[load-file] Reimplement load-file again

[alexander-yakushev]

This is the first piece of the jigsaw puzzle called "how to retain the debugger instrumentation during C-c C-k" (see clojure-emacs/cider-nrepl#946, https://clojurians.slack.com/archives/C04CAKAGADU/p1757684821897969).

To make it possible for the debugger to instrument forms when doing C-c C-k we have to abandon the Compiler/load approach for load-file middleware and instead let the interruptible-eval middleware be the driver for everything. Interruptible-eval must read through the whole file form by form and call the custom eval function on each form.

By doing this there are a few risks:

• We change a quite fundamental part of nREPL again. Not the most critical, I think – eval is still the most important, and we changed it too in the past, everything seems to be alright.
• There could be some minor behavior changes in how load-file will work, but I tried to make it as resembling Compiler/load as possible.
• I had to drop some alternative way to load a file which I marked as deprecated for 1.4.0. This is a short notice for sure, but nREPL updates don't move that fast anyway. Whoever might be impacted by this will likely stay on 1.0.0 or 1.3.0 for a long time and will be able to update to 1.4.0 safely still.
• I'm also removing nrepl.helpers/load-file-command. This looks like a command-line REPL affordance to load files using the old old way. There is no mention of it in the documentation, and no occurance of using this funciton on Github. I'd just remove it and wait for anybody to complain. If this is still useful to somebody, we can provide a better helper for that.

Don't need to merge this yet as this is experimental, I should definitely dogfood this, and backwards-compatible changes are required in cider-nrepl too.

---

• You've updated the changelog(that only applies to user-visible changes)

[bbatsov]

My main concern with the PR is that now looks a bit strange that we have a :file param and a :file-name param. Perhaps we should either rename the old param with a more descriptive name (as an alias, keeping backwards compatibility) or expand the description about the need for both.

[alexander-yakushev]

I'll edit the docs. Renaming, even with an compatibility alias, might be too much for the holy eval. Maybe, one day, we can do a sweeping take on it where we can clean up all the cruft – rename it from interruptible-eval? (gasp!) Maybe in nREPL 2?

[bbatsov]

Perhaps. I'm mostly concerned about this from the perspective of the people looking at the current implementation as the spec, but I guess we can do a cleaner "official spec" around nREPL 2 as well.

But I agree that down the road we can revisit the name of the eval middleware and potentially the need for load-file as a separate op.

[bbatsov]

Overall the code changes look good to me, but you'll have to dogfood them for a while before we can merge this, given the the relative importance of this functionality. I always start with load-file before switching to more granular evals in my own Clojure workflow. (I also often use load-file to reset all breakpoints in a namespace)

[alexander-yakushev]

So, you are saying that you rely on C-c C-k not enabling the debugger, even in forms where you have a #dbg? Or you mean that you use it to clear debugged functions that were compiled with C-u C-c C-c (so, no explicit #dbg in the code)?

[bbatsov]

Or you mean that you use it to clear debugged functions that were compiled with C-u C-c C-c (so, no explicit #dbg in the code)?

This - I was just sharing how I normally use load-file myself. I never got in the habit of using #dbg, as it doesn't exist in Elisp and I often switch between Clojure and Emacs Lisp. I do agree, that #dbg breakpoints should not disappear on load-file, so I'm fine with this PR in general.

[alexander-yakushev]

@bbatsov I'm going to merge this and publish an alpha1 release, just need a thumbs up from @ericdallo. On my side, nothing has broken in the days I've been using this.

[bbatsov]

Sounds good to me.