This was copied from here… moved it because it needs its own git repository.
Want to make a tracing function which uses nrepl-discover.
Just ignoring implementation issues for now. How would I really like this to work?
Would want to bind different execution environments to different invocation keys. I.e., I run this, and this test, perhaps in a different ns, gets run. Tracing happens when specified conditions are met. Can easily change the environment associated with a key.
There is a binding which takes one of these invocation keys and invokes it. Another for reporting/editing the environment associated with it.
Tracing data should go into a clojure data structure, as well, with a handy way of reaching it from the repl. There should be an option to turn off the stdout output of it.
The clunky part of this at the moment is the editing of the execution environment. How to specify the namespace/function to run? How to specify the tracing condition? How to specify the variable to store the trace data in? Or maybe how to access it from the repl? Maybe put them in a dbg namespace? Like dbg/t?
There needs to be a fast way and a slow way of doing these things. Slow way: open up a customization dialog which shows all the options.
Fast way: A specific binding for setting each, which takes the execution environment binding and runs it.
I am going to just do the fast way for now. Simpler, and I think most people will get it. C-c t will be the prefix.
| C-c t n <key> | bind <key> to a new execution environment. If region is |
| active, that becomes the forms to trace. Otherwise, just | |
| inject locals report at point. | |
|---|---|
| C-c t f <key> | ask for the command to run when it’s called. By default, |
| it will load the current ns, a la C-c C-k. If a bare var | |
| is passed, it will be treated as a function. If a list, | |
| the list will be treated as a function call. | |
| C-c t N <key> | will set the ns to that of the current buffer. With |
| prefix arg, it will ask for the ns to set it to. | |
| C-c t c <key> | set the tracing condition. Injected into the current |
| context. | |
| C-c t o <key> | toggle whether output is sent to the repl or not |
| C-c t d <key> | show the var the tracing output is being dumped to |
| C-c t r <key> | report all the data associated with this execution |
| environment. Allow it to be edited, too. | |
| C-c t m <key> | Set the forms which the tracing/debugging apply to. |
| This should be paredit-style, but since I only want to | |
| paint parts of the tree, just use arrow keys: up means | |
| the containing form, down means the setting prior to the | |
| last up, or nothing, left and right mean the forms in those | |
| directions. See up-list et al. in lisp.el. | |
| C-c t M <key> | Add trace to all forms touched by the currently marked |
| region. Error if region is not active. | |
| C-c t F <key> | Assume form under point is a function, and trace its inputs |
| and outputs. | |
| C-c t l <key> | Go up the tree to find the first let form, and trace its |
| definitions. |
When the tracing condition causes an exception, that should be reported wrapped, because it won’t show up in the code. Ideally, a message should be sent back to emacs which can report it specially.
The traces should contain timestamps and all info about the forms being traced, condition for tracing, etc.
First stage: Take the current point, and report the values of the locals at that point.
Next stage: Take all forms in the region, and instrument them to return their values.
Next stage: Take all forms in the region, and instrument them to drop to a debugging repl, step between them, etc.
I implemented the second stage of the above plan first. There’s been some positive response to this.
Someone suggested that it would be good to use the wrapping macro to make a test-coverage tool. It looks like cloverage already works this way. It appears that that is using something like clojure.walk/macroexpand-all, which is erroneous in some cases. But it looks like if I wanted to do this, I would be better off improving cloverage.
The main problem I ran into is how to specify the test code to be run when the tracing is requested. There needs to be a few ways of doing this. The existing one is just to call a function in the repl, which takes a function with no arguments. It should be possible to send a form for this just by pointing it out in emacs. It should also be possible to name a var to be called. And it should be possible to choose just among test vars, with higher priority given to vars in the project namespaces. There are also autotest tools out there, but I think most of them work by checking the disk for changes, so I don’t think they’re going to help, much. nrepl-discover uses nrepl-ido-read-var to choose vars. I could start with just that, I guess.
I’m just doing this with a single testing context at the moment.
For a straight report of locals, how should I decide where to inject the report? The user could have point in a quoted list or something, for instance.
The rule should be that if point is on a parenthesis, the form containing that which is actually evaluated should be wrapped. If the point is on or next to a symbol, the evaluated form containing that should be wrapped.
I guess I could do the same macro wrapping trick as before, and pass information up the chain using metadata on the forms??
Can get the top-level form with nrepl-region-for-expression-at-point. Returns start and end positions in a list.
Perhaps the problem here is that I am trying to do too much work which ought to be done by the user?
I’m going to punt on this, for now. I’m going to require the user to specify the form to wrap, wrap it with a macro which reports when it’s been run, and throw an error if the macro doesn’t run.
Still need to determine how many forms are in that section, but I guess I could do that in the clojure macro I wrap the whole thing in.
This is going to be tricky: The positions are going to be moving around. Emacs has data structure for keeping track of this, markers.
So I need a way to
This needs to be a ref of some variety. Probably needs to be accessed through an agent, since I really need to pipeline
I guess I am going to have to specify the marker insertion type, but I am not sure whether the pointer should move to the start or end of inserted text.
nrepl-discover passes the top-level defun, the current namespace, the region which the user wants traced. The clojure side inserts a macro around that. It is the user’s responsibility to get the positions of the region right, for now.
The clojure code uses the LCA method to identify portions of the macroexpanded code which come from the region.
So, first thing I need is a reader which just reads. That means turning off the syntax quote in clojure.tools.reader/macros and clojure.tools.reader/read-ctor.
Then, I need to use the earlier code to map the macroexpanded form back to the target region. Actually, there is a better way: post-traversal combination of the range containing symbols below each node. That way, I can figure out exactly which forms need to be wrapped. Awesome. I can probably
Once I’ve figured out which forms to wrap (indicated in the metadata by {::wrap true}), I can just use the previous wrapping code. To start with, I should probably just trace it, with the form as the name.
The way the tracing should work is, I have a recursive macro like before. This time, if the form has ::wrap true, I should wrap the form in (trace form form). Let’s start there. I probably want my own trace function, but maybe I can overwrite trace/tracer for this.
This is ideally going through nrepl-discover. It is going to want to know the current ns. nrepl-interactive-eval does this by passing (nrepl-current-ns). I probably want to do the same.
Actually, nrepl-discover’s framework passes this for me automatically.
What is the right way to get my hands on nrepl-discover.el? Should I just put a copy in the local directory? Might as well, hmm? Instructions need to say “put these TWO files somewhere emacs can find them. It’s just a prototype at this stage, anyway.
Note that there is nrepl-send-request-sync for synchronous communication with the server.
https://github.com/emacsmirror/expand-region/blob/master/clojure-mode-expansions.el
Extend tools.reader to thread the dispatch maps through the function calls, so that they can be changed in a thread-safe way. (User code may be using tools.reader. This argues for putting a copy of tools.reader in a subordinate ns so that that can’t happen, actually.
Given the dependency diagram for tools.reader, I think I can just take a copy of reader.clj, and leave the rest as-is. This might have a way to pull reader.clj in and still allow me to track changes to it. I will do that later, though.
XXX troncle.el:11:1:Error: Cannot open load file: clojure-mode.
Need explicit package requirements. Doesn’t matter that much, since anyone interested in this probably already has clojure-mode installed.
This was fixed by Steve Purcell’s recent patch.
The claim in the README.md that this takes vars from anywhere is currently inaccurate. nrepl-ido-read-var only takes vars from the current ns. Should extend troncle-set-exec-var to also ask for the ns, with anything similar to the current ns near the front of the list.
This is actually already possible, I just have to update the
Will need to do this to get it on melpa.
- Bump version numbers in project.clj and troncle.el
- lein deploy clojars
- upload troncle.el to marmalade. (There is an automated way to do this.)
