In 4.1.0a1, compilation throws new RecursiveCompilation error

[gjtorikian]

With zero changes to my setup, other than upgrading to 4.1.0a1, I suddenly can't build my site:

Nanoc::Int::Errors::RecursiveCompilation: The site cannot be compiled because the following items mutually depend on each other:
<Nanoc::Int::ItemRep name="default" binary=false raw_path="output/404.html" item.identifier="/404.html">
<Nanoc::Int::ItemRep name="default" binary=false raw_path="output/articles/3d-file-viewer/index.html" item.identifier="/dotcom/articles/3d-file-viewer.md">
<Nanoc::Int::ItemRep name="default" binary=false raw_path="" item.identifier="/google363a90be6f604622.html">.
....

Compilation stack:

  (empty)

Stack trace:

  0. /Users/gjtorikian/github/help-docs/vendor/cache/ruby/2.2.0/gems/nanoc-4.1.0a1/lib/nanoc/base/services/item_rep_selector.rb:27:in `each'
  1. /Users/gjtorikian/github/help-docs/vendor/cache/ruby/2.2.0/gems/nanoc-4.1.0a1/lib/nanoc/base/compilation/compiler.rb:215:in `compile_reps'
  2. /Users/gjtorikian/github/help-docs/vendor/cache/ruby/2.2.0/gems/nanoc-4.1.0a1/lib/nanoc/base/compilation/compiler.rb:122:in `block in run'
  3. /Users/gjtorikian/github/help-docs/vendor/cache/ruby/2.2.0/gems/nanoc-4.1.0a1/lib/nanoc/base/compilation/dependency_tracker.rb:18:in `run'
  4. /Users/gjtorikian/github/help-docs/vendor/cache/ruby/2.2.0/gems/nanoc-4.1.0a1/lib/nanoc/base/compilation/compiler.rb:121:in `run'
  5. /Users/gjtorikian/github/help-docs/vendor/cache/ruby/2.2.0/gems/nanoc-4.1.0a1/lib/nanoc/base/compilation/compiler.rb:106:in `run_all'
  6. /Users/gjtorikian/github/help-docs/vendor/cache/ruby/2.2.0/gems/nanoc-4.1.0a1/lib/nanoc/base/entities/site.rb:24:in `compile'
  7. /Users/gjtorikian/github/help-docs/vendor/cache/ruby/2.2.0/gems/nanoc-4.1.0a1/lib/nanoc/cli/commands/compile.rb:375:in `block in run'
  8. /Users/gjtorikian/github/help-docs/vendor/cache/ruby/2.2.0/gems/nanoc-4.1.0a1/lib/nanoc/cli/commands/compile.rb:417:in `run_listeners_while'
  9. /Users/gjtorikian/github/help-docs/vendor/cache/ruby/2.2.0/gems/nanoc-4.1.0a1/lib/nanoc/cli/commands/compile.rb:374:in `run'
  ... 16 more lines omitted. See full crash log for details.

I'm going to try and dig around in recent PRs to see what changed but I suspect this may occur for others, too.

[gjtorikian]

I dropped down to 4.0.2 and the error went away. And as one might expect, commenting out item_rep_selector.rb:27 has the site build as well.

[denisdefreyne]

This to me seems like it happens because the cached data in tmp/ has changed formats in Nanoc 4.1, and Nanoc 4.1 misinterprets the format of 4.0. The solution for that would be to bump the file format version for the rule memory store.

Can you do two things for me?

1. Is this for the GitHub developer site? If so, zip the entire directory (including tmp/) and send it to me so I can take a look.
2. Afterwards, delete tmp/ and recompile with an unmodified Nanoc 4.1.0a1. If my hypothesis is correct, the site will compile properly.

[gjtorikian]

Is this for the GitHub developer site?

It is not, it is sadly for a closed source site. 😦

[gjtorikian]

Your hypothesis sadly did not work! I rm -rf tmp/ but it still did not compile.

I can send maybe a small reproducible test case instead of the whole directory? Will that help? I can get it crashing with just a 404.html and another file.

[denisdefreyne]

Yes please! Small test cases are awesome.

[gjtorikian]

I can't repro with the public developer site, so now I'm thinking I've bungled something on my end. Looking into it a bit before providing you with a zip and/or shutting this down.

[gjtorikian]

Interesting.

I have this line in my default layout:

<% @meta_description = @item[:meta_description] || @item[:intro] || safe_embed(@item.compiled_content)[0..200] %>

If I remove @item.compiled_content, the build is successful. Is it possible these are like, getting collided somewhere?

[denisdefreyne]

Ahh, that is possible, and that gives me another idea as to where this error is coming from.

You can replace @item.compiled_content with yield in the layout, which should fix it. (I’m guessing Nanoc 4.1 has a regression where it can’t deal with @item.compiled_content in this case, because it refers to the content after it’s been laid out. Needs further investigation as to where this issue coms from.)

[denisdefreyne]

(It might be the case that Nanoc 4.1.0a1’s behavior is in fact more correct.)

[gjtorikian]

Cool, that'll work. raw_content also seems to work.

Feel free to close this or keep it open if it's something you want to work on? I'm not entirely clear what the problem is, heh. And thanks for your quick replies!

[denisdefreyne]

raw_content will get the initial content without any filters applied to it, so that might not be what you want.

I’ll investigate and when I find that there’s an actual real issue here, but for now I’ll consider this resolved.

[denisdefreyne]

I captured the problem in #762.

[gjtorikian]

Nice!

[denisdefreyne]

4.1 beta 1 has the fix (see #764).

[denisdefreyne]

This is still an issue in 4.1.0b1.

[denisdefreyne]

#766 fixes this. (Hopefully!)