v0.0.4

Adds a third feature area: Checksum Toolkit, for silencing the editor's "installation appears to be corrupt" warning after you have patched its core files.

Added

• Fix Checksums (editorConsoleToolkit.checksums.apply): recomputes the SHA-256 of every file listed under checksums in the running editor's product.json and writes back only the entries that actually differ. Entries whose file is missing from disk are skipped and reported instead of aborting the run; when nothing differs, nothing is written.
• Restore Checksums (editorConsoleToolkit.checksums.restore): writes the backup back and only then deletes it, so a failure halfway through never leaves you without either file.
• A backup is copied — not moved — to product.json.ect-orig.<version> beside the original before the first write, and is not overwritten on later runs. New content goes to a temporary file in the same folder and is renamed over product.json, so an interrupted run cannot produce a truncated or missing product.json.
• No privilege elevation, ever. Where the installation is not writable — typically Windows and Linux — the fixed product.json is staged in your temporary folder and the notification hands you the single equivalent command, with a Copy command button that puts it on the clipboard.
• Backups left behind by other editor versions are cleaned up when you run either command. Only files this extension wrote itself (product.json.ect-orig.*) are ever removed.
• Both commands refuse to run, with the reason shown, when the extension host is not the machine holding the editor: Remote (SSH, WSL, container) windows, the web, or any case where the editor cannot report its installation folder.
• 36 tests covering the checksum algorithm against fixed vectors, and the apply/restore ordering, permission-denied and partial-failure paths against an in-memory filesystem — 344 tests in total.

Notes

• Either command needs a full quit and restart of the editor. Reloading the window does not apply the change; this is a property of how the editor verifies itself, not something the extension can work around.
• Restore goes back to the state before this extension's first fix — not to a factory-fresh product.json. If checksums had already been patched by another tool, that is what the backup captured. With no backup left, the way back is reinstalling the editor.
• The first fix reformats product.json: a JSON round-trip expands the short single-line arrays the editor ships with, which makes the file about 4KB larger. The data is equivalent, and the byte-exact original is what the backup holds.
• An editor upgrade replaces product.json, so the warning returns and Fix Checksums has to be run again. There is no automatic detection of that.
• Patching the core files themselves is out of scope — this only stops the editor complaining about a patch you already made.
• Still no runtime dependencies, no telemetry, no network access, and nothing happens at activation: the new commands touch the disk only when you run them.
• Only verified on macOS. The Windows and Linux paths — including the not-writable-installation flow — are implemented and covered by tests, but have not been run on a real Windows or Linux machine.
• Behaviour derived from vscode-fix-checksums — MIT © Jürg Lehni 2018, MIT © RimuruChan 2024. Written from scratch in TypeScript; none of its code or its sudo-prompt / tmp dependencies is bundled. The attribution lives in the README licence section.

v0.0.3

Reworks where an inserted log lands, and turns the ports view's search into a permanent row.

Added

• Logs now go inside the block they belong to. With the cursor on a line that opens a block — a function signature, an if / for head, a callback body — the log is inserted as that block's first statement, indented like the rest of the block, instead of after the whole construct. This is the placement you want when you are logging a parameter.
• Blocks that cannot hold statements (switch, class, Rust's match and impl) put the log on the line before the statement, where the value being logged has already been evaluated.
• A search row is always present as the first row of the Port Toolkit view. Click it to type; the list narrows as you type, Enter keeps the filter and Esc restores what was there before you opened the box. While a filter is active the row shows the term and a 2 of 7 shown match count, and carries a ✕ to clear it; with no filter there is nothing to clear, so the ✕ is absent.
• 13 regression cases covering insertion placement across the supported languages, plus a lexer test suite — 305 tests in total.

Changed

• The ports view is now named Port Toolkit rather than Listening Processes, matching its container.
• Search Ports and Clear Port Search left the view's title bar: search is the row, and clear is the ✕ on it. Both commands remain in the Command Palette and stay bindable to a key.
• Block syntax is now read from each language's adapter rather than assumed to be C-like, which is what makes Python's indentation and the do / end blocks of Ruby, Lua and Elixir land correctly.
• Log removal now requires the log to sit on the line immediately next to its anchor, which follows from the log no longer always being after the statement.

Fixed

• The cursor being inside a multi-line string or a block comment is now detected: a template literal, a Python triple-quoted string or the body of a /* … */ refuses the insertion instead of writing a log into the middle of a literal. Previously only single-line comments and strings were recognised.
• A log inserted for an argument of a multi-line call took its indentation from the argument's line, so it did not line up with the closing );. Indentation is now taken from the statement's first line.
• if (input.type === "file") { was read as a restricted block, because the restricted-block keywords matched property names as well — every condition containing .type had its log pushed before the statement. Property accesses no longer match.

Notes

• A filter does not clear your ticks. A process ticked before you filtered stays ticked even when the filter hides its row, so Kill Checked Processes will terminate it. The confirmation dialog always lists everything that is about to go down.
• When the syntax is genuinely ambiguous — a lone { whose meaning cannot be told apart — the log falls back to landing after the whole statement: further out than you might want, but never syntactically broken.
• Still no runtime dependencies, no telemetry and no network access.

v0.0.2

Adds a second feature area: Port Toolkit, for seeing which processes hold your local ports and terminating them in batches.

Added

• Port Toolkit view in the Activity Bar, listing every process that is listening on a local port. Rows are one per process, not one per port, so a process holding several ports takes a single row and the number of rows you check always equals the number of processes that get terminated.
• Source information for telling processes apart: PID, PPID, owner, uptime, working directory, every listening endpoint, and the full untruncated command line. A this workspace marker appears when the process was started from inside one of your open workspace folders — which is what distinguishes three concurrent node dev servers from each other.
• Batch termination via native checkboxes: check any number of rows, then run Port Toolkit: Kill Checked Processes. A modal confirmation lists exactly what will be terminated, and a summary afterwards reports how many actually went down. Checked rows survive the automatic refresh.
• SIGTERM first, escalating to SIGKILL only after editorConsoleToolkit.ports.killTimeout, so a dev server gets the chance to close its sockets. Child processes are never touched (taskkill is invoked without /T on Windows).
• The command line is re-checked immediately before any signal is sent. If a process exited and its PID was reused after you checked the row, that row is skipped and reported as skipped instead of being terminated blindly.
• The editor can never be terminated. The parent chain from the extension host up to the editor's main process is derived from the live process tree at startup; every process on it is shown with a lock icon, has no checkbox and no terminate action. Being derived from the process tree rather than a list of names, this holds for VS Code, Cursor, Windsurf and other derivatives.
• Processes owned by another user are listed but marked as not terminable, since the extension cannot elevate.
• Search across process name, full command line, PID, working directory and any port number; system-process filtering; manual refresh.
• Five settings under editorConsoleToolkit.ports.*: refreshInterval (default 5000, 0 disables auto-refresh), killTimeout (3000), hideSystemProcesses (true), systemPortMax (1024), includeUdp (false).
• Default keybinding Alt+L for Insert Console Log, limited to editorTextFocus && !editorReadonly.

Changed

• New icon: a terminal window with two output lines — one dimmed to read as commented out, one highlighted — and three status lamps below it. The previous > prompt now belongs to Editor Lang Toolkit; two extensions sitting next to each other in the extensions list have to be distinguishable at a glance.

Notes

• The view rescans only while it is visible; switching to another Activity Bar icon stops the timer, so there is no background scanning when you are not looking at it. One scan is three fixed command invocations regardless of how many ports are open.
• A failed scan shows a single error row in the view rather than raising a notification, because with polling a notification per cycle would bury the editor.
• Still no runtime dependencies, no telemetry and no network access.
• No default keybindings are contributed for any Port Toolkit command.
• Port and process discovery uses lsof on macOS, ss (falling back to netstat) on Linux, and PowerShell on Windows. The Linux and Windows parsers are covered by fixture tests but have not been verified on real Linux or Windows machines.

v0.0.1

Initial release.

Added

• Insert Console Log (editorConsoleToolkit.insertConsoleLog): inserts a debug log statement for the selected expression, or for the expression under the cursor, on the line after the enclosing statement.
• Toggle Console Log (editorConsoleToolkit.toggleConsoleLog): inserts the log when it is missing and removes it when it is already there. Also removes the log when the cursor sits on the log line itself.
• Toggle All Console Logs (editorConsoleToolkit.toggleAllConsoleLogs): comments out every generated log in the current file, and uncomments them on the next run. Nothing is deleted, indentation is preserved, and a mixed file converges to fully commented on the first run.
• Support for 12 language adapters covering 15 language ids: JavaScript, JSX, TypeScript, TSX, Python, Java, Kotlin, C#, Lua, Ruby, PHP, Swift, Dart, Rust and Elixir.
• editorConsoleToolkit.prefix setting (default 🎯🎯🎯 [DEBUG]), scoped as language-overridable so it can be overridden per workspace, per folder and per language.
• Multi-cursor support. Every change made by one command invocation is a single undo step, and the document is never saved or formatted automatically.
• Generated logs carry a short ownership marker (// ect:v1) so the extension only ever removes its own logs, and keeps removing them after the prefix is changed.
• UI available in English and 简体中文, following the VS Code display language.

Notes

• No default keybindings are contributed. Bind the three commands yourself in Keyboard Shortcuts.
• The extension has no runtime dependencies, no telemetry, no network access and no background listeners. It activates only when one of its two commands runs.