Error highlighting does not work

[whysoserious]

Steps to reproduce

• Create any lein project eg. lein new quil my-sketch
• Open any .clj file eg. src/my_sketch/core.clj
• M-x cider-jack-in
• Make any syntax error which should be highlighted
• M-x cider-eval-defun-at-point - yields no highlighting
• M-x cider-eval-region - yields no highlighting
• M-x cider-eval-buffer - works as expected - I can see red underscores

Additional data

My emacs config:

https://github.com/whysoserious/my-emacs-config/blob/master/init.el#L326-L374

M-x version

GNU Emacs 24.4.1 (x86_64-apple-darwin13.4.0, NS apple-appkit-1265.21) of 2014-10-21 on builder10-9.porkrind.org

repl

; CIDER 0.9.1 (package: 20150625.946) (Java 1.8.0_25, Clojure 1.6.0, nREPL 0.2.10)
WARNING: The following nREPL ops are not supported:
extract-definition rename-file-or-dir stubs-for-interface warm-ast-cache
Please, install (or update) refactor-nrepl and restart REPL.
You can mute this warning by changing cljr-suppress-middleware-warnings.
WARNING: clj-refactor and refactor-nrepl are out of sync.
Their versions are 1.1.0-SNAPSHOT and 1.0.5, respectively.
You can mute this warning by changing cljr-suppress-middleware-warnings.
user>

[Malabarba]

There are two things going on here...
This initial comment was just stupidity on my part, the following is what I've gathered now.

The problem here is that the eval op produces an uninformative error.
Right now, highlighting is done entirely by calling cider-handle-compilation-errors on the error message sent by the REPL. The error message is very uninformative in the case of interactive eval:

  "CompilerException java.lang.RuntimeException: Unable to resolve symbol: reduc in this context, compiling:(/tmp/form-init1923343092107590387.clj:2:3) \n"

The cider-handle-compilation-errors is called as part of the error callback, which is the same callback that calls the function that creates the stacktrace. Now, the stacktrace has a lot more information to work with, so it would be possible to improve things by applying the error overlay as part of creating the stacktrace buffer.
This wouldn't handle all cases (there are some errors for which even the stacktrace doesn't contain enough information), but would handle some.

[bbatsov]

There's not much cider can do, the eval op simply doesn't check for errors (I think it doesn't compile the function?). It's not just about highlighting, you don't even get a stacktrace buffer. If you issue load-buffer instead, then you will get a stacktrace and highlighting, because the load-file op does compile.

In order to trigger an error, you have to actually use the function, for instance by evaluating (should-error). Then you will get an error and a stacktrace, but you still don't get highlighting. That's the second problem.

I'm pretty confused by this command. eval actually compiles the code and reports compilation errors just as load-file does. The problem is that eval doesn't attach proper locations to the stacktraces frames, because as far as it's concerned they happened in the temp (tooling) file used to eval (compile) the code in question.

"ArityException Wrong number of args (1) passed to: core/reduce clojure.lang.AFn.throwArity (AFn.java:429)\n"

Hmm, that doesn't sound quite right. I recall the messages were pretty much the same.

[Malabarba]

@bbatsov

I'm pretty confused by this comment. eval actually compiles the code and reports compilation errors just as load-file does. The problem is that eval doesn't attach proper locations to the stacktraces frames, because as far as it's concerned they happened in the temp (tooling) file used to eval (compile) the code in question.

If I evaluate the defn above with any of the eval commands, I get no errors and no stacktrace frames. I set (setq nrepl-log-messages t) and I verified that the only messages I get back form the nrepl are the "done" status and the return value (which is #'should-error, as if everything were fine). Is that not what you get?

My understanding (from looking at the source of interruptible_eval.clj) is that the "eval" op simply calls the eval function on the code.

[bbatsov]

But eval in Clojure is actually compile. There's no evaluation, only compilation.

[Malabarba]

Strange. But I still don't get errors when evaluating that. I also don't get errors if I paste that in a regular lein repl (not sure if it's related).

[bbatsov]

The definition or its invocation? For me evaluating invalid code yields the expected results:

(shit)

  err  "CompilerException java.lang.RuntimeException: Unable to resolve symbol: shit in this context, compiling:(/private/var/folders/f0/cm24s0qd0s5cch5jvj6ctnb00000gn/T/form-init8023551865063802074.clj:1:1) \n"
  id  "8"
  session  "3253b0d0-386c-4169-8936-c1181f8e9924"

[bbatsov]

Hmm, you're right - with your example there are no precise code locations. Likely a Clojure bug...

[Malabarba]

I'm taking about both separately.

1. Evaluating a definition that contains invalid code doesn't yield errors. But doing load-file on the file that contains that definition does yield an error.
2. Evaluating the invocation does yield an error (that's the second part of my comment).

[bbatsov]

Evaluating a definition that contains invalid code doesn't yield errors. But doing load-file on the file that contains that definition does yield an error.

Hmm, that definitely sounds like some bug. //cc @cemerick

Evaluating the invocation does yield an error (that's the second part of my comment).

Although this error lacks column info for some reason.

[Malabarba]

Although this error lacks column info for some reason.

Frankly, that's the least of our problems. :) You'll note that the file info is simply wrong. It points to this AFn.java file, instead of pointing to a tooling file (which is where the function was defined).

ArityException Wrong number of args (0) passed to: core$reduce  clojure.lang.AFn.throwArity (AFn.java:437)

[bbatsov]

This probably has to be reported upstream. If this is not a bug I don't know what it is. :-)

[Malabarba]

Straight to clojure right? Or to nrepl?

[bbatsov]

http://dev.clojure.org/jira/browse/CLJ

[Malabarba]

Actually, let me rectify. I was doing something stupid before, and there's actually no difference for me between eval or load-file. Neither of them throw errors when defining a "bad" function. Both only throw errors when the function is actually called.