A couple questions that arose when I was looking into this:

• should we remove the static files from the collection objects instead? It seams cleaner to have them in one place only
• I'm not sure but this might be related to [Bug]: jekyll reports two files have conflicts but they are the same file.  #9139?

Thank you for submitting this proposal @yboulkaid, but I am not liking the double iteration in the new implementation.

Can you give an example of the stated false positives? For context, two Ruby strings "foo" and "foo" could be two different objects (different object_id) unless we are dealing with frozen string literals.
Similarly, two Jekyll resource objects with seemingly same attribute(s) could be two different instances of the same class if the file has been read into memory twice.

To maintain plausible backwards-compatibility, we cannot remove existing behavior, only add to it.
The commit that introduced the duplicate reference to static files has not been released yet. So that change can be freely altered.

In #9139, I am suspecting one of the used plugins because the path attribute is different and they are using a released version of Jekyll.

Thank you for submitting this proposal @yboulkaid, but I am not liking the double iteration in the new implementation.

I agree that the double iteration isn't very elegant, but it's the only way to remove duplicates after they have been added :/ Do we have any performance benchmarks we rely on to see how much of a hit this can be? I suspect not that much, but I only have managed small Jekyll sites

Can you give an example of the stated false positives? For context, two Ruby strings "foo" and "foo" could be two different objects (different object_id) unless we are dealing with frozen string literals. Similarly, two Jekyll resource objects with seemingly same attribute(s) could be two different instances of the same class if the file has been read into memory twice.

The test in this PR can reproduce the false positives. The duplicates also have the same object_id, so they are "true" duplicates. I checked by putting:

# in jekyll/commands/doctor.rb:130
# in destination_map(site)

binding.irb if thing.path.match?(/extensionless_static_file/)

To maintain plausible backwards-compatibility, we cannot remove existing behavior, only add to it. The commit that introduced the duplicate reference to static files has not been released yet. So that change can be freely altered.

If we decide that static files in collections should be included both in collection.documents and site.static_files, then I don't think we can escape some form of de-duplication step when looping through all objects in each_site_file, as by design this now includes both sets.

Looking through this again, I realized that we were also actually writing the same file twice even in Site#write, since we are also using the same each_site_file to loop through the files.

So as it stands today we can't assume uniqueness when looping in each_site_file, which could pose issues in other places. The message output by Doctor was just the most visible symptom.

(For context, adding static files the the collection document list has been added in 2014)

In #9139, I am suspecting one of the used plugins because the path attribute is different and they are using a released version of Jekyll.

True 👍

If we decide that static files in collections should be included both in collection.documents and site.static_files, then I don't think we can escape some form of de-duplication step...

It's not collection.documents but collection.files that holds references to static files from a collection.

The root cause behind the issue is that Site#documents contains both collection.docs and collection.files.

As of now, I am not able to come up with a better solution than what you have proposed. Yet, I am not willing to approve your proposal in current state either.

It's not collection.documents but collection.files that holds references to static files from a collection.
The root cause behind the issue is that Site#documents contains both collection.docs and collection.files.

You're right, thanks for the correction. Now we have an idea about what the issue is at least.

As of now, I am not able to come up with a better solution than what you have proposed. Yet, I am not willing to approve your proposal in current state either.

Could you elaborate about what in the proposed solution poses a problem? Is it the concept of the double loop, the performance implications, or something else? I can spend some more time looking into this, and knowing what is wrong with this solution helps giving me a direction

Could you elaborate about what in the proposed solution poses a problem?

Primarily the double iteration. Secondarily (to a very less extent because modern Ruby versions are better optimised for handling the scenario) stashing the block as a proc for subsequent use.
Regarding the double iteration, you may be able to better understand my disproval if you take a closer look at Site#docs_to_write and Site#static_files_to_write. Both of definitions are (intentionally) non-memoized yet involves iterations of their ow on every call.

The ideal solution according to me involves defining new variables and methods to ensure consistency and backwards-compatibility but that could involve making changes to classes other than Jekyll::Site.

I rewrote the each_site_file method in a way that avoids both the double loop and passing the &block as an argument to the method. Does that address your concerns?

Thank you for getting back at tackling this @yboulkaid. I must admit that the current diff looks a lot better than the prior ones. Jekyll uses a similar procedure to traverse through layouts of a rendered resource.

However, the attached test needs some cleaning up. You have currently created a new context under previously existing context. That is not correct since we're testing a separate / different Doctor command procedure. Additionally, the expectation is a bit vague: static files in a collection should not trigger false positives ?? I have no idea what the test is about until I read the block contents.
One of the reasons, is that we are not changing the Doctor command implementation at all but something in Jekyll::Site. Consider the following scenario:

# test/test_site.rb

context "static files in a collection" do
  should "not be revisited in `Site#each_site_file`" do
    ...
  end
end

Thanks for the detailed comments. I agree that the tests were not optimal, they were testing a symptom rather than the root cause (because that's what got me into this rabbit hole 😄)

I rewrote the test following your recommendation, I think it became much nicer :) What do you think?

The diff looks very nice now. 👍
But I am still request an additional review to rule out familiarity bias.

Thanks @yboulkaid
@jekyllbot: merge +fix