui: Fix tasks with file variables missing for single-file worktrees (#66195) (cherry-pick to preview) (#66751)
Cherry-pick of #66195 to preview ---- ## Summary When a single file is opened without a folder (e.g. via `ctrl-o`), it gets a worktree whose root is the file itself. `task_contexts` only built the editor context for items from worktrees rooted at a directory, so variables like `ZED_FILE` were never set. Tasks referencing them, including global tasks from `~/.config/zed/tasks.json`, failed to resolve and were silently dropped from `task: spawn`. Tasks without variables were still listed, which made the bug look inconsistent. This PR: - builds the editor context for any active item with a project path, while keeping the worktree context limited to directory worktrees (as introduced in #25605); - makes `BasicContextProvider` set `ZED_WORKTREE_ROOT`, the relative path variables and the default cwd only for visible directory worktrees, so they don't point at the file itself or at an unrelated directory such as the Zed config directory; - moves the task picker's source-aware context selection into `TaskContexts` and uses it when spawning tasks by name or tag, so worktree tasks are resolved against their own worktree even when an external file is active; - gives items outside of visible directory worktrees (e.g. the global `tasks.json` opened via `zed: open tasks`) the visible project's `ZED_WORKTREE_ROOT` and cwd, preserving the behavior from #55727, and keeps only their file and cursor variables, so nothing derived from their own worktree (relative paths, git repository, language-provided values such as Python test targets) leaks next to the project's root; - uses the first visible directory worktree, skipping single-file ones, as the active worktree when the active item is not in a visible directory worktree, so the fallback above and the worktree context still apply when a standalone file was opened before a folder was added to the project. As a side effect, language tasks (e.g. Python runners) now also show up for single-file worktrees, since the active location is no longer dropped. Closes #66148 ## Testing - Added `test_single_file_worktree_provides_file_context`: the item context of a single-file editor contains `ZED_FILE`, has no `ZED_WORKTREE_ROOT` and no `cwd`. - Added tests that spawn a worktree-local task by name, or rerun it with `reevaluate_context: true`, while a file-backed external editor is active (single-file and invisible directory worktrees): the task resolves against its own worktree and doesn't get the external file's `ZED_FILENAME`. - Added `test_spawn_by_name_resolves_global_task_with_global_tasks_file_editor_active`: with a file-backed global `tasks.json` editor from an invisible worktree active, a global task gets the visible project's root and cwd, keeps `ZED_FILENAME` and has no `ZED_RELATIVE_FILE`. - Added `test_spawn_by_name_resolves_global_task_with_single_file_worktree_listed_first`: the same check with a visible single-file worktree listed before the visible directory worktree. - Extended `test_external_item_context_does_not_carry_its_worktree_environment`: an external item's `ZED_MAIN_GIT_WORKTREE` and language-provided variables don't reach global tasks. - Added `test_spawn_by_name_does_not_use_language_targets_derived_from_external_relative_path`: with the built-in languages registered and a Python file from an invisible worktree active, the real `PythonContextProvider` derives a relative `ZED_CUSTOM_PYTHON_TEST_TARGET` for the editor, but a global `pytest` task spawned by name gets the `.` default and the visible project's cwd. - `cargo test -p tasks_ui` and task-related tests in `project` pass. - Manually reproduced on Linux before the fix: with global tasks using `$ZED_FILE` and a task without variables, only the latter was listed for a file opened via `ctrl-o`, while both were listed with a folder opened. - Manually verified on Linux after the fix: global tasks using `$ZED_FILE` are listed and run for a file opened via `ctrl-o`. - `./script/clippy` passes. To test: add a global task with `"command": "echo $ZED_FILE"`, open a file via `ctrl-o` in an empty window, run `task: spawn`. ## Self-review - [x] I've reviewed my diff for quality, security, reliability, and performance. - [x] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). - [x] Tests cover the new or changed behavior. Release Notes: - Fixed tasks that use file variables such as `$ZED_FILE` not being listed in `task: spawn` when a single file is opened without a folder. Co-authored-by: Aliaksandr Zhyburtovich <graduss@tut.by> |
||
|
|
62e797e4b4 |
remote: Retry the POSIX environment probe without stdin (#66737) (cherry-pick to preview) (#66741)
Cherry-pick of #66737 to preview ---- Closes https://github.com/zed-industries/zed/issues/66587 Follow-up to https://github.com/zed-industries/zed/pull/66097 Release Notes: - Fixed SSH connections failing with "Missing remote shell" when the remote shell's startup reads from stdin Co-authored-by: Kirill Bulatov <kirill@zed.dev> |
||
|
|
f35ab74cbb |
gpui_macos: Handle unavailable pasteboards (#66736) (cherry-pick to preview) (#66739)
Cherry-pick of #66736 to preview ---- Closes https://github.com/zed-industries/zed/issues/66694 Follow-up to https://github.com/zed-industries/zed/pull/65713 Restores the objc1 behavior that handled `nil` case. Release Notes: - Fixed SSH connections to macOS hosts getting stuck on "Starting Proxy" when the remote server cannot access the system pasteboard - [GPUI] Fixed a crash when macOS fails to provide the general or find pasteboard Co-authored-by: Kirill Bulatov <kirill@zed.dev> |
||
|
|
8e1ba5b4e0 | v1.28.x preview for @JosephTLyons v1.28.0-pre | ||
|
|
784374c679 |
remote: Fix Windows SSH commands when the default shell is Nushell (#65580)
## Summary Closes #65575 When connecting over SSH to a Windows remote server, Zed forwards commands that need to be executed (for example terminals and tasks) through `ssh`. The logic lives in `build_command_windows()` in `crates/remote/src/transport/ssh.rs`. The command built on the client looks like this: ``` ssh -t <remote_server> powershell -E <command_base64> ``` Zed passes `powershell -E <command_base64>` to `ssh` as the remote command. The Windows OpenSSH server appends it as-is to `<ssh_remote_shell> <shell_command_option>`, so the command that actually runs on the remote server is: ``` <ssh_remote_shell> <shell_command_option> powershell -E <command_base64> ``` For `cmd.exe`, which is the default Windows OpenSSH server shell, the final command is: ``` cmd.exe /c powershell -E <command_base64> ``` This works fine. However, if the default shell is changed to `nu`, the final command becomes: ``` nu -c powershell -E <command_base64> ``` Here `nu` treats `powershell` as the only command to execute and passes `-E <command_base64>` to itself as flags. `-E` is not a flag of `nu`, so we get the following error: ``` The nu command doesn't have the flag '-E' ``` To fix this, the command is now wrapped in literal quotes when it is sent, so that the command line executed on the remote server becomes: ``` <ssh_remote_shell> <shell_command_option> "powershell -E <command_base64>" ``` This way, `powershell -E <command_base64>` is always interpreted as a single command, regardless of the shell. ## Testing I checked the process creation on the Windows remote with Procmon: - Before the change, `nu` was launched as `nu -c powershell -E <command_base64>`. - After the change, it is launched as `nu -c "powershell -E <command_base64>"`, which runs successfully. I also manually tested `nu`, `pwsh`, `powershell`, and `cmd` as the Windows OpenSSH server default shell. Terminals and tasks work correctly with all of them. ## Self-review - [x] I've reviewed my diff for quality, security, reliability, and performance. - [x] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). - [ ] Tests cover the new or changed behavior. Release Notes: - Fixed the remote terminal and tasks failing on Windows SSH servers whose default shell is Nushell |
||
|
|
3c4f6d6e90 |
openai_subscribed: Preserve cached prefix during compaction (#66527)
Follow-up to #64938 Part of #59528 ChatGPT subscription models now use `PreserveRequestPrefix`, so prompt-based compaction keeps the tool definitions and sends `tool_choice: none` instead of rebuilding the prompt, which keeps the cached prefix. These models already support `tool_choice: none` and go through the same Responses request conversion as the OpenAI provider, where OpenAI [recommends this approach](https://developers.openai.com/api/docs/guides/prompt-caching#manage-tools-with-append-only-updates). Tested with GPT-6.1-Sol: `/compact` went from 0 cached input tokens to 16,128 of 16,361. Release Notes: - Fixed compaction not reusing the prompt cache for ChatGPT subscription models. |
||
|
|
9fa6fd3f44 |
Worktree picker: search also by branch name (#66384)
## Summary In the worktree picker, can now search by both branch and worktree name instead of only by worktree name. With the current implementation whenever the worktree and a checked out branch do not share same name and the user wants to switch worktree, they have to remember the worktree name, or do some workaround like launching `zed .` from the worktree. It's not possible to switch worktree using the branch picker, as it will try to check out the branch in the current worktree (this change doesn't touch that behaviour). An example where it exists a branch named `a-random-branch` checked out in worktree `zed-deliberate-purpose` Before, when searching by `random` the worktree does not appear: <img width="696" height="197" alt="image" src="https://github.com/user-attachments/assets/cd1674c1-3ca6-4337-ad16-439f58df9e03" /> After, it is fuzzy matched because of the branch name: <img width="668" height="189" alt="image" src="https://github.com/user-attachments/assets/ed521e76-bc1b-4e4e-b662-7caba46ff91a" /> ## Testing I manually tested the change, but only on macOS. I have been using it for a few weeks with megarepos and tens of worktrees/branches. ## Self-review - [x] I've reviewed my diff for quality, security, reliability, and performance. - [x] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). - [x] Tests cover the new or changed behavior. Release Notes: - Added the possibility to search also by branch name in the worktree picker. |
||
|
|
ead5672e16 |
agent: Fix compaction being billed as un-cached input tokens (#64938)
## Summary Compaction currently bills the entire context window as input tokens instead of cached input tokens, because tool definitions change during compaction requests. Compaction here is separate from native provider compaction. Closes #59528 The fix is only able to be applied to providers which have a way of disabling tools without changing the tool definitions. This notably does not include Anthropic. This PR adds a compaction strategy enum, which describes whether a provider supports disabling tools without causing a cache miss. The new variant checks whether the model supports `tool_choice: none` and includes it in the request, while keeping the tool definitions the same. A provider may support `tool_choice: none`, but still bust the cache, or have models which don't support it. So we need both the enum and `tool_choice: none` check. ## Testing I've added tests for checking request prefix equality during compaction, where this new fix is applied. ## Self-review - [x] I've reviewed my diff for quality, security, reliability, and performance. - [x] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). - [x] Tests cover the new or changed behavior. Release Notes: - Fixed compaction being billed as un-cached input tokens |
||
|
|
ad54befbc6 |
workspace: Ensure panel is activate before opening dock (#66472)
## Summary When handling a `PanelEvent::Activate` event, the dock where the panel lives was being opened before the panel from which the event was emitted had been activated. This ends up calling the `Panel::set_activate` method for whatever panel was active on the dock before the event was handled, which could lead to weird cases where the panel from which the `PanelEvent::Activate` event was emitted ended up having its focus stolen by another panel. For example, if the dock's previous active panel was the terminal panel, calling `Panel::set_activate` would start spawning a terminal session which could finish loading after the new panel had been activated and would end up stealing the focus back to itself. Closes #66197 ## Testing Besides manually testing this, as shown in the screen recordings below, a test was added to ensure this behavior was fixed when handling the `PanelEvent::Activate` event but, in order for us to be able to assert whether a panel's `set_active` method was called, `TestPanel::activation_history` was introduced. ### Showcase <details> <summary>Before</summary> https://github.com/user-attachments/assets/1762362f-65fc-49ae-b7ed-9b925e179710 </details> <details> <summary>After</summary> https://github.com/user-attachments/assets/163b410e-13dd-4f07-8371-01b48f7e3076 </details> ## Self-review - [x] I've reviewed my diff for quality, security, reliability, and performance. - [x] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). - [x] Tests cover the new or changed behavior. Release Notes: - Fixed an issue where using `pane: reveal in project panel` after closing the last terminal tab would unexpectedly reopen the terminal panel |
||
|
|
8a41a5cc04 |
crates_io_client: Add GitHub Trusted Publishing configuration endpoints (#66473)
As a next step, this PR adds the trusted publishing endpoints to the crates.io API client. Also slightly improves our pagination handling for crates. Release Notes: - N/A |
||
|
|
ba36af2050 |
Revert "Reduce waits for navigation, suggestions, and file comparisons (#66487)" (#66517)
This reverts commit
|
||
|
|
dd34370114 |
Add organization management permission and follow organization changes after account reloads (#66494)
This PR does two things for organizations in Zed Cloud: 1. It adds one yes/no value, `can_manage_organization`, to the organization data that Zed and Delta receive from Cloud. It is true when the signed-in user is an owner or admin of that organization. 2. It fixes Zed so that, when the user loses access to their selected organization, AI features and edit predictions switch to the new organization right away instead of after a restart. ## Why **The new value.** Delta will show a "Manage Organization" link, which opens the organization's page in the dashboard, only to owners and admins of a team organization. Cloud now sends this value ([zed-industries/cloud#4489](https://github.com/zed-industries/cloud/pull/4489)). Delta reads Cloud's account data through `cloud_api_types`, so the crate must describe the new value first. **The fix.** The same Cloud change makes Cloud send a "user updated" message when someone's role changes, when they are removed from an organization, when they accept an invitation, or when they create an organization in the dashboard. Zed already reloads the user's organizations when that message arrives. Before, Cloud never sent it for these changes, so Zed did not notice them until a restart. Now Zed notices, but one step was missing. If the user was removed from their selected organization, the reload selects a different organization. It did not send Zed's "organization changed" signal. That signal is what makes Zed get a new AI token for the selected organization. Without it, AI features and edit predictions kept using the old organization until the token expired or Zed restarted. ## What changed - `cloud_api_types`: `OrganizationUiPermissions` gets `can_manage_organization`, next to the existing `can_install_github_app`. It is true for owners and admins, and false for billing managers and members. It is also true for the owner of a personal organization. Zed does not read it. - `client`: `UserStore::update_authenticated_user` now emits `Event::OrganizationChanged` when a reload replaces the selected organization. The first load after sign-in does not emit it, as before. ## Data impact - The new field defaults to false when the server does not send it, so older servers and older clients keep working. - After a removal, Zed now requests one new AI token for the newly selected organization. Nothing else changes when the selected organization stays the same. ## Limits - When "user updated" arrives, Zed also starts an AI token refresh for the organization that is selected at that moment. If the user was just removed from it, that one request fails and Zed logs an error. The token refresh triggered by the new signal then succeeds. ## Testing - `cargo test -p client --lib`: 23 passed. The new test reloads the account twice: an unchanged reload sends no signal, and a reload after removal sends exactly one and selects the remaining organization. It failed before the fix. - `./script/clippy -p client`: clean. - `cargo test -p cloud_api_types`: 13 passed. Release Notes: - Fixed AI features and edit predictions continuing to use an organization after you were removed from it. |
||
|
|
7e7311ad20 |
Reduce waits for navigation, suggestions, and file comparisons (#66487)
# Objective Reduce waiting when navigating code, loading details for code suggestions, and comparing files. Zed currently starts some independent file reads and requests only after the previous one finishes. ## Solution - Load up to 16 definition or reference results at once. Keep their original order. Repeated results for the same file share one read and one open copy of the file. - Request details for up to 16 code suggestions at once. Repeated item indexes produce one request. Successful items still update if one request fails. Keep the checks that each language server supports the request. - Read both sides of a file comparison together. Load up to 16 comparison pairs at once, while keeping their display order. - Add controlled-delay tests. Saved file contents and the text ranges returned by navigation stay the same. ## Testing On current `main` at `8a47b35109`: - 748 tests passed across `project`, `git_ui`, and `git_ui_core`. Four tests were already ignored. Each suite ran one test at a time. - Seven timing and behavior tests passed across 100 task execution orders each: 700 runs. They check result order, repeated files and suggestions, the request limit, errors, and cleanup. - Required `./script/clippy` passed for all three packages, with all targets and all features. Its dependency check also passed. The check used no debug symbols and `profile.release.opt-level=0` to limit build storage. - Formatting and `git diff --check` passed. The original checkout measured the following controlled delays. The port to current `main` passes the same 40 ms checks: | Operation | Before | After | | --- | ---: | ---: | | Go to definition across three unopened files | 120 ms | 40 ms | | Find references across three unopened files | 120 ms | 40 ms | | Load details for three code suggestions | 120 ms | 40 ms | | Open one file comparison | 80 ms | 40 ms | | Open three file comparisons | 240 ms | 40 ms | Each independent read or request has a deliberate 40 ms delay. These results prove reduced waiting in these cases. They do not establish a whole-application speed improvement. Remote network performance and the best concurrency limit for each machine remain unmeasured. The earlier checkout also passed all five checks in a clean-built macOS app. Those built-app checks predate this port. They are separate from the current-main checks above. Reviewers can run the focused tests with: ```sh ITERATIONS=100 cargo test -p project -p git_ui -p git_ui_core --features project/test-support concurrently -- --nocapture ``` ## Self-Review Checklist - [x] Reviewed the changes for correctness and cleanup of completed or failed work - [x] Added tests for the changed behavior and measured waiting time - [x] Kept result order and bounded active work ## Suggested .rules additions Proposed text for the project test guidance, for review in a separate rules change: > In GPUI tests that check the release of invisible worktrees, drop the final buffer handles inside `cx.update(...)`, then check the remaining worktrees. `run_until_parked()` alone does not flush deferred entity releases. Release Notes: - Improved responsiveness when navigating code, loading completion details, and comparing files. |
||
|
|
b7a6e5cdc0 |
call_hierarchy: Add gap between call hierarchy name and detail (#66144)
## Summary Closes #64379 ### 1: Space between signature and detail Add `gap_1` in `render_match`. Before: <img width="908" height="115" alt="Screenshot 2026-10-26 at 5 39 36 PM" src="https://github.com/user-attachments/assets/f4d7d582-dd0c-459d-85c8-24744e463f10" /> after: <img width="596" height="119" alt="Screenshot 2026-10-26 at 5 40 09 PM" src="https://github.com/user-attachments/assets/9d863061-8246-40a0-b8f7-5660db91fdf6" /> ### 2: Showing detail in tooltip + bug where tooltip would not show needs_tooltip was computed from the width of the label text alone (e.g. func helper), ignoring the separate detail (the package path). So even when the rendered row (func helper path/to/go/repo/package) was wider than the modal, the tooltip was never attached. Fixes: - Width measurement: when the label doesn’t already contain the detail, measure label + detail instead of just the label. - Tooltip content: pass the detail into CallSignatureTooltip and render it when the label doesn’t already contain it. Complete signature labels (e.g. Rust’s fn helper(&self) -> i32) already include the detail, so it isn’t appended twice. Go example: <img width="1232" height="344" alt="image" src="https://github.com/user-attachments/assets/a648054b-19eb-44d6-b465-315ccf14af44" /> Rust example: <img width="733" height="142" alt="Screenshot 2026-10-26 at 11 59 51 PM" src="https://github.com/user-attachments/assets/b854dd54-ce1f-4ef1-9849-bc6c5ae8413d" /> Tooltip now correctly shows on hover and includes detail while not duplicating in the case the label already has it. ## Testing all tests in `call_hierarchy` pass and I've added two tests: - Added a regression test that the tooltip appends the separate detail for name-only labels (e.g. Go package paths). - Added a regression test that a complete signature label (e.g. Rust) does not get its detail appended twice. ## Self-review - [x] I've reviewed my diff for quality, security, reliability, and performance. - [x] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). - [x] Tests cover the new or changed behavior. Release Notes: - Improved call hierarchy item details: added spacing between the name and detail, and fixed the hover tooltip to show the full detail (including package paths) when the row is truncated --------- Co-authored-by: Kirill Bulatov <kirill@zed.dev> |
||
|
|
ce6c5413f1 |
ui: Open submenus and popovers on the side that has room (#66496)
Follow up of https://github.com/zed-industries/zed/pull/66482 Wide submenus could open on top of their own menu, and popovers near the window bottom got pushed up over their button. The side was picked while rendering, before the menu's real size was known, so submenus guessed 200px wide and popovers never switched sides at all. This adds `Anchored::attach(target, corner)` to GPUI, which picks the side during layout using the real size and moves the element to the other side of the target when it doesn't fit. Submenus and `PopoverMenu` use it. `DropdownMenu` now opens left-aligned under its button, and settings row dropdowns opt into right alignment. Left out on purpose: - Pickers can switch between above and below as the list shrinks while you type. Using the side that fits seems better than a stale one. Before: <img width="300" src="https://github.com/user-attachments/assets/9a45aba2-6149-4835-8aa5-c2e8ae49c9b2" /> After: <img width="300" src="https://github.com/user-attachments/assets/148e205d-dd6b-4c59-9ed4-5ad37054725e" /> <img width="300" src="https://github.com/user-attachments/assets/88a24ac6-e1c5-4e92-b9be-26f813b1b576" /> Release Notes: - Fixed wide submenus and popovers covering the menu or button they open from. - [GPUI] Added `Anchored::attach` to place an element next to a target and flip it to the other side when it doesn't fit. |
||
|
|
b61f33f8b6 |
ui: Support persistent submenu refresh (#66336)
Allow submenus to refresh their contents without closing the menu. This supports updates to checkmarks, counts, labels, enabled states, and removed entries. - Retain submenu builders and use consistent confirmation behavior for mouse and keyboard input. - Add opt-in parent refresh, using stable submenu IDs to preserve an open child when its trigger changes label or position. - Repair keyboard selection after entries disappear. Close only the child when it has no selectable entries or its trigger becomes disabled or is removed. ## Self-review - [x] I've reviewed my diff for quality, security, reliability, and performance. - [x] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). - [x] Tests cover the new or changed behavior. Release Notes: - N/A |
||
|
|
57dfd54ade |
Fix Vim visual-block hang at wrapped EOF (#66118)
Closes https://github.com/zed-industries/zed/issues/66116 Release Notes: - Fixed Vim visual-block hang at wrapped EOF |
||
|
|
8a47b35109 |
Do not panic on wrong grammar results (#66474)
Closes https://github.com/zed-industries/zed/issues/65051 Adjust the code to not panic if an underlying grammar returns an error. This does not fix the slim extension per se, instead just not showing any highlights at all. To fix Slim extension, it has to update to a never tree-sitter grammar that contains the fix. Release Notes: - Fixed a panic when underlying tree-sitter highlights fail |
||
|
|
d36a706614 |
acp: Support background work in ACP v2 (#66446)
Implements the background work additions to the ACP v2 spec ([RFD](https://agentclientprotocol.com/rfds/v2/background-work)). - **`paused` tool calls:** `paused` is now its own `ToolCallStatus`. The tool call card shows a pause icon with no spinner. Before this, `paused` displayed as in progress. - **Background list:** `AcpThread::background_work()` returns the `background` list from the agent's latest state update. If the update leaves it out, background work is unknown rather than empty. While the session is idle and the agent reports background work, the label next to the message editor shows "N tasks in background". - **Stopping one tool call:** while a tool call advertises `capabilities.cancel` and hasn't finished, its card has a Stop button that sends `tool_call/cancel`. `AcpThread::cancel_tool_call` refuses locally otherwise, because Clients must not send it unadvertised. - **Stopping an agent-run terminal:** the terminal card's Stop button also appears for agent-run terminals that advertise cancel and haven't exited. It sends `terminal/cancel`, and its tooltip says the output stays available. Release Notes: - N/A |
||
|
|
601a7cc8ac |
acp: Track subagent working directories (#66440)
Implements the `cwd` and `additionalDirectories` fields that the [subagents RFD](https://agentclientprotocol.com/rfds/subagents) adds to `subagent_update`. Zed previously dropped them. - `SubagentAssociation` stores both fields as nullable patches: omitted leaves a field unchanged, `null` clears it, and a value replaces it. - A field the parent hasn't reported is inherited from the parent's own working directories, and the two fields inherit independently. Each ACP subagent's effective directories become its `work_dirs`. When a parent's directories change, nested subagents are updated too. Native subagents are unchanged. - v1 updates now carry both fields. A relative path is ignored with a warning, the same way v2 ignores a malformed field. - In v1, a subagent's `terminal/create` without a `cwd` runs in the subagent's effective `cwd`, as the RFD requires. Root sessions are unchanged. Release Notes: - N/A |
||
|
|
4ebf3eac14 |
crates_io_client: Add endpoints for listing, getting, and updating crates (#66438)
As a next step, this adds endpoints to the API client for crates.io to query a list of crates as well as patch a crate and get crate info. Also follows the OpenAPI spec available from crates.io. Release Notes: - N/A |
||
|
|
ccb25cbc63 |
crates_io_client: Add crates.io API client (#66437)
Adds a `crates_io_client` tooling crate for the crates.io REST API: - `RealCratesIoClient` authenticates with an API token and follows the crates.io data access policy (https://crates.io/data-access). - Errors distinguish missing permissions (401/403), missing resources (404), and other upstream failures. - `SimulatedCratesIoClient` (behind `test-support`) serves in-memory responses. For testing this, this PR also adds the `GET /api/v1/site_metadata` as a first endpoint. Release Notes: - N/A |
||
|
|
0294cfb58e |
gpui_windows: Preserve native tab advances (#66468)
## Summary Fix zero-width tabs in GPUI text on Windows by preserving DirectWrite’s native tab advances. DirectWrite represents tabs as zero-glyph drawing runs. GPUI previously skipped those runs and accumulated only glyph advances, losing tab spacing. Query each tab’s native hit-test width and include it in the accumulator, with an invisible space-glyph entry for caret/index mapping. Keep the original text, DirectWrite’s native tab-stop policy, and GPUI’s existing logical-order glyph placement. This does not introduce a four-space policy or attempt to unify tab stops across platforms. ## Attribution Follow-up to #66455 by @leafyoung, which identified the zero-width tab issue and proposed a fix. This takes a different implementation approach using native layout advances. Release Notes: - [GPUI] Fixed zero-width tabs in text on Windows by preserving native DirectWrite tab advances and caret positions. Co-authored-by: Yang Ye <yangye@techntea.com> |
||
|
|
8e8f54c67e |
Actualize all-settings.md contents (#66410)
Update stale defaults and add missing entries from default.json Release Notes: - N/A --------- Co-authored-by: Finn Evers <finn@zed.dev> |
||
|
|
504abf602a |
feat: config to toggle edit prediction button in the status bar (#66450)
## Summary Resolves https://github.com/zed-industries/zed/discussions/26095 https://github.com/user-attachments/assets/fca26dcc-e526-4e69-b611-4b3cdcc91066 ## Testing Manually test your changes; report results, mention platforms if relevant. How can reviewers test this easily? Include setup and steps. ## Self-review - [x] I've reviewed my diff for quality, security, reliability, and performance. - [x] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). - [ ] Tests cover the new or changed behavior. Release Notes: - Added an `edit_predictions.button` setting to toggle the edit prediction button in the status bar |
||
|
|
37568ad7dc |
ui: Support embedded views in context menu submenus (#66461)
Needed for: https://github.com/zed-industries/zed/issues/66458 I ran into this while adding branch mode to the git panel, where I needed a searchable picker inside a context menu submenu. This lets a submenu host a focusable view like that. What changed: - While a submenu is open, the parent menu no longer pulls focus away from content inside it, such as a search field. - `ContextMenu::embedded_view(view)` adds a view without the menu's row padding. A menu that only holds embedded views also drops the list padding. - `ContextMenu::focus_content(handle)` is optional. It sends a submenu's focus to that handle when the submenu opens by hover, the arrow keys, or Enter, and after a click on a part of the content that can't take focus. Without it, menus behave as before. The git panel's base branch picker uses this to focus its search field. Left out on purpose: - Accessibility roles for embedded views. I don't know much about this area, and I think it can be skipped for now. <img width="350" src="https://github.com/user-attachments/assets/8436653c-12fc-48dd-953d-bf0ee88fe8bc" /> Release Notes: - N/A |
||
|
|
504a7f2cf2 |
Allow mouse wheel zoom in terminals and diff views (#66296)
## Summary
I like to use the mouse wheel zoom functionality on the code to change
the size in Zed. It's very useful when switching between screens of
different sizes for example.
Unfortunately it only works when a file is opened, not in the diff view
or the terminal, and diff view is pretty much the only view I use
nowadays. So whenever I want to change the font sizes, I have to click
"Open file" to zoom there.
This has been annoying me for months so I fixed it with the help of
Claude Opus 5.5 High.
The agent diff and terminals in the agent panel still don't zoom on
purpose (decision made in #53598 apparently).
### Before
https://github.com/user-attachments/assets/eb47fae2-d82f-4bfc-adbe-3a3c7223bacb
### After
https://github.com/user-attachments/assets/84cd86ae-d589-457b-8ab0-733cfb60177f
## Testing
I tested on macOS 27.0.1 (26A434) and Zed 1.27.0 (
|
||
|
|
00b86477f8 |
debugger: Allow configuring the default panel size (#66435)
Closes #66412 Add `debugger.default_width` for left/right docking and `debugger.default_height` for bottom docking, along with controls in the Settings Editor. Both default to 300 pixels, preserving the existing default size. ## Testing - All 22 debugger panel tests and 76 Settings Editor tests pass. - Built and visually checked the Settings Editor on macOS with defaults of 300/300 and custom values of 495/275 after relaunch. - Rust formatting, documentation formatting, and `git diff --check` pass. Release Notes: - Added support for configuring the debugger panel’s default size via `debugger.default_width` and `debugger.default_height`. Co-authored-by: Amp <amp@ampcode.com> |
||
|
|
f4d40a2ccb |
git_ui: Drag files from the git panel to the terminal (#66423)
Files in the git panel's Changes list can now be dragged, just like entries in the project panel. Dropping a file on a terminal inserts its absolute path, and dropping it on a pane opens it. The rows reuse the project panel's `DraggedSelection` payload, so existing drop targets accept them without changes. When the dragged row is marked, the drag carries all marked files and directories, and the preview shows the entry count. Deleted files have no project entry and are not draggable. <!-- Add a GIF of dragging files from the git panel to the terminal. --> ## Testing https://github.com/user-attachments/assets/0d63ce06-057b-432b-97aa-747f6e8d09dd ## Self-review - [x] I've reviewed my diff for quality, security, reliability, and performance. - [x] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). - [x] Tests cover the new or changed behavior. Release Notes: - Added support for dragging files from the git panel to the terminal and editor panes. |
||
|
|
f73feed9de |
http_client: Remove anyhow reexports (#66430)
## Summary What problem does this solve, how, and why does this change matter? Use "Closes #123" for each issue resolved. Include artifacts produced when testing: before/after screenshots, GIFs, videos, benchmarks. ## Testing Manually test your changes; report results, mention platforms if relevant. How can reviewers test this easily? Include setup and steps. ## Self-review - [ ] I've reviewed my diff for quality, security, reliability, and performance. - [ ] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). - [ ] Tests cover the new or changed behavior. Release Notes: - [GPUI] http_client no longer re-exports anyhow types. --------- Co-authored-by: Amp <amp@ampcode.com> |
||
|
|
4ec98df500 |
acp: Update agent-client-protocol to 5.0.0 (#66425)
Upgrades `agent-client-protocol` from 3.6.0 (schema 1.13.0) to 5.0.0 (schema 1.15.1), and brings Zed's v1 and v2 ACP clients in line with the spec and schema changes made since then. ## SDK 5 port - Schema types move from `schema::v1`/`schema::v2` to `v1::schema`/`v2::schema`, builders to `Client.v1()`, and v2 handle types to `agent_client_protocol::v2`. - `agent_client_protocol::Error` is now the SDK's own type, not the v1 schema error. Zed uses it at the SDK boundary, and `ProtocolErrorInfo` also recognizes it so auth-required detection keeps working. - Foreground dispatch helpers are generic over the protocol marker, and v2 request IDs come straight from the handles. - SDK 5 only matches transport-closed errors that the SDK itself produced, so peers can't forge them. A test that built one by hand now covers only the local disconnect case. The `agent_servers` tests still cover real transport loss. ## Spec changes - `session/close` and `session/delete`: a `-32002` response counts as success, on both v1 and v2. - Permission and elicitation requests that name an unknown session get `-32002` with `data.sessionId` instead of `-32603`. The v2 migration guide makes this required for Clients too. - v2 `usage_update.cost` is a patch field: omitted leaves it unchanged, `null` clears it. A v1 update without a cost still leaves it unchanged. - A `plan_update` whose type Zed can't render (`file`, `markdown`, or an unknown type) still replaces the plan with that `planId`, so the old entries stop showing. - A diff `patch` with an unrecognized `format` renders from `changes`, as if `patch` were omitted. - v2 never sends an empty prompt. - On v2, `session/cancel` no longer answers pending permission requests locally. They stay open until the agent cancels them with `$/cancel_request`, because requests for background work survive a cancel. v1, including v1 subagents, keeps answering them with `cancelled`. ## Follow-ups These are left for separate PRs: - Background-work UI: the `paused` status, the `background` list, and cancel buttons for tool calls and terminals. Until then, `paused` displays as in progress, which the spec allows. - Subagent `cwd` and `additionalDirectories`. - The new unstable methods: `auth/status`, `session/set_title`, `session/status`, `session/archive`. Release Notes: - Improved compatibility with the latest Agent Client Protocol, including handling of missing sessions, plan updates, diff formats, and cancellation of permission requests. |
||
|
|
e38d4ec980 |
Fix active repository selection after removing remote worktrees (#66299)
Removing a project folder containing the active Git repository from a remote workspace could leave the Git panel showing “No Git Repositories” and an “Initialize Repository” button, even when other Git repositories remained open. Select a remaining repository when the active repository is removed. Preserve the current selection when removing another repository, and clear it when the last repository is removed. Testing: - Added a regression test covering removal of an inactive repository, the active repository, and the last repository. - Confirmed the regression test fails before the fix and passes afterward on Windows. - Passed the existing remote worktree add/remove/re-add test. Release Notes: - Fixed the Git panel showing “No Git Repositories” after removing the active repository's project folder from a remote workspace while other repositories remain open. --------- Co-authored-by: dino <dinojoaocosta@gmail.com> |
||
|
|
7a349ce42d |
Fix remote windows merging during session restoration (#66289)
## Summary When two windows on the same remote host open directories `[A, B]` and `[B, C]`, restarting Zed merges them into a single workspace containing `[A, B, C]`. During restoration, the shared directory `B` causes the second saved workspace to match and reuse the first restored workspace. Disable workspace matching and explicitly open a new window for each saved remote workspace, preserving the original directory sets `[A, B]` and `[B, C]` in separate windows. ## Testing - Manually verified that `[A, B]` and `[B, C]` remain in separate windows after restarting on Windows. - Passed 6 related regression tests on Windows/MSVC with `--test-threads=1`. - Formatting and whitespace checks passed. Release Notes: - Fixed remote workspaces sharing a directory being merged into one workspace after restarting Zed. --------- Co-authored-by: dino <dinojoaocosta@gmail.com> |
||
|
|
1e47770804 |
debugger: Fix CodeLLDB schema validation for build-inferred configurations (#64133)
## Summary Fix the false CodeLLDB schema warnings reported in the linked issue. - Allow configurations with `build` to omit `request`. - Apply launch/attach target checks only when `request` is present, preserving target requirements for explicit requests even when `build` is provided. No debugger runtime behavior changes. Closes #64132 ## Validation I created a tiny project with debug configuration following the guidance, and turns out the build with the fix does not report issues any more. Thus I consider that the fix is working properly. ### Before: Zed reports 3 issues <img width="1022" height="720" alt="Image" src="https://github.com/user-attachments/assets/d42e5656-0b6c-4c25-b3ce-c991305902ba" /> ### After: Zed reports no issues <img width="1500" height="642" alt="Image" src="https://github.com/user-attachments/assets/0bf4d05f-6e58-410e-8ae6-450763a2a48c" /> Release Notes: - Fixed false missing-property warnings for CodeLLDB debug configurations using build-based target inference. --------- Co-authored-by: dino <dinojoaocosta@gmail.com> |
||
|
|
0967dd028c |
Fix C++ lambda specifier highlighting (#66406)
## Summary Closes #66169 Fixed missing syntax highlighting for C++ lambda specifiers and the defined preprocessor operator. - Added lambda_specifier to the C++ keyword captures so mutable and static lambda specifiers are highlighted correctly - Highlighted defined as a preprocessor keyword - Highlighted the identifier inside defined(...) as a macro constant - Add a regression test to verify that all four tokens receive the expected captures. ## Testing - cargo test -p languages: 109 passed, 0 failed - cargo test -p languages test_cpp_highlights_query_captures_lambda_and_defined_keywords: passed - cargo fmt --all -- --check: passed. Manually tested on Windows using a locally built Zed instance: <img width="512" height="471" alt="zed_test" src="https://github.com/user-attachments/assets/f4ae3b5b-5e3e-4eb9-9d96-caf8923a37a0" /> ## Self-review - [X] I've reviewed my diff for quality, security, reliability, and performance. - [X] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). - [X] Tests cover the new or changed behavior. Release Notes: - Fixed missing syntax highlighting for C++ lambda specifiers and the defined preprocessor operator |
||
|
|
9e23c1f43e |
docs: Load c15t from the hosted CDN (#66413)
Replace the vendored c15t v2 bundle with the Inth-hosted c15t v3 script. Release Notes: - N/A |
||
|
|
6569db7470 |
remote: Accept shell-quoted ssh executable in SSH command lines (#66360)
## Why Remote Projects accepts either a bare destination or a full `ssh ...` command. Commands that tools print with every token shell-quoted (for example `'ssh' '-p' '64042' 'user@127.0.0.1'`) work in a shell but Zed rejects them with `unsupported argument`. The goal is that any valid SSH command pasted into **Connect New Server** parses the same way whether or not its tokens are quoted. Closes #61101 ## Summary `SshConnectionOptions::parse_command_line` stripped a literal, unquoted `ssh ` prefix from the raw string before tokenizing. With a quoted `'ssh'`, the prefix didn't match, so after tokenizing `ssh` became the hostname and the real destination was rejected as an extra argument. This change tokenizes first and then drops a leading `ssh` token when more tokens follow. A lone `ssh` is still treated as a hostname, as before. ## Testing - Added `test_parse_command_line_with_ssh_executable` covering unquoted, single-quoted (including an identity path with spaces), and double-quoted `ssh` executables, plus `ssh` used as a host. It fails on `main` with `unsupported argument: "user@127.0.0.1"` and passes with this change. - `cargo test -p remote`: 56 passed - `cargo fmt -p remote -- --check` - `./script/clippy -p remote` ## Self-review - [x] I've reviewed my diff for quality, security, reliability, and performance. - [x] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). (No UI changes.) - [x] Tests cover the new or changed behavior. AI disclosure: this change was AI-assisted. I reviewed the diff and the test. Release Notes: - Fixed Remote Projects rejecting SSH commands whose `ssh` executable is shell-quoted. |
||
|
|
1e77078b3a |
zed: Don't let a second launch overwrite the session ID (#66364)
## Summary Closes #65166 When launching `Zed.exe` on Windows while an instance is already open, the second process spawns `Session::new`, which immediately writes a fresh session ID to `kv_store`, before hitting the single-instance check and exiting. As the issue traces, on the next cold start Zed looks up workspaces under that orphaned ID and restores nothing. This moves `Session::new` below the single-instance check so only the primary instance records a session ID. Nothing before the check uses `session` or `session_id` (both the crash handler and `app.run` come after). *(Note: This only fixes the secondary-launch overwrite, not the `app_will_quit` timeout in the issue logs.)* ## Testing No unit test added since the ordering is inline in `main()` behind the instance mutex (happy to factor out the startup sequence for a test if you'd like). Tested manually on Windows 11 (debug build with `ZED_RELEASE_CHANNEL=nightly` so the single-instance check runs, `ZED_UPDATE_EXPLANATION` set, and a temp `--user-data-dir`): 1. Opened a workspace, launched `zed.exe` 3 more times, and checked `kv_store.session_id` in `db/0-nightly/db.sqlite` after each. - **Before:** Overwritten on every extra launch while the active workspace kept the initial ID. - **After:** Stayed `b09be725-a782-4231-9686-90b43ba1c2a1` across all 3 launches. 2. Force-killed the instance (`taskkill /F`) and relaunched: wrote a new ID (`91a430ae-3040-472c-b8cb-e873867d0ffe`) and restored the workspace as expected. I haven't tested an orderly quit (Ctrl+Q) and relaunch. Not run on Linux or macOS, which go through the same check block. Local `cargo fmt --all -- --check` and debug `cargo clippy -p zed --all-targets -- --deny warnings` pass. ## Self-review - [x] I've reviewed my diff for quality, security, reliability, and performance. - [ ] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). (No UI changes.) - [ ] Tests cover the new or changed behavior. (See Testing.) Release Notes: - Fixed launching Zed on Windows while it was already running causing the previous session not to be restored. |
||
|
|
bc6fc11740 |
gpui_windows: Keep child windows above GPUI content (#66368)
## Summary
> On Windows, a WebView2 attached as a child window inside a GPUI window
isn't visible. Because of this, I can't work with Claude/Codex
artifacts.
Windows GPUI 창 안에 webview2를 자식 창으로 붙이면 안보인다. 그래서 claude/codex의 아티팩트를 조작할
수 없다.
> The cause analysis and test notes below were investigated and written
up together with an AI assistant (Claude). I checked the before/after
screens myself.
아래 원인 분석과 테스트 기록은 AI(Claude)와 함께 조사하고 정리한 내용이다. 전후 화면은 내가 직접 확인했다.
> **AI-generated (Claude):**
>
> **Cause.** `DirectComposition::new` creates the composition target
with `CreateTargetForHwnd(hwnd, true)`. Per the [`CreateTargetForHwnd`
docs](https://learn.microsoft.com/en-us/windows/win32/api/dcomp/nf-dcomp-idcompositiondevice-createtargetforhwnd),
a window is composed bottom to top as: the window's own content, the
non-topmost composition tree, child windows, the topmost composition
tree. GPUI creates its DirectComposition windows with
`WS_EX_NOREDIRECTIONBITMAP`, so everything it draws lives in the
composition tree. With `topmost = true` that tree is drawn over every
child HWND, and a WebView2 hosted with wry's `build_as_child` is covered
completely. With `false` the tree goes below child windows.
>
> The target is created only in `DirectComposition::new`, which runs at
startup and again when recovering from a lost device, so this one line
covers both paths.
>
> The workaround today is `GPUI_DISABLE_DIRECT_COMPOSITION=1`. It
switches every window to an HWND swap chain and gives up transparent,
blurred and Mica backgrounds. Related: #34374 (suggested hosting the
webview as a composition visual instead), #54433 (contained this same
one-line change inside a larger WebView feature PR, closed for scope).
#62379 keeps `true` and doesn't change this.
>
> **Trade-off.** Child HWNDs now draw above GPUI-drawn menus, popovers
and tooltips, as they did before the DirectX renderer. Zed itself
doesn't create child windows, so Zed's own rendering is unaffected.
>
> **Testing.** Windows 11 26100, NVIDIA GTX 1080, WebView2 156, debug
builds of main @
|
||
|
|
1d76f828ee |
Fix SVG images in remote Markdown previews (#66404)
## Summary <img width="3024" height="1898" alt="zed-svg-ssh-after" src="https://github.com/user-attachments/assets/d35454cc-9d27-4165-900a-00990f483e4e" /> Closes #64517 Fixes SVG images failing to display in remote Markdown previews by adding content-based SVG support to ImageStore and using the existing GPUI renderer. ## Testing - Integration tests and Clippy passed. - Manually verified with a macOS client connected to a Linux SSH project. - To verify: open a remote Markdown file containing `` and confirm the SVG displays in the preview. ## Self-review - [x] I've reviewed my diff for quality, security, reliability, and performance. - [x] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). - [x] Tests cover the new or changed behavior. Release Notes: - Fixed SVG images failing to render in Markdown previews for remote projects. |
||
|
|
b41a8b60a4 |
Fix Windows shell detection for uppercase executable names (#66387)
## Summary
Zed was detecting `CMD.EXE` as PowerShell because the filename check was
case-sensitive. This fixes shell detection on Windows while keeping Unix
behavior unchanged.
## Testing
- Built and ran Zed on Windows. A task using `"shell": {"program":
"CMD.EXE"}` and `"command": "echo zed-shell-case-probe"` printed
`zed-shell-case-probe`.
- Added regression tests covering 22 Windows and Unix cases.
- `cargo fmt --all -- --check` and `./script/clippy.ps1 -p util
--locked` passed.
- `cargo test -p util --locked`: 128 unit tests and 2 doctests passed; 1
doctest ignored.
- Workspace doctests: 150 passed, 22 ignored.
- Full Windows workspace suite: 10,809 tests run, 26 skipped. Initially
10,798 passed, 3 failed, and 8 timed out. Nine setup-related cases
passed on retry after preparing the extension fixture, fetching the
remaining locked dependencies, and setting `CRITERION_HOME` to avoid
Criterion's runtime Cargo metadata lookup.
Two failures remain in unchanged tests, so the full suite is not
entirely green locally:
-
`extensions_ui::extension_suggestions::tests::test_suggestion_returns_after_failed_install`:
one pending installation where zero is expected; also fails in
isolation. This path downloads an extension through the fake HTTP
client, outside shell detection.
- `gpui_windows::window::tests::test_double_click_interval`: the test
assumes a 700 ms gap resets the click count, while this machine's
double-click interval is 760 ms.
## Self-review
- [x] I've reviewed my diff for quality, security, reliability, and
performance.
- [x] UI checklist: not applicable (no UI changes).
- [x] Tests cover the new or changed behavior.
Release Notes:
- Fixed Windows shell detection for executable names containing
uppercase letters.
|
||
|
|
decd2399b1 |
gpui: Keep subscriptions for entities read by reused cached views (#65952)
When a cached view was reused, entities it read outside its recorded scope lost their subscriptions, so notifying them no longer redrew the window. Two cases: - **Deferred draws.** `deferred(...)` elements are prepainted and painted at window level after the element tree, outside the cached view's recorded reads. Each deferred draw now records its own reads alongside its reusable ranges and restores them when it's replayed. - **Paint-phase reads.** A cached view recorded the entities read during render, layout and prepaint, but not during paint, so a model read only in a paint callback (for example a `canvas`) was dropped on reuse. Paint reads are now added to the view's recorded set. In both cases one set covers every phase and is restored when the prepaint is reused, since the window only consumes it at the end of the draw. It also changes how scoped reads are recorded. `detect_accessed_entities` used to clone every entity read so far in the frame and diff against it afterwards, which cost O(frame reads) per cached view (and now per deferred draw), and dropped entities that something earlier in the frame had already read. Reads are now recorded in a stack of scopes: each read goes into the innermost scope, and a scope's reads are added to its parent when it ends. Each scope now gets exactly its own reads, at a cost proportional to those reads. Restoring a recorded set now skips entities released since it was recorded. Previously the window subscribed to them again after `App` had removed their subscriptions, leaking those entries for as long as the cache kept being reused. Tests cover a canvas paint read under a reused cached panel, with and without a deferred draw around it, and with the panel itself deferred. Each fix's case fails without it. A unit test covers nested scopes, entities read before a scope began, and released entities. A unit test covers nested scopes recording entities read before them. Known limitation (unchanged): notifying a non-view entity read by a cached view redraws the window but doesn't mark that view dirty, because `mark_view_dirty` only resolves views through the dispatch tree. Release Notes: - [GPUI] Fixed entity notifications failing to redraw windows when the entity was read by a deferred draw or a paint callback inside a reused cached view. --------- Co-authored-by: Lukas Wirth <lukas@zed.dev> |
||
|
|
26abd1c959 |
gpui_platform: Extend benchmarks and add a two-revision comparison script (#66316)
Extends the GPUI benchmarks in `gpui_platform` and adds `script/gpui-bench-compare` for quick before/after comparisons. No runtime behavior changes. The only new GPUI APIs are `bench-support`-only: `Window::rendered_scene()` and `BenchWindowContext::simulate_resize`. ## Benchmarks - **`RandomizedTree/mixed` and `CachedTree/mixed`:** the general benchmark. Each iteration applies a fixed, seeded sequence of changes to a freshly built tree, one per frame: restyles, resizes, structural edits, reorders, pointer moves, scroll events, focus moves and window resizes, while some entities animate. Every iteration does exactly the same work. - **Single-change groups:** element style, leaf bounds, root layout, insert/remove, reorder and hover. They explain where a difference in the mixed workload comes from. - **`CachedTree`:** the same workloads with entities wrapped in `.cached()`. - **`TextPanels`, `Scroll` and `LayoutChurn`:** text updates and first frames, three scrolling workloads, and entity/ID/text churn. `LayoutChurn` also prints a heap check before timing. - **`Primitives`:** seeded random quads, shadows, underlines, paths and text, plus fixed overlap patterns, painted through a window and directly on `Scene`. - **The randomized tree fixture:** now draws paths, shadows, scroll containers, text, hover styles and optional animations. The benchmarks don't check their output; that's left to the fixture's tests. On macOS and Linux, frames include GPU submission through the headless renderer. Text benchmarks are skipped on Windows and on macOS without `font-kit`, because neither has a real headless text system. ## Comparison script `script/gpui-bench-compare BASE NEW` builds both revisions once and alternates short runs of 25 core cases. It reports foreground instructions per frame, wall time and run-to-run spread, in about 3 minutes. It also accepts `--filter`, `--full`, `--repetitions` and `--slow`. Both revisions must include this PR. ## Validation - A/A runs: within ±0.15% on the mixed cases and ±0.34% or better on the core suite. - All 193 `--test` cases pass (`SEED=0 ITERATIONS=1`, `bench-support,font-kit,inspector`). - The fixture's tests and `./script/clippy` pass. - A full `cargo bench` takes about 13 minutes on an M-series Mac. I haven't run it on Linux. ## Follow-ups - Add hiding the root to `CachedTree/mixed` once the `.cached()` visibility bug is fixed. - Add text edits, keyboard input, scale changes, overlays, virtual lists, images and drag and drop to the mixed workload. - Add images and SVG icons to the `Primitives` random case. - Measure GPU time, memory and upload bandwidth next to CPU instructions. - Cover hover and animation in the fixture's render-equivalence test. - Bring the window-painted `Primitives` random case into the core suite once GPU noise is separated out. Release Notes: - [GPUI] Added a mixed-workload benchmark, plus cached-view, hover, text, scrolling, layout-churn and primitive benchmarks, and `script/gpui-bench-compare` for quick instruction-count comparisons between two revisions |
||
|
|
f816e1f926 |
gpui: Render macOS window screenshots offscreen (#66344)
`MetalRenderer::render_to_image` rendered a window's screenshot into the layer's next drawable. Since macOS windows began presenting through a `CAMetalDisplayLink`, the display link owns the layer's drawables, and calling `nextDrawable` raises `CAMetalLayerInvalidOperation: -nextDrawable should not be called when using CAMetalDisplayLink`, which aborts the process. Every caller of `Window::render_to_image` on a live macOS window crashed on its first screenshot, which broke automated UI testing in apps built on GPUI. This renders the screenshot into a texture of its own instead. It moves the offscreen path that `render_scene_to_image` already used for headless rendering into a shared `render_offscreen` helper, including the blit that managed textures need on Intel Macs, and has `render_to_image` call it at the layer's drawable size. Nothing is presented, so taking a screenshot no longer touches what the display link draws. Tested by taking automation screenshots of a running GPUI app on macOS, which aborted before this change and succeed with it. Release Notes: - [GPUI] Fixed a crash when capturing a macOS window with `Window::render_to_image` while a display link drives its layer --------- Co-authored-by: Anthony Eid <hello@anthonyeid.me> |
||
|
|
c408f8efc4 |
gpui: Improve borrow and window error diagnostics (#66358)
## Summary - Preserve the attempted borrow location and existing App borrow origin in fallible borrow errors, using a compact anyhow result without a redundant standard borrow error. - Propagate caller locations through window-update wrappers and share diagnostic formatting between returned errors and panics. - Add operation and window context to all 34 log_err sites in window.rs. Context formatting runs only on failure. ## Validation - cargo nextest run -p gpui app::test:: — all 5 existing tests passed. - Formatting checks passed for the changed Rust files. - No test changes are included in this PR. - The original Delta reentrancy trigger was not reproduced; this improves diagnostics, not the underlying reentrancy behavior. Release Notes: - [GPUI] Improved App borrow and window-operation errors with caller locations, borrow origins, and operation context. |
||
|
|
d94f32a1d9 |
Terminal header corners (#66367)
## Summary The terminal tool header in the agent panel paints its own background, but the card it sits in clips its contents to a rectangle, so the header's square corners poke out past the card's rounded corners. The command block inside the header also painted its own copy of the header background with square corners, which showed at the bottom of collapsed cards. This change rounds the header's top corners for standalone terminal cards, and its bottom corners too when nothing is rendered below it (i.e. the card is collapsed and not awaiting confirmation). The command block no longer paints its own background inside the card, since the header already paints the same color behind it. When the card is expanded or showing a confirmation prompt, the header's bottom corners stay square so it joins the content below it. Follow-up to #64328, which fixed the same issue for the terminal output's bottom corners. The issue is barely visible in the bundled themes, because their `element.background` (which the header is based on) is the same or nearly the same as `panel.background`: **Stock One Dark:** the corner mismatch is nearly invisible. <img width="603" height="208" alt="Terminal tool card in stock One Dark" src="https://github.com/user-attachments/assets/1392f06e-3d31-4906-8e32-13b482a1e5c3" /> It's clearly visible in themes where those colors differ, which is common in theme extensions. To make it obvious, give the header a contrasting background with a theme override: ```json "theme_overrides": { "One Dark": { "element.background": "#ff5555" } } ``` ### Before **With the override:** the header's square corners show past the card's rounded corners. <img width="600" height="202" alt="Terminal tool card with a red header background, square corners showing past the card" src="https://github.com/user-attachments/assets/8d4e0701-d3cc-4bd2-9046-bab139a1459a" /> **Collapsed card:** the bottom corners are square too. <img width="605" height="122" alt="Collapsed terminal tool card with square corners showing past the card" src="https://github.com/user-attachments/assets/4c186b96-c899-4cba-bf79-898015d7cb5a" /> ### After **Same card with the fix:** the header follows the card's rounded corners. <img width="640" height="202" alt="Terminal tool card with a red header background, corners matching the card" src="https://github.com/user-attachments/assets/7dbaffb1-668a-4e88-a81e-40952f8174f4" /> **Collapsed card with the fix:** all four corners are rounded. <img width="641" height="122" alt="Collapsed terminal tool card with all four corners rounded" src="https://github.com/user-attachments/assets/33f19131-3e06-4d6f-a9ea-7c96f41e512e" /> ## Testing Tested manually on Windows. With the theme override above, run a terminal command in the agent panel and check: - Collapsed card: all four header corners are rounded and match the card. - Expanded card: top corners are rounded; bottom corners are square where the header meets the output. - Card awaiting confirmation: top corners are rounded; bottom corners are square. ## Self-review - [x] I've reviewed my diff for quality, security, reliability, and performance. - [x] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). - [ ] Tests cover the new or changed behavior. (Not added: this is a purely visual styling change, verified manually with the screenshots above.) Release Notes: - Fixed square corners showing on terminal tool cards in the agent panel with themes whose element background differs from the panel background. |
||
|
|
a4b39c38e0 |
markdown_preview: Stop rounding image corners (#65474)
## Summary Closes #65311 Images in the Markdown preview were always drawn with rounded corners, because `MarkdownElement` applied `.rounded_md()` to every image. Badges such as shields.io looked different from how GitHub and browsers render them. The rounding was added on purpose for images in the agent panel (#56427), so this PR doesn't remove it everywhere. It adds a `rounded_images` field to `MarkdownStyle`, which defaults to `true`. The preview style overrides set it to `false`. Other uses of `MarkdownStyle`, including the agent panel, look the same as before. If reviewer prefer that images stay square everywhere, the alternative is to drop the `.rounded_md()` call and the new field. That's simpler but also changes the agent panel. I can change current apporach to this if needed. ### Before <img width="1920" height="1080" alt="Screenshot from 2026-10-11 17-55-18" src="https://github.com/user-attachments/assets/6e96bfd2-b516-4c8e-815e-fb49b7a394af" /> ### After <img width="1920" height="1080" alt="image" src="https://github.com/user-attachments/assets/07d7ca1b-69ff-4a5d-a66b-1a2026c6e561" /> ## Testing - `cargo check -p markdown -p markdown_preview` passes. - `cargo test -p markdown`: all 186 existing tests pass. No new test was added because this is a visual change. To manual test, create a Markdown file containing:  Open the Markdown preview. The badge should have square corners. Then paste an image into the agent panel and check that it's still rounded there. ## Self-review - [x] I've reviewed my diff for quality, security, reliability, and performance. - [x] UI changes follow the [checklist](https://zed.dev/docs/development/ui-checklist). - [x] Tests cover the new or changed behavior. Release Notes: - Fixed images in the Markdown preview being rendered with rounded corners. |
||
|
|
ba63a022c9 |
recent_projects: Show remote picker scrollbar (#63909)
Make overflowing remote lists discoverable and size the scrollbar accurately from the first layout, matching the Recent Projects picker. Closes #63856. ## Self-Review Checklist: - [x] I've reviewed my own diff for quality, security, and reliability - [ ] Unsafe blocks (if any) have justifying comments - [ ] The content adheres to Zed's UI standards ([UX/UI](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist) and [icon](https://github.com/zed-industries/zed/blob/main/crates/icons/README.md) guidelines) - [ ] Tests cover the new/changed behavior - [x] Performance impact has been considered and is acceptable (the impact on [large remote lists](https://github.com/zed-industries/zed/blob/84d9a73c1957a8181c8ae017e703d1d56a18426a/crates/gpui/src/elements/list.rs#L196-L201) was not tested) --- Release Notes: - Fixed a missing scrollbar in the Open Remote picker, making additional remote servers easier to discover |