Changelog

All notable changes to ClojureElisp are documented here.

Format follows Keep a Changelog; versioning follows Semantic Versioning.

0.7.0 - 2026-09-05

The theme of this release is that the interactive loop is the primary way to write ClojureElisp, and the tooling around it is no longer heavier than the elisp REPL it replaces.

Why

Public feedback on the project said, in substance: an external transpiler process is not ergonomic while you are writing elisp functions, unless there is a way to invoke the transpiler, and that comes with a lot of tooling when elisp already has a good REPL model.

Half of that was already wrong about ClojureElisp. The nREPL middleware (clojure-elisp.nrepl) and the CIDER minor mode (resources/clojure-elisp/cider-clojure-elisp.el) have shipped for several releases: C-c C-c compiles the form at point and evaluates the resulting Elisp in the running Emacs, redefining the function in the live image with nothing written to disk. That is the same loop the critique credits elisp with.

The other half was fair, on three counts, and this release addresses all of them:

1. The README documented the interactive loop in exactly one table row, with no workflow section. A reader could not discover it, so concluding it did not exist was reasonable.
2. Setup really was heavy: a JVM, a .nrepl.edn edit, cider-jack-in, then M-x cider-cljel-start.
3. The interactive path had defects that made it feel second-class, the worst of which silently disagreed with the compiler.

Added

• clel nrepl, a ClojureElisp nREPL server that needs no JVM. Starts in roughly 275 ms under Babashka. Connect with M-x cider-connect-clj; there is no deps.edn to write, no .nrepl.edn middleware entry, and no jack-in. Implemented as clel.nrepl-server, a bencode socket loop speaking clone, close, describe, ls-sessions, eval, load-file, cljel-start and cljel-stop.
• clojure-elisp.nrepl-kernel, the transport-independent core: session registry, compile modes, and op semantics. handle-op returns a vector of response maps and writes nothing, so a transport only has to merge its correlation keys and serialize. Both servers now run this one definition of compile-code.
• compile-string-in-ns (in clojure-elisp.compile, re-exported from clojure-elisp.core): compiles forms in the namespace context of an (ns ...) source string, emitting the forms alone with no file header and no (provide ...).
• cider-cljel-runtime-file, a defcustom pointing at clojure-elisp-runtime.el for setups where it is not on load-path.
• README "Interactive Development" section, placed ahead of Features, Installation and CLI Usage, with both server routes, the keymap, and a worked session that builds an interactive fence-region command from the ns form through M-x to clel compile. Every code block in it is verified compiler output.

Fixed

• C-c C-e and C-c C-k no longer define different functions. handle-eval hardcoded :expr mode while handle-load-file used :file, so evaluating (defn greet ...) inside (ns my.pkg) installed greet while compiling the same buffer installed my-pkg-greet. The running image and the compiled artifact disagreed about every namespaced definition. The CIDER client now sends the buffer's (ns ...) form as cljel-ns and the kernel compiles in that context, so all four paths agree.
• First interactive evaluation no longer dies on void-function clel-str. Expression mode emits no runtime require and cider-cljel-start loaded nothing, so any first eval touching str failed. cider-cljel-ensure-runtime now loads the runtime from load-path or cider-cljel-runtime-file, and reports clearly when it cannot find it instead of leaving a void-function to surface later.
• Evaluation results appear at point. The response handler accepted a point argument it never used and sent everything to the echo area. It now renders a CIDER inline overlay, falling back to message where that is unavailable.

Changed

• clojure-elisp.ast/gen-node resolves malli.generator lazily through requiring-resolve rather than requiring it at load time. It is used by tests only, and requiring it put clojure.test.check on the compile path, which no lightweight host provides. Pinned by a test asserting no compile-path namespace aliases malli.generator.
• bb.edn now puts src and resources on :paths with hive-dsl and malli, so the compiler loads in the Babashka process rather than shelling out to a jar.
• clojure-elisp.nrepl is now a transport only. wrap-cljel and the handle-* functions delegate to the kernel; the public surface (cljel-sessions, cljel-active?, compile-code, handle-eval, handle-load-file, handle-start, handle-stop, wrap-cljel) is unchanged.
• clojure-elisp-runtime.el regenerated from runtime.cljel with the current emitter. The only differences beyond the version header are (when x y) forms emitted as the equivalent (if x y nil); behaviour is identical. The checked-in file had been generated by an older emitter.
• Version headers in clojure-elisp-mode.el and cider-clojure-elisp.el were stale at 0.5.0 and now track the release.

Notes

Sessions on the standalone server start with compilation active, because that server has no Clojure evaluator to fall through to. M-x cider-cljel-start is therefore no longer a required step there, and cider-cljel-mode is only about keybindings. The JVM middleware is unchanged in this respect: a session there still opts in.

ClojureWasm was evaluated as a faster host, since cljw starts in 36 ms and already ships cljw nrepl. The transport is ready for it and needs no changes. malli is not: malli.core/-memoize reaches java.util.concurrent.atomic.AtomicReference, which ClojureWasm does not provide, and that is malli's own memoization rather than anything ClojureElisp can route around. Clearing the earlier blockers in that chain is what produced the malli.generator change above, which was worth making on its own.

Verification

598 tests, 3021 assertions, 0 failures. The standalone server was driven over a real socket through clone, namespace-aware eval, bare eval, load-file, a compilation error and describe. Neutralizing compile-string-in-ns so it ignores its namespace argument turns the new suite red (6 failures, 1 error), so the namespace-parity coverage is not vacuous.

0.6.2 - 2026-07-21

Fixed

• The release job no longer writes the version back to a protected main. The v0.6.2 release had failed because git push origin HEAD:main is rejected by branch protection. README install coordinates use a placeholder and the Clojars badge carries the current version, guarded by a test that fails if a concrete version is ever pinned again.
• load-file compiles as a whole file, and every compiled form is evaluated.

Changed

• Clojure 1.12.4, nREPL 1.7.0, cider-nrepl 0.62.2.
• The release patch auto-bumps so main always ships.

0.6.1 - 2026-07-11

Fixed

• The analyzer keeps every subform instead of destructuring it away, closing a class of silently dropped forms.
• Pre-grouped cond is rejected loudly instead of being mis-compiled.

Added

• dev/migrate_cond.clj, a source rewriter for the elisp-cond migration.

Earlier releases

Releases before 0.6.1 are recorded in the GitHub releases and in the Progress Log in CLAUDE.md.