Having following code:

(def rng1 (java.util.Random. 2288)) 

^{::clerk/no-cache true}
(clerk/example
 (repeatedly 3 #(.nextDouble rng1))
 (repeatedly 3 #(.nextDouble rng1))
 (do (.setSeed ^java.util.Random rng1 2288) rng1)
 (repeatedly 3 #(.nextDouble rng1))) ;; <------ seed is not set!

(def rng2 (java.util.Random. 2288)) 

^{::clerk/no-cache true}
(repeatedly 3 #(.nextDouble rng2))
^{::clerk/no-cache true}
(repeatedly 3 #(.nextDouble rng2))
^{::clerk/no-cache true}
(do (.setSeed ^java.util.Random rng2 2288) rng2)
^{::clerk/no-cache true}
(repeatedly 3 #(.nextDouble rng2)) ;; <------- seed is set back to 2288

Moreover. With this code (aftere clearing cache):

(def rng1 (java.util.Random. 2288)) 

^{::clerk/no-cache true}
(clerk/example
 (.nextDouble rng1)
 (.nextDouble rng1)
 (do (.setSeed ^java.util.Random rng1 2288) rng1)
 (.nextDouble rng1))

(def rng2 (java.util.Random. 2288)) 

^{::clerk/no-cache true}
(.nextDouble rng2)
^{::clerk/no-cache true}
(.nextDouble rng2)
^{::clerk/no-cache true}
(do (.setSeed ^java.util.Random rng2 2288) rng2)
^{::clerk/no-cache true}
(.nextDouble rng2)

Everything looks ok, unless I add another

^{::clerk/no-cache true}
(.nextDouble rng2)

Then everything is messed...

(it's not deterministic though 😕 )

Global no-cache solves the issue. well, doesn't solve...

@U5H74UNSF can you look at this? thanks!

@U1EP3BZ3Q unlikely I’ll have much time to look into this before next week

don’t fully understand looking at the pictures what the problem is

is it related to clerk/example or not?

Originally I thought it was related to clerk/example

My later experiments have shown that it appears also with regular forms

could you create an issue with the minimal repro?

Sure!

thank you!

well... 🙂 when I apply no-cache to def forms as well it works. I don't know why I missed that previously :/

Sorry guys for generating false positive case

(anyway still experimenting)

Ok, found it. It affects clerk/exemple only when sequences are returned: https://github.com/nextjournal/clerk/issues/435

Hi @U1EP3BZ3Q, thanks for the repro. I think this behaviour is not strictly due to a caching issue, but rather something in clerk’s eval phase which forces the top-level lazy-sequences constructed with repeatedly (Case 2) but not when they’re nested into other collections, like it’s the case for the code emitted by the clerk/example macro (Case 1). If in your example, you force your lazy seqs with e.g. a doall in your first case, you get back the expected behaviour. In Case 3, no lazyness is involved. That said, if we’d expect lazy seqs to be forced only at display time and not at eval time, then .setSeed would always be called before calls of .nextDouble. Is this correct? I don’t have a clear picture of what’s going on exactly.

I've started to suspect that lazyness could be involved here but didn't check this. Yes, if lazy seqs are evaluated at display time then setSeed is called before generating random numbers.

Anyway it would be great to have clerk/example evaluating examples step by step.

> if lazy seqs are evaluated at display time and I don’t think this is the case, at this spot https://github.com/nextjournal/clerk/blob/23214a719e191c822901b749bf5f8be867421066/src/nextjournal/clerk/eval.clj#L139 I think we force seqs to check they stay within some limits prior to computing their hash and that’s right after their evaluation

> evaluating examples step by step but then you’d also have to use eager seqs, right?

It works like that without clerk/examples , ie. forms are evaluated in the order they appear (case 2).

I treat clerk/examples as a nice way to display (kind of) repl session, so the order of evaluation and print should be preserved.

the order is correct in both cases, but the clerk/example macro wraps single forms in a vector which gets evaluated all at once

Yep, got it, so the only way is to force lazy seqs to be eager. Wrapping into vec works.

Still, from dumb user perspective there should be no difference between forms wrapped into clerk/example (or possible other forms) and unwrapped. 🙂 Just the opinion.

Wrapping into clerk/table also reveals this issue.

> there should be no difference I agree 🙂, the somewhat confusing part is that the unwrapped behaviour is somehow off wrt clojure lazyness

Ok, corrected issue title and left a comment as a summary of our discussion.

Thank you

Thanks for the discussion!

Example: Say I’ve got notebooks/mathpunk/some_problem/day_01.clj. If I start a day_02 file, I can reference that namespace. But if I close the clerk watcher, when I go back and edit day_02, it doesn’t find the namespace, until I go and edit day_01, and then day_02 will be able to find it. Okay so, maybe instead the Right Way is to extract functions for reuse into the src directory. But when I’ve edited something in the src directory, I’m not seeing Clerk automatically show file while I’m editing it.

I could be holding things wrong, I could be a few commits behind, etc

The issue might be that “notebooks” is not listed in the paths specified in deps.edn

So when you require, day1 isn’t seen on the class path until you manually load it

The watcher issue is probably that you’ve only passed “notebooks” to the watcher in dev/user.clj. You can add “src” to that list if you like!

Apologies if this is my template and it’s been confusing; let me know what we should change here to make the experience better. Adding notebooks to deps.edn seems like a no-brainer here

hmmm

:paths ["src" "dev" "notebooks" "artifacts" "data"]

I think notebooks started out in the template

as did src

Interesting… @U5H74UNSF any ideas?

it’s POSSIBLE i added these things after observing that…

sidenotes, I haven’t gotten the hang of dev dependencies…. do people use that for their repl utility belts then?

I’m going to try and reproduce what I’m complaining about, lest I encourage you to chase an unmannered goose

When I am developing a library I will have dev dependencies for things like test runners, or maybe something like SCI where I don’t want to force the dependency on someone

it really looks like editing something in src is not displaying it

while notebooks, yes

there’s two .edn files right? a bb.edn and a deps.edn?

https://github.com/mentat-collective/clerk-utils-custom-template/blob/796a0a96cee59f81adb26ce76ff6e808919e3308/dev/user.clj#L21

This setting controls what directories get watched

Some people don’t like source to get watched since maybe it’s a collection of utilities for example and you don’t want Clerk navigating away from your notebook while you work on that

ohhhhh ok

The paths here should definitely build the class path. https://github.com/mentat-collective/clerk-utils-custom-template/blob/796a0a96cee59f81adb26ce76ff6e808919e3308/deps.edn#L1

I thought that it was the :paths in deps.edn that mattered here

Is there a chance that your namespaces don’t match the file path?

I’ve barely looked at dev.clj

there is absolutely a chance my namespaces don’t match the file path

That might be it. Is this a Clojure problem too? Like is it just clerk that can’t find notebook 1 from 2, or the repl as well

I guess you wouldn’t know if you’re only using the bb task! That is another guess though, that would break the namespace resolution in require

Should be mathpunk.some-problem.day-01

Cool. Notebooks can refer to other notebooks, if the user has the namespaces right

And if I want editing src to trigger a view change, put it in dev.clj

Dev/user.clj, that’s right

thanks, looks like it was a liveware problem :)

Haha well, the tech should mold to us ideally

But keep notes on what you want the “explore problems with Clerk” experience to feel like, I’m super interested

will do

broadly: • explore in a literate way — writing is as or more important than coding to get me to understand a new thing • reference previous work • extract what’s reusable to a library or “script” file

Hard agree on all! I want a world where maybe 3 isn’t as necessary (the extract part), or at least the extraction brings along a bunch of the writing and exploration

and probably • generate a ✌️site ✌️ or something for people who aren’t running Clojure

That should work from the template already

And I agree it’s critical

yes, I just haven’t gotten there yet. Also we’re on Gitlab rather than Github, there might be differences in the way they do their “Pages”

I want a https://Maria.cloud mode that opens a clerk notebook for live editing / exploration there

totally

also i should mention, I haven’t even been turning on my repl in my editor, just because I think I misconfigured some stuff

so it’s nice to be able to still get this live coding experience, but i also should take some time to fix that