There should be a warning about code outside of modules being slow

[scvalex]

Hi,

I'm was solving Advent of Code in Livebook today and I noticed that evaluation in Livebook was much slower than in iex. Here's a simple example:

1..30_000_000
|> Enum.count(fn n ->
  n > 20_000_000
end)
|> IO.inspect()

This takes about 15s in Livebook, but evaluates in less than 1s in iex.

I was going to ask about this, but then I remembered @josevalim saying in some video that code is much slower to evaluate if it isn't in a module. Indeed, putting it in a module makes it evaluate very quickly just like in iex.

I think this should be documented somewhere. Maybe there should be a warning under a cell that takes too long to evaluate or something.

[josevalim]

The issue is that this is tough to warn. What if it is not the loop in the cell that's slow but rather the function you call? Enum.sort(some_really_large_list) will be slow, but not because of code being evaluated.

We do talk about this on the "Welcome to Livebook" notebook, but I think it is really hard to automate/detect this.

[scvalex]

We do talk about this on the "Welcome to Livebook" notebook,

Oh, indeed you do. I guess I read that a year ago and forgot.

It didn't show up in any search results I tried earlier today (all of which sounded like "why is Livebook slow"). Maybe it's not rendered as HTML anywhere so the only place search engines could find that is in the .livemd on Github.

The issue is that this is tough to warn. What if it is not the loop in the cell that's slow but rather the function you call? Enum.sort(some_really_large_list) will be slow, but not because of code being evaluated.

Fair enough. I guess there's no way to tell without running some profiler all the time, but that doesn't sound worth it.

[josevalim]

Yeah, we should consider including those Livebooks in our docs. /cc @hugobarauna

[hugobarauna]

@josevalim I was planning to repurpose the content inside the Learn notebooks in different areas of our future docs. But it can take a while until I have the time for that.

So, maybe as a workaround, until I evolve the docs, I can create a section in the current hex docs just listing all the learn notebooks we have inside the docs themselves.

Something like that:

What do you think?

[josevalim]

Something along those lines, but we need to be careful because maybe not all notebooks work. See for example that the <i class="..."> example in the image you shared. So we may want to do some pre-processing.

For example, before generating the docs, we could copy all of these notebooks to a separate directory, and we add an admonition text at the top saying "This notebook is distributed with Livebook inside the Learn section. It is reproduced here for discoverability. Cluck on the button above to run it."? And we can use some regexes to remove unwanted content?

[hugobarauna]

@josevalim ok, I added that as a task for me on our internal backlog. As soon as I finish my current task, I'll work on that. 👍