feat: Add "force_implicit_dirs" to dir/tree.

[volkadav]

This allows users to specify a list of directories to treat akin to the list of system directories, i.e. things that the constructed package should not take ownership of in e.g. the rpm database. (I'm 100% open to a better name for that, btw!)

Example use case:

1. user belongs to FooCorp
2. FooCorp has many packages to install under /opt/foocorp
   e.g. /opt/foocorp/service[A,B,C...]
3. Using tree w/ forced implicit dirs keeps generated package(s)
   from claiming ownership of /opt/foocorp so that the
   service[A,B,C...] packages can be bundled and installed
   separately without path conflicts.

The particular scenario I ran into this for was with RHEL8+ being much pickier about each dir being owned by one and only one package. (We pair this with having one "foocorp-dirs" package that just owns the shared dir creation, that's used as a pre-req for the other packages.) We could have done something like explicitly list every sub-dir needed or every file for all the service packages, but using tree was much more concise and flexible with respect to package contents changing over time, and in effect this is giving the user the ability to specify dirs to treat like the hard-coded list of system directories already checked towards the same end.

I have run task ci and the normal tests pass. There seems to be some issue with opkg going away on the latest OpenWRT images that keeps the ipk acceptance test from working though:

#9 ERROR: process "/bin/sh -c opkg install /tmp/foo.ipk" did not complete successfully: exit code: 127
                                ------
                                 > [min 1/1] RUN opkg install /tmp/foo.ipk:
                                0.287 /bin/sh: opkg: not found
                                ------
                                ipk.dockerfile:12
                                --------------------
                                  10 |     # ---- minimal test ----
                                  11 |     FROM test_base AS min
                                  12 | >>> RUN opkg install /tmp/foo.ipk
                                  13 |
                                  14 |
                                --------------------

Sadly I don't know enough about OpenWRT/opkg/Docker to guess what the right solution would be for that, so I haven't included any changes for it.

Thank you for a cool tool! :)

[erikgeiser]

I like this proposal. However, we would need a test as well. Don't worry, I'll update the PR myself.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 73.76%. Comparing base (432fb67) to head (5419c42).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #948      +/-   ##
==========================================
+ Coverage   73.64%   73.76%   +0.12%     
==========================================
  Files          23       23              
  Lines        2758     2771      +13     
==========================================
+ Hits         2031     2044      +13     
  Misses        507      507              
  Partials      220      220              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[erikgeiser]

Before I merge this, I would like to ask @caarlos0 or @djgilcrease what they think about the name disown_subtree for this settings and whether they also think that this is a reasonable feature.

I also added a fix to an issue which is currently present in main because I encountered it when writing tests. If I add a tree to /, nfpm will try to add the explicit directory . which gets normalized to // to the package. I'm pretty sure we don't want that and it was currently not covered by tests because all tree tests have non-root directory as destination. The fix is simple: https://github.com/goreleaser/nfpm/pull/948/changes#diff-326e77b3dbf4c7c47df40989fb759cc07692519b73e1698190193f266678c8f7R484

[volkadav]

Thank you, Erik! I had sort of forgotten about this (and have moved on from the company that needed it), but it's good to see it getting some traction. The disown_subtree name makes total sense. If there's anything further I can do to help, please let me know!

(edit to add: I looked at the failing checks. The semgrep failure is complaining about something in msix/msix.go which afaict is unrelated to this PR and the docs build appears to relate to populating author info in an rss feed; sadly I don't know enough about the project to know what the right fixes might be for either, but if needed I can make some educated guesses.)

[erikgeiser]

Sorry for not being able to review this PR when you actually needed it.

Yes the CI issues are not related to this PR, so it could be merged as-is.