What happens when you start a CLJS REPL?

[thheller]

Hey, just saw this in my Twitter feed.

I recently spent some time with REPL related ideas myself and one of the key requirements for me is that I want to be able to launch a CLJS REPL seamlessly. I like your approach but what happens if you start another REPL that is not aware of this protocol?

Take the default CLJS REPL as an example. It will takeover the *in* and only write to *out*. While that should produce proper [:out ...] EDN forms (assuming you provide a custom *out* binding) that will be the only thing it sends. No :eval or :prompt. The UI I imagine you'd build out of this would seem "stuck"?

[cgrand]

You nested them in the wrong order: (CLJ -> )CLJS -> EDN

[cgrand]

However the real question to address is how to ensure that CLJ -> EDN -> CLJS -> EDN works.

[cgrand]

Right now I'm thinking of having the server sending [:upgrade nil] before yielding control to the inner repl. Then the client can attempt upgrading plain cljs repl to an unrepl cljs repl.

[thheller]

The CLJS REPL has some other hidden "problems" in that print and eval happen in the JS runtime. This makes a custom eval and print tricky without having the code already present in CLJS.

Maybe the REPL should support arbitrary output streams (not just :out) as well. Since in the CLJS world you'll have the CLJS compiler producing output (eg. compile warnings) as well as the connected JS process.

[thheller]

As far as the output streams go, it would probably be useful to make them thread aware to avoid garbled output like

user=> (dotimes [x 10] (future (Thread/sleep 1) (prn :foooo)))
nil
user=> :foooo
:foooo
:foooo
:foooo:foooo

:foooo:foooo
:foooo:foooo


:foooo

So the tool can re-assemble the streams easily.

[cgrand]

That's what the third optional component ("group id") is for in the tuples.

[cgrand]

@thheller 1da6ef7 enables the use of "group id" in the prototype impl.