On Windows, a project file named after a device (nul, con, aux, prn, com1-9, lpt1-9; nul is typically left by a 2>nul redirect in Git Bash) aborts the Java import: Files.exists reports it missing, Eclipse's refresh builds a child path with an empty name and throws "Path must include project and resource name", so m2e never creates the Java project.
When JDT LS reports that failure on Windows, the server root is scanned for reserved device names (bounded, ignore files not honoured) and a one-time error notification lists them with how to recover. The name check covers COM0-9/LPT0-9 including superscript digits, case-insensitive, with or without extension or trailing spaces. Backend notification texts now go through txTemplate so templated catalog entries translate; zh text added.
Refs #976
Maven launches put a snapshot of each module's class directory on the classpath in place of the original. The originals were built as module.join("target/classes") (`...\app\target/classes` on Windows) while JDT LS reports `...\app\target\classes`; compared as text, the original stayed next to its snapshot, so every resource was on the classpath twice and MyBatis failed with "XML fragments parsed from previous mappers already contains value". Launch class directories are now joined per component, and classpath entries are compared by a key that ignores the verbatim prefix, separators, trailing separators and, on Windows, case.
Fixes#704
Tab switches and mode toggles unmount MarkdownPreview, whose only
scroll state was the DOM scrollTop; Preview Only started over at the
top. Record the preview's own scrollTop in the per-path live markdown
store and restore it on mount (split mode keeps its editor-driven
sync).
* fix(markdown): use editor-side tokens in the preview
The preview paints on var(--bg-editor) but took its text colors from
the UI tokens (--fg, --fg-strong, --fg-secondary), so with a light UI
theme plus a dark editor scheme (e.g. an imported dark .icls) it
rendered black-on-black. The code editor was unaffected because it
uses --fg-editor.
The preview now uses editor-side tokens: text var(--fg-editor),
headings var(--syn-markup-heading), code/pre/th background
var(--bg-editor-inline), blockquote var(--fg-editor) at 80%.
* chore: oxfmt src/features/run/types.ts (fixes format:check broken on main by 002401e8)
* chore: regenerate bindings after rebasing onto main
`@{u}..HEAD` fatally fails when the tracking config remains but the
remote-tracking ref is gone (the upstream branch was deleted on the
remote and the fetch pruned): the repository was dropped from the
multi-repository Push dialog, or, when every repository failed, the
dialog showed the raw fatal and disabled Push. Fall back to recent
history like the single-repository dialog does, so the branch stays
listed and pushable (a plain push recreates the remote branch).
The overview lane lies over the vertical scrollbar, so a mark the height of the lane (a new file added to git is one green VCS mark) hid the native thumb completely. The lane now draws the thumb itself, above its marks and without taking the pointer, at the position of the native one (scrollbarThumb, inset by its 2px border); it switches to the hover color while the pointer is on the thumb. The native thumb is made transparent in editors that have the lane, so the two don't add up, and still takes the drag. Editors without the lane (notebook cells, previews) keep the native thumb.
Fixes#1014
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
A recursive symbol got a fresh id on every level (ids carry the parent
chain), so sibling dedupe never matched and each expansion requested the
same level again, forever. Identify items by their position key instead:
a child whose key repeats an ancestor is kept a leaf, marked recursive,
and never expanded.
In a workspace the tree's root is the workspace directory, and the linked projects are separate top-level roots outside it. Select Opened File (and other reveals) only looked inside that root and contributed folders, so a file of a linked project was never expanded, selected or scrolled to. Reveal now resolves the innermost linked project containing the path first; the Project Files pane also counts linked projects as shown.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Any ``` line closed an open fence, so a ```` block showing a ``` block ended early and the inner code was spell checked as prose. Fences now close as in CommonMark: the same character, at least as long as the opening fence, and no info string.
Co-authored-by: fhrmrk <81476595+fhrmrk@users.noreply.github.com>
Comments and Markdown went through the string-escape rule, which drops only the letter after a backslash, so the Doxygen command `\brief` was reported as the typo "rief". Outside string literals the whole word after a backslash is skipped now (Doxygen / TeX commands, escapes quoted in comments); string literals still skip only the escape letter.
Co-authored-by: fhrmrk <81476595+fhrmrk@users.noreply.github.com>
A request URL without a scheme defaults to http://, but the check was `contains("://")`, so `GET localhost:8080/login?next=http://…` was sent as is and failed to parse. Only a `://` after a leading RFC 3986 scheme counts now, both for HTTP Client requests and the proxy connection check.
Co-authored-by: fhrmrk <81476595+fhrmrk@users.noreply.github.com>
`/<token>/<key>` has no segments after the key, so the trailing-slash flag kept its initial value and the root's index.html was served at a URL without the slash; its relative links then resolved above the root and returned 404. The flag now comes from the path, so the root redirects to `/<token>/<key>/` like any other directory.
Co-authored-by: fhrmrk <81476595+fhrmrk@users.noreply.github.com>
After Reformat Code rewrote a file's indents (e.g. spaces to tabs per Code Style), the editor kept the indents detected when the file was opened: tabs rendered at the old width, the status bar showed the old indent, and Enter inserted the old indent, mixing tabs and spaces until the file was reopened.
Whole-file format edits from language servers and contributed formatters are now tagged `format.reformat`, and after such a transaction the document runs the open-time indent detection again and reapplies its settings, so every split picks up the new indent.
Refs #935.
Co-authored-by: jaw52 <53564781+jaw52@users.noreply.github.com>
JDT LS 1.63's ECJ 3.46.200 keeps a constructor's super()/this() call as the first of its statements. Lombok 1.18.48 wrapped that call in the @SneakyThrows try block (ECJ internal error, whole workspace build failed), emitted @NonNull checks before it, and generated constructors without it.
The bundled agent is now 1.18.48-openidea.2: v1.18.48 with lombok#4059 and the upstream constructor-call fix lombok#4089. The new file name gives JDT LS a new workspace, so classes from the previous agent are rebuilt.
Fixes#1000Fixes#830
Navigation targets that start at a string literal (a Spring @Value("${key}") consumer reached from a configuration key) are named by the literal instead of the first identifier fragment inside it, which showed as "$". Applies to the declaration chooser, the quick-definition popup and the gutter line-marker chooser.
The run list applies Languages & Frameworks to every configuration it returns (interpreter or toolchain command, PATH, VIRTUAL_ENV, JAVA_HOME, GOROOT, GOPATH, Go module variables, RUSTUP_TOOLCHAIN), and every save sends the user configurations back from that list, so the derived values were written to run.toml: reopening a configuration showed the IDE's whole PATH in its environment, and since applied values never replace existing ones, a saved configuration kept the old toolchain after the interpreter or SDK changed.
Saving now drops what settings derive: an environment entry or an absolute command counts as derived when applying settings to the configuration without it produces it again. The user's own values, including ones for the same variables, are kept. Covers python, pytest, django-test, npm, cargo, go, maven, tomcat, ruby, rake, rspec and godot; Gradle stays as saved, as the run list already leaves it unapplied.
Java Go to Definition consults the Spring configuration command only inside @Value / getProperty / getRequiredProperty arguments (tree-sitter check), so ordinary Ctrl+click and Ctrl+hover go straight to JDT instead of waiting on a project-wide index rebuild.
The java-usages bundle (1.2.1) keeps its configuration snapshot across editor open/close, builds, saves of non-Java resources and reconciles; it invalidates on Java model changes and unsaved Java edits only.
Fixes#998Fixes#1002
Right-clicking a project tab in a merged window group offers Close Other Tabs, Close Tabs to the Left and Close Tabs to the Right (left/right disabled at the ends); the same commands are in the Command Palette and can take shortcuts.
The backend closes the projects one at a time, focusing each so its own unsaved-changes prompt is visible, as Quit does. Cancelling a prompt stops the batch and leaves that project and the rest open; the clicked tab stays and is selected when the batch finishes. Starting Quit clears a pending batch.
Closes#942.
Co-authored-by: jaw52 <53564781+jaw52@users.noreply.github.com>
Opening a project failed outright when the OS file watcher could not start (inotify limits used up, a file system without change notifications). The watcher now falls back to polling the same non-ignored directories every 2s, per directory unless flat so excluded trees like `node_modules` and `target` are not rescanned, and the project opens with an "External file changes sync may be slow" warning that gives the reason. Directory mtime changes seen only by polling are dropped so folders are not reported as modified.
Fixes#891.
Co-authored-by: azeemhassni <6503258+azeemhassni@users.noreply.github.com>
JSP files had no highlighting and showed a `?` icon; the HTML mode left directives, scriptlets, JSP comments and EL as plain text and flagged `<c:out/>` / `<jsp:include/>` as unclosed elements.
- A JSP grammar layer (comments, directives, scriptlets, expressions, declarations, EL) with HTML and Java overlays, so markup split by JSP elements still nests and braces match across scriptlets.
- EL as its own grammar top: member/index/method calls, namespaced functions, keyword operators, ternaries, lambdas and collection literals.
- Block comments use `<%-- --%>` in markup and `//` in scriptlets; JSP comments and scriptlets fold.
- `.jsp .jspf .jspx .jsph .tag .tagf .tagx` map to the JSP language with a file icon; `gen:jsp-grammar` regenerates the parser and a test checks it matches the grammar.
Closes#870.
Co-authored-by: jaw52 <53564781+jaw52@users.noreply.github.com>
In Chinese the Update Project notification read "已在 2 commits 中更新 3 files" and the Hot Swap result "已重新加载 3 classes". Both sentences were translated, but the counts inside them were English phrases passed in as placeholder values, which are only looked up as exact catalog entries.
Each count is now its own source text per variant (`{count} file` / `{count} files`, `{count} commit` / `{count} commits`, `{count} class` / `{count} classes`), built in the current language before it goes into the sentence. These were the only two places where a translated sentence receives an English count phrase as a value.
After a push the notification only said "Push successful". It now says what went out, for example "Pushed 3 commits to origin/main", counting the commits the branch was ahead of its upstream before the push. This covers Push and pushing a branch to its own upstream. A push across several repositories stays generic unless exactly one of them had commits to send. The title is translated as a whole for each count.
Fixes#954.
Co-authored-by: a-rookie-of-C-language <186315048+a-rookie-of-C-language@users.noreply.github.com>
`fs_read_dir` was a synchronous Tauri command, so every directory listing ran on the main thread. On a slow file system such as a network share (`\\wsl.localhost\...`), each stat while expanding the project tree stalled the whole UI. The listing now runs on the blocking pool, as the other disk-heavy commands do, both for the desktop command and over RPC.
Refs #916.
Co-authored-by: ntbang0901 <121003585+ntbang0901@users.noreply.github.com>
Deleting a worktree-isolated agent task together with its worktree failed for good with "fatal: '<path>' is not a working tree" once git no longer recognized the worktree: its registration was pruned, the repository moved, or the worktree's `.git` file was deleted.
When `git worktree remove --force` fails but the folder is still there and lies inside the agent worktrees directory, the folder is now removed directly, so the task can always be deleted. A stale registration is pruned by git later. Paths outside that directory (the task's cwd comes from its stored record) keep git's error and are never deleted by hand.
Fixes#969.
Co-authored-by: a-rookie-of-C-language <186315048+a-rookie-of-C-language@users.noreply.github.com>
Hovering an editor tab with a long path showed a tooltip whose text ran past the bubble: tooltips never wrapped (`white-space: nowrap`) and were capped at 360px.
Tooltip text now wraps inside the bubble, breaking long paths where needed, and the shortcut stays on one line next to it. The bubble grows up to 720px or the window width less 32px, so most paths still fit on one line. Short tooltips look as before (same size and line height), and newlines in tooltip text are still collapsed.
Fixes#968.
Co-authored-by: yangyezhuang <68811197+yangyezhuang@users.noreply.github.com>
Ruby gets an ERB editor mode (HTML with embedded Ruby), Ruby / Rake / RSpec run configuration types, and a Settings › Languages & Frameworks › Ruby page for choosing the interpreter (Auto picks the `.ruby-version` / mise / asdf install, else `ruby` on PATH).
- Run: `.rb` scripts run from the editor context menu, `_spec.rb` / `_test.rb` files narrow an RSpec or Rake (`SPEC=` / `TEST=`) configuration, and Rakefile tasks are detected; Rake and RSpec go through `bundle exec` when the project has a Gemfile. The chosen interpreter's `ruby`, `bundle`, `rake` and `rspec` are used and its bin directory is put on PATH.
- Configuration forms rebuild from a raw command line keeping option values with their options (`--format documentation`, `-I lib`), and a new Ruby configuration opens in the Ruby form with the script still to fill in.
- External Libraries lists the Ruby standard library and the bundle's gems, including those Bundler installs inside the project (`vendor/bundle`).
- Ruby is part of the frontend settings defaults and normalization, so the selected interpreter survives a settings round-trip.
Closes#719
Co-authored-by: retsef <2175889+retsef@users.noreply.github.com>
Settings › Editor › Code Style › Java › Tabs and Indents had no effect on Reformat Code: a file indented with 4 spaces stayed that way after choosing "Use tab character" with tab size and indent 1.
There were two causes. The reformat request sent the editor's indent, which comes from the indents detected in the file, instead of the Code Style; and jdt.ls takes the per-level indent width (`indentation.size`) only from its formatter profile, never from the request. With an explicit Tabs and Indents override for the language, Reformat Code now asks for the Code Style indent (EditorConfig still applies, the detected indents don't; without an override the editor's indent is kept). The jdt.ls profile carries the same override: it is patched into the bundled IntelliJ profile, or put into a minimal generated profile for the default formatter, and custom profiles are left alone. An unset Indent follows the tab size, as in Code Style. With tabs the request sends the tab size; every editor of a document gets the override when it is created, splits and reopened tabs included, not only after a settings change.
Checked with jdt.ls 1.63 on JDK 25: tab character, tab size 1 and indent 1 reformat a 4-space file to one tab per level with both the default and the IntelliJ formatter.
Fixes#935.
Co-authored-by: a-rookie-of-C-language <186315048+a-rookie-of-C-language@users.noreply.github.com>
Add Interpreter (creating a virtualenv), the Python packages list and pip install/uninstall, Settings › Git › Test, JVM listen-mode debugging and Create Patch › Save to file failed with "no project is open in this window", or ran without the project.
Their commands passed an empty window label to ops that resolve the calling window's project. They now pass the real window label, as the other project-scoped commands do. A sweep of every command that passes an empty label found exactly these seven ops using it: `sdk_locate_tools`, `sdk_python_create_venv`, `sdk_python_packages`, `sdk_python_pip`, `vcs_git_version`, `debug_listen_for_jvm` and `vcs_write_patch`. Without the window, the last one skipped the project's disk sync, so a patch saved inside the project showed up as an external change.
Fixes#955.
Co-authored-by: a-rookie-of-C-language <186315048+a-rookie-of-C-language@users.noreply.github.com>
Pyrefly (`pyrefly lsp`) joins basedpyright, Pyright and ty as a fallback Python language server. It is installable from Tools as a managed virtualenv tool and is listed among the interchangeable type checkers, before Ruff, so Ruff still runs after them.
Pyrefly only reports type errors in projects with a Pyrefly configuration (`pyrefly.toml` or `[tool.pyrefly]`); hover, completion and navigation work without one. A real-server test (ignored by default, needs `pyrefly` on PATH) covers startup, go-to-definition and diagnostics.
Co-authored-by: pcastro93 <2058578+pcastro93@users.noreply.github.com>
The changes tree in the Commit and Local Changes views now gives test files a light green background, so test changes stand out from production ones. A file counts as a test when it lies under a test source root (detected or marked by hand) or is named like a test (`*.test.ts`, `*Test.java`, …), the same check the project tree uses; folders under a test root get the tint too. Hover and selection still win.
The colour is a new `--bg-file-tests` token derived from `--success` and `--bg-island`, so every theme gets a matching tint, and it follows test roots as they are marked or unmarked.
Fixes#931.
Co-authored-by: jaw52 <53564781+jaw52@users.noreply.github.com>
Running a main class in a multi-module Maven project could fail with "Cannot resolve the classpath: jdtls: Attempted to beginRule: P/<dependency>, does not match outer scope rule: P/<module>".
`java.project.getClasspaths` and `getSettings` ran outside the jdt.ls build lock, so they could overlap a build or an m2e refresh of a sibling module that takes a nested Eclipse scheduling rule. The launch now resolves the classpath and JDK under the same build lock the build commands use. A delegated external build already holds that lock across the launch, so its callers pass `build_lock_held` instead of waiting on it again, which would deadlock.
Fixes#933.
Co-authored-by: a-rookie-of-C-language <186315048+a-rookie-of-C-language@users.noreply.github.com>
Check for Updates on the welcome screen showed nothing while the request was out: the check ran only as a status-bar task, and the welcome screen has no status bar. Leaving a project also dropped every notification, including a running update download.
A check started with no project or LightEdit window open now shows its progress in a balloon (keyed like the result, so the result replaces it); project windows keep it in the status bar. A project switch clears only project notifications and keeps the app-wide update ones.
Co-authored-by: youngtseup <53266026+youngtseup@users.noreply.github.com>
Tools and the install log named "pip (virtualenv)" for Python language servers (basedpyright, Pyright, ty, Ruff) even when the virtualenv is made with uv, which `install_venv` prefers whenever uv is on `PATH`.
The label now comes from the installer that will actually run: "uv (virtualenv)" when uv is on `PATH`, unchanged otherwise. debugpy's private copy still says pip, since it is always installed with `python -m pip`.
Co-authored-by: pcastro93 <2058578+pcastro93@users.noreply.github.com>
A linked worktree created below the project while it was open never joined the repository set: its files had no status entries or gutter marks until the project was reopened.
A linked worktree's `.git` is a gitfile and its HEAD lives under `<main>/.git/worktrees/<name>/`. The watcher classified both as non-metadata, `.git/worktrees` was never watched, and the repository rescan only reacted to `<repo>/.git/HEAD`. The watcher now reports the `.git` path and `.git/worktrees/**` (without `.lock` files) as VCS changes and watches `.git/worktrees`; the rescan accepts a worktree's `.git`; and the Add/Remove Worktree commands refresh the repository set directly, so both the in-IDE and the external `git worktree` paths are covered.
Fixes#927.
Co-authored-by: a-rookie-of-C-language <186315048+a-rookie-of-C-language@users.noreply.github.com>
tao makes frameless Wayland windows client-side decorated (an empty GTK
title bar, so the compositor adds no frame) but leaves them undecorated,
so GTK draws no shadow or resize margin and dark windows blend into
whatever is behind them. Decorating the window after build keeps that
empty title bar: GTK adds the theme's shadow and resize edges, nothing
else. Applies to project windows and the Settings window; X11 is
unchanged (decorating there would bring back the WM frame).
- Detect Wayland from GDK's backend (tao's own test), not env vars.
- Measure windows by GTK's content size, which excludes the shadow
margin, so remembered geometry and Settings placement don't grow.
- Skip the in-page resize handles on Wayland; GTK provides the edges.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Local Changes tab of the Git tool window showed only the changes tree; a file's diff was visible only after opening a diff tab.
It now previews the selected file's working-tree diff beside the tree when the window is docked at the bottom and below it when docked on a side. The preview follows the selection (arrow keys included), shows a placeholder for no or multiple selection, and can be turned off with View Options › Show Diff Preview (persisted; the selection is kept when toggling).
The preview reuses the diff editor, whose toolbar did not fit a side-docked window: the editor is now a size container, the whitespace and highlighting pickers move to the settings menu below 640px and the separators go below 380px, buttons no longer shrink and the differences counter truncates.
Fixes#710.
Co-authored-by: a-rookie-of-C-language <186315048+a-rookie-of-C-language@users.noreply.github.com>
GitHub Copilot, Qwen Code and Mistral Vibe were listed in the ACP Registry under their registry ids (github-copilot-cli, qwen-code, mistral-vibe), so they showed as unrelated third-party agents and could not be installed over the built-in providers. Registry ids now map to the built-in ids: the rows use the built-in name and icon, install and update into <config>/agents/<id>.toml, show "x.y.z available" when the registry has a newer version, and fall back to the built-in command on PATH after Remove. Installing never overwrites a hand-written agent config, and Remove only deletes a registry install. Changing installed providers reloads them and drops stale probes, so New Task sees the installed version right away.
Co-authored-by: dengliu <5108773+dengliu@users.noreply.github.com>
Structured issue forms for bug reports and feature requests: environment, steps, expected/actual behaviour and logs for bugs; problem, proposal and alternatives for features. Both ask reporters to search existing issues first and keep one problem per issue; blank issues are disabled and the latest release is linked from the chooser. Forms do not require a particular language.
Refs #860
Since #920, saves skip clean documents, so the detach-tab save-failure test never reached its mocked failing save. The test now edits the document first, keeping its intent: a failing save keeps the tab and opens no window.
Co-authored-by: a-rookie-of-C-language <186315048+a-rookie-of-C-language@users.noreply.github.com>
Root cause, two parts:
- An explicit save (Save All before a checkout, before running, before Commit) also wrote documents that were not modified, running save actions such as stripping trailing whitespace and adding a final newline on them — files the user never touched gained whitespace-only changes. Saves now skip clean documents, as autosave already did.
- After our own VCS operations rewrote files (checkout, rollback, stash, patch, ignore, commit/worktree hooks, Directory Compare), clean editor buffers were only refreshed by the file watcher, which can miss a rewrite that keeps the length and the timestamp's millisecond. Those operations now reload clean buffers synchronously before returning, keeping each document's chosen encoding; modified buffers keep the existing conflict handling.
Fixes#893
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Root cause: shelved file paths are relative to the repository they were shelved from, but Compare with Local, Show Diff, navigation and Unshelve joined them to the selected VCS root. In a project folder that is not itself a repository (or holds several), the local file was not found and Compare with Local stayed disabled. Shelves now record their repository (older shelves fall back to the primary repository), imported patches record the current one, and every shelf action resolves paths against it, with Windows separators handled. Stash and Apply Patch already kept their repository.
Fixes#875
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
l0ng-ai
arookieofc
Marius Ciepluch
JawandClaude Opus 5.5
Park Jong Hyun
Azeem Hassniandazeemhassni
ntbang0901
Kevin
Roberto Scinoccaandretsef
Piettro
youngtseup
arookieofc
Piettro CastroandClaude Opus 5.5
Deng Liuanddengliu