External/contextzip
1
0
Fork 0
mirror of https://github.com/akadeepesh/contextzip.git synced 2026-10-20 08:36:20 +02:00
Code Issues Packages Projects Releases Wiki Activity

fix(applier): stop skipping wrapper-strip when project has a same-named folder

Browse Source 
apply-zip's wrapper-folder detection refused to strip a zip's outer
folder whenever the project already had a real directory of that same
name — which backfires on the single most common shape of AI-returned
zip: the whole project wrapped in a folder named after the project (or
its own package dir), e.g. 'project/' or 'contextzip/contextzip/...'.
Since the project's real package dir almost always shares that name,
the guard treated the wrapper as legitimate structure and dumped a
nested copy at the project root instead of updating the real files.

Now, whenever a manifest is available, the decision is made by
comparing manifest match-rate with vs. without stripping (ground
truth about what paths this project actually has) instead of
inferring intent from what merely exists on disk. Falls back to the
old on-disk heuristic only when no manifest exists to check against.

Fixes the case where apply-zip silently mis-applies a returned zip.
This commit is contained in:
akadeepesh
2026-09-20 11:56:37 +05:30
parent c84ba3367b
commit 6ac6b85070
1 changed files with 21 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
contextzip/applier.py
+21 -3
View File
@@ -446,8 +446,24 @@ def build_plan(
strip_prefix: str | None = None
wrapper_note: str | None = None
candidate = _detect_common_wrapper(raw_names)
if candidate and not (project_dir / candidate).is_dir():
if candidate:
if manifest_files:
# Decide using the manifest — actual ground truth about what
# paths this project has — rather than whether a directory of
# that name merely happens to exist on disk. That on-disk check
# used to gate this whole branch, which backfired on the most
# common real case: an AI tool wraps the whole project in a
# folder named after the project itself (e.g. "contextzip/"),
# and the project's own top-level package/dir happens to share
# that exact name — so the wrapper looked "legitimate" and was
# never stripped, dumping a nested copy of the project instead
# of updating the real files. Comparing manifest match-rates
# with vs without stripping gets both cases right: a genuine
# wrapper matches far better once stripped, while a zip whose
# top-level folder is an intentional part of the project (e.g.
# updates confined to an existing "docs/" folder) matches
# better *unstripped*, since those paths are already in the
# manifest as-is.
unstripped_rate = _match_rate(raw_names, manifest_files)
stripped_names = [n[len(candidate) + 1:] for n in raw_names]
stripped_rate = _match_rate(stripped_names, manifest_files)
@@ -462,10 +478,12 @@ def build_plan(
f"project manifest after stripping vs {unstripped_rate:.0%} "
"before."
)
else:
elif not (project_dir / candidate).is_dir():
# No manifest to confirm against, so this is a softer call —
# still strip (matches how `tar` and GitHub's own zip downloads
# behave), but say so plainly since it's not manifest-verified.
# behave) as long as the project doesn't already have a real
# folder of that name, but say so plainly since it's not
# manifest-verified.
strip_prefix = candidate
wrapper_note = (
f"Removed wrapping folder '{candidate}/' present in every zip "