I knew I was being daft!

@dominicm, I'm curious, are you working on something based on unrepl?

Speaking of :set-source, there’s the open question of knowing reliably when it starts to apply. Scenario; the user submitted three benchmark forms at once, the 1st form is being evaluated, the other two are waiting to be read. Meanwhile the user wants to evaluate some form from a file (with corresponding source info set). This raise at least one edge cases and several questions: • if :set-source occurs on a separate connection it shouldn’t apply bafore the 2 pending forms are evaluated • evaluating from file also require to set the ns, should eval from file and repl share the same session? • if :set-source is async how do we know when it applies (or when to apply it)?

Currently I’m leaning towards adding an extra param to :set-source: the offset where it must become active. Still unconvinced

Yeah. I had wondered about the async and multi form versions of that. This, I suppose, is another part of the trade off of not "wrapping" input forms in a special way.

wrapping on the “aux” connection is acceptable

It’s a related concern but there’s something that bugs me with eval-form-from-file

after some time the lines/cols are all wrong, ideally the editor should version files at each eval-form and makes the version id part of the file path

Back to my question on :set-source, should “eval-form-from-file” happen on the repl connection? Should it alter *1 and friends? Should it be concurrent? In my experience (I’d like to know yours) the only integration I care about is having it appears in the repl logs so as to be able to review the log more easily. This may be a pure client feature

I use *1 and *e for things evaluated from a file

From a tooling perspective *e is invaluable for formatting exceptions

True, most of the time it’s only defstuff but when it fails *e is useful (would a nice exception browser meets this need?)

@cgrand perhaps. I hear Cider's is really good (you can click on lines and stuff).

@dominicm yeah but it asks too much of the client

@cgrand indeed. I'm starting to look at how vim can even have something close (although, it is based on a generic exception viewer I think, not dived in yet)

so new small change, the :prompt message lists :column and :offset.

This si interesting for two reasons: the client can keep track of the progress of evaluation without counting chars itself

By checking if last :echo‘s :offset+ :len is equal to current prompt :offset

the second reason is that you can easily spot when :set-source applies (you don’t have control over when it applies yet)