abrignoni Validating a path glob
A pattern that matches nothing fails silently. The artifact simply reports no data, which is indistinguishable from the app not being present. Never commit a pattern you have not run.
What you are testing against
The matcher is Python fnmatch, so reproduce it exactly rather than approximating with
glob or a regex:
*and**both span/. They are interchangeable.- Matching goes through
os.path.normcase, so it is case-sensitive on macOS and Linux and case-insensitive on Windows. Test the case-sensitive behaviour; it is the stricter one. - The directory, zip and tar seekers prepend a synthetic
root/. iLEAPP's iOS-backup seeker does not, and does not normcase either.
Method 1: against recorded path listings (fastest)
iLEAPP carries real file listings from published research images at
admin/data/filepath-lists/*.csv.zip, covering several iOS generations.
Unzip to a scratch directory, then run each candidate pattern over the path column with
fnmatch.fnmatchcase and count hits per generation. Run the old and new patterns both,
and compare. A change that reduces the count on an older image is a regression, not a
tightening, unless you meant it.
Method 2: against a real extraction (authoritative)
Instantiate the matching seeker from scripts/search_files.py directly so the semantics are
exact by construction, then call search() with the pattern. This is the only way to catch
a mismatch between what you think the seeker does and what it does.
To confirm the artifact then produces the rows you expect, call the artifact function's
.__wrapped__, exposed by @artifact_processor's @wraps, with a mock context providing
get_files_found() and get_relative_path(). len(data_list) is the count to record in
sample_data.
Patterns that need extra care
Case variants. Use one bracket class, */[Bb]iome/*, never a tuple of two patterns. A
tuple double-counts on Windows, where normcase folds both to the same string and the
entry point extends files_found once per pattern with no dedup.
Vendor moves. When a file changes location between OS versions, keep the old pattern alongside the new one. Examiners run these tools against extractions going back years.
Over-generic components. A pattern whose filename component is bare *, ** or *.*
matches every file at that level and is rejected. Anchor on something real.
Hygiene
Print counts and value shapes from test images, never actual values, and delete anything you extracted when you are done.