Accept broken symlinks as path arguments of dandi validate

[CodyCBakerPhD]

Another pre-PR splintered out of #1933 since it does actually affect the CLI today (granted, IDK why someone WOULD use the CLI in that way today on a datalad dataset, but they COULD)

dandi validate already handles annexed files whose content has not been fetched (the broken symlinks of a DataLad dataset) according to --missing-file-content: error, skip, or only-non-data. Those policies were applied only when such a file was reached through its directory, though: given directly on the command line, the file never got past argument parsing, because click.Path(exists=True) follows the link and rejected it with "does not exist". Check the path with lexists() instead, so that

dandi validate --missing-file-content=skip sub-01/sub-01.nwb

works the same as validating sub-01/, while paths that truly do not exist are still rejected with the same error.

Claude-Session: https://claude.ai/code/session_01HFhMPs6vF5zyeHDvPTvRcA

[codecov]

Codecov Report

❌ Patch coverage is 87.50000% with 7 lines in your changes missing coverage. Please review.
✅ Project coverage is 78.35%. Comparing base (817818b) to head (bc533a6).

Files with missing lines	Patch %	Lines
dandi/cli/base.py	61.53%	5 Missing ⚠️
dandi/cli/tests/test_base.py	92.85%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1937      +/-   ##
==========================================
+ Coverage   78.32%   78.35%   +0.03%     
==========================================
  Files          92       92              
  Lines       14055    14109      +54     
==========================================
+ Hits        11008    11055      +47     
- Misses       3047     3054       +7     

Flag	Coverage Δ
unittests	78.35% <87.50%> (+0.03%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 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.

[yarikoptic-gitmate]

Subclassing click.Path and checking with os.path.lexists() first is the right approach. click.Path.convert always follows links through os.stat() and has no option to use lstat. A callback= would be harder to reuse, and dropping exists=True would turn a clean usage error for a mistyped path into an unclear failure later on. The test covers the cases that matter.

A few things to tighten before merging (details inline):

• Stop exists=True and resolve_path=True from being passed back in, since either one undoes the fix.
• Handle allow_dash, and build the error message the same way click does (format_filename, _()).
• The deprecated validate-bids still uses click.Path(exists=True). Either switch it or say that leaving it is intentional.

---

Generated by Claude Code

[CodyCBakerPhD]

codecov/project (78.27%, -0.04%) is red only because there is no codecov config, so any drop fails. The 8 patch lines it lists as uncovered are:

• in dandi/cli/base.py, the new imports and the class/def lines of PathOrBrokenSymlink, which run at import time before coverage starts (the bodies of __init__ and convert are covered by test_path_or_broken_symlink*);
• in test_base.py, the pytest.skip fallback for platforms without symlinks.

Leaving as is.

---

Generated by Claude Code

[CodyCBakerPhD]

codecov/project (78.27%, -0.04%) is red only because there is no codecov config, so any drop fails. The 8 patch lines it lists as uncovered are:

Which I will say, Claude also offered to fix separately from all this validation effort, if you are interested? Might increase coverage %

[yarikoptic]

I hate to do it, but I started to wonder if we could choose a better name.
Also I think docstring should not talk in such gory detail about why -- move into commit msg

[yarikoptic]

stroke me only now that the name is a bit misleading/could be improved given that it is forever to be part of our public interface ...
I kinda see the rationale for it but it kinda begs a question "What about a not broken symlink?". Overall any symlink is also a Path strictly speaking, so it is really the absense of lexists option in addition to exists (also at pathlib level somewhat lacking such parallel to os.path. interfaces) hindering the core situation here...

so I wonder what is a potentially better name here?

overall it could even be just an overloaded Path but accepting the lexists flag in addition to exists (exists=False, lexists=True even makes sense if someone is making command like 'fix-broken-links'). Type checking etc would help to ensure using our overloaded one (dandi.cli.Path) where we want to allow providing broken links as well. WDYT?

[CodyCBakerPhD]

Agreed, done in 56bb1a5: dandi.cli.base.Path, a click.Path with a lexists flag next to exists; validate now uses Path(lexists=True, dir_okay=True). exists=True keeps click's meaning (and wins if both are given); lexists=True only refuses resolve_path=True, which would swap a link for its target. Tests cover lexists, exists, and both. I kept it in dandi/cli/base.py next to EnumChoice/ChoiceList rather than re-exporting from dandi.cli; happy to add that if you prefer.

---

Generated by Claude Code

[CodyCBakerPhD]

Hmm sorry about that - Codybot should have proposed some naming alternative before assuming

I'd propose either UnresolvedPath or LinkAwarePath: your choice or propose your own

[CodyCBakerPhD]

@yarikoptic (going with working name of LinkAwarePath for now since it assumes the least about the target)

[yarikoptic]

Great, thanks! Let's just squash for merge

[CodyCBakerPhD]

Let's just squash for merge

Need the gitmate to grant their approval as well since official changes were requested

also

will have the same effect on the commit history without losing all the nice history of this PR, please use that instead

[github-actions]

🚀 PR was released in 0.81.0 🚀