Scripts and the REPL

Running code

clove                      # REPL
clove script.clj a b       # run a script; *args* is ("a" "b")
clove -e '(+ 1 2)'         # evaluate and print
clove -m my.app            # require my.app and call its -main
echo '(+ 1 2)' | clove     # piped stdin runs as a script
clove watch script.clj     # rerun on every save

A script can name its interpreter. #! is a comment to the reader, so this runs directly once it's executable:

#!/usr/bin/env clove
(println "hello," (or (first *args*) "world"))

The REPL

The prompt is a multi-line editor. Enter runs the form if it's complete and opens a new, indented line if it isn't. ↑ brings back whole forms, brackets pair themselves, Tab completes, and Ctrl-R searches history. The arglists of the function you're calling show dimly after the cursor. If your ~/.inputrc says set editing-mode vi, you get vi mode, with counts, text objects and ..

Results are numbered, and *1, *2, *3 hold the last three. A few functions help you look around:

(doc map)              ; docstring and arglists, special forms included
(source take-while)    ; the definition
(dir clojure.string)   ; a namespace's public names
(apropos "index")      ; names containing "index"
(find-doc "lazy")      ; docstrings mentioning "lazy"
(inspect big-map)      ; browse a value full-screen; q returns

After an error, *e holds the exception and (pst) prints it again.

Debugging

dbg prints a form and its value, then returns the value, so you can wrap anything without changing what the code does. #p is shorthand for it.

Uncaught errors underline the form that failed, suggest a fix when they can, and print a trace that reaches into function bodies:

── ERROR  report.clj

I don't know what `printn` refers to.

   1 │ (printn total)
     │  ▔▔▔▔▔▔

Did you mean `print`?

  at (printn ...) line 1

Asking Claude

A REPL line that starts with ? goes to Claude instead of the evaluator, along with the session so far. Claude can look up docs and source on its own but never runs code. When running something would help, it puts the line in your prompt, and you decide whether to press Enter. A bare ? after an error asks it to explain the error. This needs curl and an ANTHROPIC_API_KEY.

Projects and tasks

clove new my-app makes a project: src/, test/, and a clove.edn that works like babashka's bb.edn. A string task is a shell command.

{:paths ["src"]
 :tasks {test "clove test"
         ci   {:depends [test] :task (println "ok")}}}

clove tasks lists them and clove run ci runs one, dependencies first.

Tests

clojure.test is built in. clove test finds every *_test.clj under the current directory, runs it, and exits 1 if anything failed. A failed (is (= ...)) prints where the two values differ.

clove judge is for snapshot tests. Write (test expr) forms with no expected value and clove judge -a file.clj fills in what they return. Later runs fail when the value changes. It's borrowed from Janet's Judge, and Snapshot tests walks through it.

Editors

clove nrepl serves nREPL on port 7888 and writes .nrepl-port, which CIDER, Calva and Conjure pick up. Evaluation and load-file work in all of them. Completion and lookup work too, but clove doesn't implement the full CIDER middleware, so some editor features won't.

The repo has a .clj-kondo/config.edn tuned for clove. clove lint does a quicker check, reading the files without running them, and clove fmt formats with cljfmt if you have it.