run tsc, cargo build or pytest like you always do. gleaner catches the errors as they scroll by and puts them where you can use them: squiggles in your code, a list in the problems panel, one click from the line.
no tasks.json. no problem matchers. nothing to set up.
why it's nice
- zero setup. install it and keep typing the commands you already type.
- 40+ tools. compilers, type checkers, linters, test runners and crash traces, out of the box.
- more from your extensions. picks up their problem matchers for commands it doesn't know yet.
- live results. long builds report errors as they go. huge output gets read to the end, with a warning if anything gets cut.
- whole-project typescript errors. run
tsc --watch in a terminal to see errors for files you haven't opened.
- always current. fix an error and rerun: it's gone.
tsc --watch and vitest update on every save. every other command keeps its own list.
- no noise. a problem only shows up if its file really exists, so random output never turns into fake errors.
- no doubles. if your language server already shows an error, gleaner doesn't list it again. two commands that report the same error show it once.
- follows your edits. add lines above an error and its problem moves with it, in the editor and the problems panel.
- the whole story. "declared here", compiler notes and call stacks come along with each problem.
- yours alone. runs on your machine. no network, no telemetry, no account.
works with
| language |
tools |
| typescript, javascript |
tsc, vue-tsc, tsgo, eslint, biome, esbuild, vite, jest, vitest, bun test, deno |
| js runtimes |
crashes from node, tsx, ts-node, bun, deno |
| c, c++, swift, zig |
gcc, g++, clang, clang++, cc, swiftc, swift build, cppcheck, zig (errors and panics) |
| build files |
make (its own errors), cmake, compiler errors from cmake --build |
| rust |
cargo build, cargo check, cargo clippy, cargo test, rustc |
| go |
go build, go vet, go test, go run panics |
| python |
tracebacks, pytest, pyright, mypy, ruff, flake8, pylint |
| java |
javac, mvn compile, mvn test, gradle build, uncaught exceptions from java |
| c# |
dotnet build, msbuild |
| ruby, lua |
crashes and syntax errors from ruby, ruby -c, lua, luajit |
| shell |
shellcheck |
npx tsc, uv run pytest, cd app && cargo build and package scripts like npm run build all work too. gleaner reads your package.json to see what a script really runs.
install
get it from the vs code marketplace, or from the command line:
code --install-extension thesecondcoming123.gleaner
use it
open a terminal in vs code and run your build, linter or tests. that's it.
- the status bar counts what gleaner found. click it to open the problems panel.
- turn on
gleaner.runSummaries to see what's fixed and new after a complete rerun of the exact same command. this pilot compares full runs and tsc watch passes, skips incomplete results, and stays until the next command starts, even when all problems are fixed.
- rerun a command to refresh its problems. if you stop a run with ctrl+c, the last list stays.
tsc --watch, vitest, jest --watch and nodemon refresh the list after each pass.- reload the window and the last list comes back, marked "last session". files with unsaved changes are skipped. rerun to refresh it.
- gleaner: show output opens a recent run's output, even when its problems are hidden as duplicates. use a problem's code action to jump to its whole output block.
- output views are read-only, with the command, folder, time and exit code or watch pass. gleaner keeps up to 512 kib per run and 4 mib across 50 recent runs, evicting the oldest first. cut output is marked.
- logs stay in memory. last-session problems have no saved output; rerun to get it back.
- gleaner: copy problems lets you pick a run and copy its problems, command, state and output excerpts as markdown.
- gleaner: clear problems in the command palette clears everything.
| setting |
default |
what it does |
gleaner.enable |
true |
turn gleaner on or off |
gleaner.runSummaries |
false |
pilot: show fixed and new counts after a complete, comparable rerun |
gleaner.hideDuplicates |
true |
hide a problem another extension, like a language server, already shows at the same spot |
gleaner.extensionMatchers |
true |
use problem matchers from installed extensions for commands gleaner doesn't know |
gleaner.pathMappings |
{} |
map build or container paths to your workspace, like { "/app/src": "src" }; in multi-root workspaces use ${workspaceFolder:name} as the target |
gleaner.ignoredCommands |
ls, cat, grep, rg, git, ... |
commands to never read, for output that only looks like errors |
faq
which shells work?
anything with vs code's shell integration: zsh, bash, fish and powershell on macos, linux and windows, plus git bash, remote ssh and wsl. cmd, tmux and a plain ssh you typed yourself don't report their commands, so gleaner can't see them.
does it work with watch mode?
yes, for tsc --watch, vitest (which watches by default), jest --watch / --watchAll and nodemon. each pass updates the list as you save, and stopping the watcher keeps the last one. other watchers report when you stop them.
it missed an error.
open an issue with the command and its raw output. every parser is tested against real output from the real tool, so your sample becomes a test.
i already use tasks.json.
keep it. gleaner covers everything you type outside of tasks.
build
pnpm install
pnpm test # every parser against real tool output
pnpm lint
pnpm typecheck
pnpm e2e # drives a real vs code window
license
MIT. made by alan bagel (github).
| |
