Pruner should check for excluded files without considering to output_dir

[agross]

When a pattern excluded from pruning is contained in the output_dir config variable all files below output_dir get excluded.

Steps to reproduce

1. Set output_dir: output in nanoc.yaml (the default)
2. Set prune.exclude: ['output'] in nanoc yaml
3. Compile site
4. touch output/something
5. Run nanoc prune --yes

Expected behavior

output/something should be pruned.

Actual behavior

output/something remains.

Details

When checking whether a file is excluded from pruning, the output_dir path component(s) should not be considered.

[agross]

Background: I have a site with output_dir: build/bin but there is an external tool that creates files in build/bin/bin and I do not want nanoc to prune build/bin/bin/**/*, hence I added bin to prune.exclude.

[gpakosz]

I believe the behavior you're witnessing is normal. As per Nanoc manual:

The exclude option determines which files and directories you want to exclude from pruning. If you version your output directory, you should probably exclude VCS directories such as .git or .hg.

Now I would acknowledge there's an ambiguity in Nanoc's wording.
Should prune.exclude: ['output'] exclude output or output/output?

The current behavior makes it output/ is excluded from pruning. And by definition, output/something being inside output/ (which is excluded from pruning) doesn't get deleted.

In any case, the exclusion logic is implemented in terms of `String#include? so it's definitely loose and can't achieve what you're asking.

Finally, given your additional example of having output_dir: build/bin and a tool creating files in build/bin/bin it seems to me you want to exclude build/bin/bin and not build/bin right?

[agross]

Finally, given your additional example of having output_dir: build/bin and a tool creating files in build/bin/bin it seems to me you want to exclude build/bin/bin and not build/bin right?

Right, but given that the exclude patterns are checked against each path component I cannot achieve that either.

Adding bin as a pattern essentially disables pruning for all files.

[gpakosz]

I mean prune.exclude: ['build/bin/bin']

[agross]

Now I would acknowledge there's an ambiguity in the Nanoc's wording. Should prune.exclude: ['output'] exclude output or output/output?

From my point of view it should only exclude output/output.

Consider this:

1. output_directory: foo and prune.exclude: ['bar']
2. You use pruning and like it
3. Later, you decide you want your output in bar, hence you set output_directory: bar
4. Pruning gets disabled without anybody telling you

[agross]

I mean prune.exclude: ['build/bin/bin']

Not possible, as each path component is checked against a pattern. What you suggested never is a component. For build/bin/bin/foo.dll these components will be checked against exclude patterns: build, bin, bin, foo.dll.

I remember a discussion we had earlier that I would prefer pruning to use globs. It allows for fine-grained control, is rather standard, but I do still believe these globs should be checked against paths below output_dir, whatever that value is at the moment.

[agross]

#883

[denisdefreyne]

Having globs would certainly still be useful, though it wouldn’t fix this particular bug.

Fix is coming up.

[agross]

Thank you!

[agross]

I'm sorry but it still doesn't work. I made a little repro: https://github.com/agross/nanoc-pruning

Run run which creates a file untracked by nanoc in output_dir which is not pruned during compilation.

[denisdefreyne]

Reopening!

[denisdefreyne]

@agross Can you verify that #1324 works for you?

[agross]

Looking good!

[denisdefreyne]

Nice! That PR needs better test coverage, so I’ll spend a bit more time to finish it up and then release 4.9.2, hopefully this weekend.