Merged
Conversation
|
Important Review skippedThis PR was authored by the user configured for CodeRabbit reviews. By default, CodeRabbit skips reviewing PRs authored by this user. It's recommended to use a dedicated user account to post CodeRabbit review feedback. To trigger a single review, invoke the You can disable this status message by setting the Use the checkbox below for a quick retry:
WalkthroughReplaced numerous procedural patch utilities with a typed Patch API ( Changes
Sequence Diagram(s)sequenceDiagram
participant Cmd as format_patch command
participant FS as Filesystem
participant Patch as dfetch.vcs.patch.Patch
participant SubP as SubProject / SuperProject
participant Output as Patch file writer
rect rgba(100,149,237,0.5)
Cmd->>FS: iterate patch_file paths
FS-->>Cmd: return patch_file
end
rect rgba(60,179,113,0.5)
Cmd->>Patch: Patch.from_file(patch_file)
Patch-->>Cmd: Patch instance
Cmd->>Patch: .convert_type(target: PatchType)
Cmd->>Patch: .add_prefix(prefix)
Cmd->>Patch: .dump()
Patch-->>Output: write patch text
end
rect rgba(255,165,0,0.5)
SubP->>Patch: Patch.for_new_files(paths, PatchType)
Patch-->>SubP: Patch instance
SubP->>Patch: .combine() / .reverse()
Patch-->>Output: final dump()
end
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45 minutes Possibly related PRs
Suggested labels
Suggested reviewers
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 inconclusive)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 4
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
dfetch/vcs/patch.py (1)
402-424:⚠️ Potential issue | 🔴 Critical
convert_patch_tois broken:PatchTypeenum members never equal rawpatch_ngconstants.
required_typeis aPatchTypeenum, but lines 406, 409, and 419 compare it againstpatch.type(a rawpatch_ngconstant) andpatch_ng.GIT/patch_ng.SVNdirectly. PythonEnum.__eq__won't match these, so:
- Line 406: early-return is never taken (conversion always runs even when unnecessary).
- Lines 409/419: neither GIT nor SVN branches execute — the function falls through and sets
patch.type = required_type(aPatchTypeenum), whichpatch_ngwon't understand downstream.Use
.valueconsistently, or compare againstPatchTypemembers.Proposed fix
def convert_patch_to( patch: patch_ng.PatchSet, required_type: PatchType ) -> patch_ng.PatchSet: """Convert the patch to the required type.""" - if required_type == patch.type: + if required_type.value == patch.type: return patch - if required_type == patch_ng.GIT: + if required_type == PatchType.GIT: for file in patch.items: file.header = [ b"diff --git " + _rewrite_path(b"a/", file.source) + b" " + _rewrite_path(b"b/", file.target) + b"\n" ] - file.type = required_type - elif required_type == patch_ng.SVN: + file.type = required_type.value + elif required_type == PatchType.SVN: for file in patch.items: file.header = [b"Index: " + file.target + b"\n", b"=" * 67 + b"\n"] - file.type = required_type - patch.type = required_type + file.type = required_type.value + patch.type = required_type.value return patch
🤖 Fix all issues with AI agents
In `@dfetch/project/svnsuperproject.py`:
- Around line 125-132: The code mixes Patch objects and raw strings causing
type/runtime errors: replace uses of Patch.from_file(filtered) (where filtered
is a diff string from create_diff) with Patch.from_bytes(filtered.encode() or
Patch.from_bytes(filtered)) to construct a Patch, change the handling of
Patch.for_new_file (which currently returns str) to return a Patch object (or
wrap its result into Patch.for_new_file_from_str/ Patch.from_bytes) so all
entries in patches are Patch instances, and finally use Patch.combine(patches)
(or call the class/instance method that combines Patch objects) instead of
combine_patches which expects Sequence[bytes]; update or call the corrected
Patch.combine API so combine is invoked on Patch objects rather than passing
list[Patch] to combine_patches.
In `@dfetch/vcs/patch.py`:
- Around line 145-156: combine() currently only appends items from other and
omits self; update Patch.combine so it includes both self._patchset.items and
other items (preserving intended order), e.g. initialize combined_patchset.items
with self._patchset.items (accessing self._patchset.items) and then extend/+=
with each patch._patchset.items from other (still handling the case where other
is a single Patch by wrapping it), then return Patch(combined_patchset); keep
existing use of patch_ng.PatchSet and the Patch constructor and respect the
protected-access pattern used elsewhere.
- Around line 127-133: The filter method creates a new patch_ng.PatchSet but
doesn't preserve the original patchset's type, so is_git/is_svn and dump_patch
can be wrong; update the filter method (function name: filter, class: Patch,
variable: filtered and source: self._patchset) to initialize or assign the new
PatchSet's type from self._patchset (e.g., construct filtered with the same
.type or set filtered.type = self._patchset.type) before appending items so the
filtered Patch preserves the original type.
In `@dfetch/vcs/svn.py`:
- Line 336: Guard against empty patch_text before calling Patch.from_bytes to
avoid RuntimeError: if patch_text is empty (falsy) return the empty result
(e.g., "") rather than calling Patch.from_bytes; change the code around
Patch.from_bytes(patch_text).filter(ignore).dump() to first check patch_text and
short-circuit, otherwise proceed to construct Patch and call filter and dump.
Ensure you reference the patch_text variable and the Patch.from_bytes / filter /
dump chain when making the change so behavior matches previous empty-output
behavior.
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (5)
dfetch/vcs/git.py (1)
381-413: Note:GitLocalRepo.create_diffstill returnsstrwhileSvnRepo.create_diffnow returnsPatch.This asymmetry may be intentional (git's diff is consumed differently), but if further unification is planned, it's worth aligning these return types in a follow-up.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@dfetch/vcs/git.py` around lines 381 - 413, The GitLocalRepo.create_diff currently returns a raw str while SvnRepo.create_diff returns a Patch object, so make them consistent: update GitLocalRepo.create_diff (the create_diff method in dfetch/vcs/git.py) to return a Patch (or the same type used by SvnRepo.create_diff) by converting the git command output into a Patch instance before returning (e.g., parse or construct a Patch from result.stdout.decode()), and update the method signature/return annotation accordingly so callers expect the unified type.dfetch/vcs/patch.py (3)
185-229:add_prefixmutatesselfdespite docstring claiming it returns a new Patch.The docstring says "Return new patch with all file paths rewritten" but this method mutates
self._patchset.itemsin-place (modifyingfile.source,file.target,file.header) and returnsself. This is inconsistent withfilter(),combine(), andreverse(), which all return genuinely newPatchinstances. A caller doing:original = Patch.from_file(...) prefixed = original.add_prefix("src") # original is also mutated hereCurrent callers aren't bitten because they reassign (
patch = patch.add_prefix(...)) or chain on a fresh object, but it's a latent footgun. At minimum, fix the docstring to say "Rewrite file paths in-place and return self for chaining." The same applies toconvert_type.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@dfetch/vcs/patch.py` around lines 185 - 229, The add_prefix method currently mutates self._patchset.items in-place (modifying file.source, file.target and file.header) but its docstring claims it "Return new patch"; update the add_prefix docstring to accurately state it rewrites file paths in-place and returns self for chaining (and do the same for convert_type), referencing the add_prefix method and the in-place mutations of file.source, file.target and file.header on self._patchset.items so callers know it does not produce a new Patch instance.
253-265:combineshares item object references — later mutation of source/combined patch will affect both.
combined_patchset.items += self._patchset.itemscopies list references, not the items themselves. Sinceadd_prefixandconvert_typemutate items in-place, calling.add_prefix()on either the original or combinedPatchwould corrupt the other. This is safe only if callers never reuse aPatchafter combining. Worth documenting or shallow-copying items.Proposed defensive copy
combined_patchset = patch_ng.PatchSet() - combined_patchset.items += self._patchset.items + combined_patchset.items += list(self._patchset.items) for patch in other: - combined_patchset.items += ( - patch._patchset.items # pylint: disable=protected-access + combined_patchset.items += list( + patch._patchset.items # pylint: disable=protected-access )Note: this only gives a new list — the individual patch item objects are still shared. A true deep copy would require copying each item, which may not be worth the complexity if you document the aliasing contract.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@dfetch/vcs/patch.py` around lines 253 - 265, The combine method is currently concatenating lists by reference (combined_patchset.items += self._patchset.items and similar), causing shared list objects between the original and returned Patch; change combine in class Patch to create new lists of items (e.g., replace in-place concatenation with list(self._patchset.items) and extend with list(patch._patchset.items) for each incoming patch) so the combined Patch has its own items list and later calls like add_prefix or convert_type on one Patch won't mutate the other's list structure; alternatively explicitly document the aliasing contract if you prefer to keep references, but prefer the defensive shallow-copy approach for combined_patchset before returning.
119-122:Patch.empty()creates aPatchthat will crash ondump()but silently succeed, whilereverse(),from_bytes()etc. will raise.
Patch.empty()creates aPatch(patch_ng.PatchSet())with no items. Calling.dump()returns""(fine),.filesreturns[](fine), and.apply()would callself._patchset.apply(...)which may returnFalseand raiseRuntimeError. Meanwhile.reverse()would calldump()→""→reverse_patch(b"")→""→from_bytes(b"")→RuntimeError.Consider documenting that
Patch.empty()is only suitable for sentinel/no-op uses (e.g., as returned bysvn.create_difffor empty diffs) and should not be reversed or applied.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@dfetch/vcs/patch.py` around lines 119 - 122, Patch.empty() currently constructs Patch(patch_ng.PatchSet()) which behaves as a sentinel/no-op for dump() and files but will raise or fail when used with reverse(), from_bytes(), or apply() (via _patchset.apply); update the empty() implementation docstring to clearly state it is only a sentinel/no-op (e.g., suitable for svn.create_diff empty diffs) and explicitly warn that reverse(), from_bytes(), and apply() are unsupported and may raise RuntimeError, so callers must not attempt to reverse or apply Patch.empty() instances.tests/test_patch.py (1)
304-310: Consider testingPatch.from_fileon an invalid/empty file to confirmRuntimeErroris raised.The test suite covers the happy path well. A small addition could verify that
Patch.from_fileraisesRuntimeErrorfor invalid input, which is the expected behavior relied on bysubproject.py(immediate failure on bad patches).Do you want me to generate a test case for this?
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@tests/test_patch.py` around lines 304 - 310, Add a new unit test that verifies Patch.from_file raises RuntimeError for invalid or empty patch files: create an empty/invalid temporary file and call Patch.from_file(path) inside pytest.raises(RuntimeError) (or catch and assert the exception) to ensure the function errors deterministically; reference Patch.from_file and ensure the test is colocated in tests/test_patch.py alongside the existing reverse-patch test so it fails fast if invalid input is provided.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@dfetch/project/svnsuperproject.py`:
- Line 120: The code currently uses truthiness of the Patch instance
("filtered") which is always true for dataclass Patch, causing empty patches to
be included and later crash when combined.reverse() calls Patch.from_bytes(b"");
change the check when building patches to detect and exclude empty patches (use
Patch.empty() or equivalent) — e.g. only append filtered if filtered is not None
and not filtered.empty() so that create_diff-produced empty Patch objects are
omitted before any reverse() call; update the logic around the variable filtered
and the patches list to prevent passing an empty Patch into combined.reverse().
---
Nitpick comments:
In `@dfetch/vcs/git.py`:
- Around line 381-413: The GitLocalRepo.create_diff currently returns a raw str
while SvnRepo.create_diff returns a Patch object, so make them consistent:
update GitLocalRepo.create_diff (the create_diff method in dfetch/vcs/git.py) to
return a Patch (or the same type used by SvnRepo.create_diff) by converting the
git command output into a Patch instance before returning (e.g., parse or
construct a Patch from result.stdout.decode()), and update the method
signature/return annotation accordingly so callers expect the unified type.
In `@dfetch/vcs/patch.py`:
- Around line 185-229: The add_prefix method currently mutates
self._patchset.items in-place (modifying file.source, file.target and
file.header) but its docstring claims it "Return new patch"; update the
add_prefix docstring to accurately state it rewrites file paths in-place and
returns self for chaining (and do the same for convert_type), referencing the
add_prefix method and the in-place mutations of file.source, file.target and
file.header on self._patchset.items so callers know it does not produce a new
Patch instance.
- Around line 253-265: The combine method is currently concatenating lists by
reference (combined_patchset.items += self._patchset.items and similar), causing
shared list objects between the original and returned Patch; change combine in
class Patch to create new lists of items (e.g., replace in-place concatenation
with list(self._patchset.items) and extend with list(patch._patchset.items) for
each incoming patch) so the combined Patch has its own items list and later
calls like add_prefix or convert_type on one Patch won't mutate the other's list
structure; alternatively explicitly document the aliasing contract if you prefer
to keep references, but prefer the defensive shallow-copy approach for
combined_patchset before returning.
- Around line 119-122: Patch.empty() currently constructs
Patch(patch_ng.PatchSet()) which behaves as a sentinel/no-op for dump() and
files but will raise or fail when used with reverse(), from_bytes(), or apply()
(via _patchset.apply); update the empty() implementation docstring to clearly
state it is only a sentinel/no-op (e.g., suitable for svn.create_diff empty
diffs) and explicitly warn that reverse(), from_bytes(), and apply() are
unsupported and may raise RuntimeError, so callers must not attempt to reverse
or apply Patch.empty() instances.
In `@tests/test_patch.py`:
- Around line 304-310: Add a new unit test that verifies Patch.from_file raises
RuntimeError for invalid or empty patch files: create an empty/invalid temporary
file and call Patch.from_file(path) inside pytest.raises(RuntimeError) (or catch
and assert the exception) to ensure the function errors deterministically;
reference Patch.from_file and ensure the test is colocated in
tests/test_patch.py alongside the existing reverse-patch test so it fails fast
if invalid input is provided.
There was a problem hiding this comment.
Actionable comments posted: 3
🧹 Nitpick comments (1)
dfetch/vcs/patch.py (1)
189-233: Inconsistent mutation semantics acrossPatchmethods.
add_prefix()andconvert_type()mutateselfin-place and returnself, whilefilter(),reverse(), andcombine()return newPatchinstances. This mixed API can surprise callers who expect uniform copy-on-write or uniform in-place behavior.Current callers work correctly because they reassign (
patch = patch.add_prefix(...)), but this inconsistency is a foot-gun for future use.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@dfetch/vcs/patch.py` around lines 189 - 233, The add_prefix method currently mutates self (via self._patchset and file.source/file.target and file.header) then returns self, creating an API mismatch with filter()/reverse()/combine() which return new Patch instances; change add_prefix to be non-mutating by creating a copy of the Patch (e.g., clone the Patch and/or deep-copy self._patchset and its File entries), apply the path rewrites to the copy’s file.source/file.target/file.header using the existing _rewrite_path and regex logic, and return the new Patch instance; also update convert_type to follow the same non-mutating pattern so add_prefix and convert_type have the same copy-on-write semantics as filter/reverse/combine.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@dfetch/vcs/patch.py`:
- Around line 115-117: The plain-patch branch is joining diff lines with
"\n".join(diff) which inserts extra blank lines because difflib.unified_diff
already includes trailing "\n"; change the plain fallback to use "".join(diff)
(same approach as the GIT/SVN branches) so the output isn't double-spaced—update
the return in the function handling patch_type (compare with the existing
PatchType.SVN branch) to use "".join(diff) or otherwise avoid adding an extra
delimiter.
- Around line 257-269: The combine() method creates a new patch_ng.PatchSet but
does not copy the original patchset type, causing is_git/is_svn and dump() to be
incorrect; update combine (in the Patch.combine method) to set
combined_patchset.type = self._patchset.type (same approach used in filter())
after creating combined_patchset and before returning the new Patch, ensuring
that when you merge other Patch or Sequence[Patch] instances the resulting Patch
retains the original PatchSet.type.
- Around line 124-126: The is_empty method is inverted: it returns True when
items exist, causing non-empty patches to be treated as empty; update
dfetch.vcs.patch.Patch.is_empty to return the negation of the current expression
(i.e., return not bool(self._patchset.items) or equivalently check
len(self._patchset.items) == 0) so callers like the usage in svnsuperproject.py
correctly treat only truly empty Patch instances as empty.
---
Duplicate comments:
In `@dfetch/project/svnsuperproject.py`:
- Around line 118-125: The condition that initializes patches uses
filtered.is_empty() inverted, causing non-empty diffs to be dropped; change the
logic so that when filtered is NOT empty the list starts with filtered (i.e.,
include filtered when filtered.is_empty() is false) and then append new-file
patches from Patch.for_new_file and Patch.from_string as before; update the
initialization around the symbols filtered, Patch, Patch.for_new_file,
Patch.from_string, and repo.untracked_files (inside in_directory(path)) so the
code includes non-empty filtered patches rather than excluding them.
---
Nitpick comments:
In `@dfetch/vcs/patch.py`:
- Around line 189-233: The add_prefix method currently mutates self (via
self._patchset and file.source/file.target and file.header) then returns self,
creating an API mismatch with filter()/reverse()/combine() which return new
Patch instances; change add_prefix to be non-mutating by creating a copy of the
Patch (e.g., clone the Patch and/or deep-copy self._patchset and its File
entries), apply the path rewrites to the copy’s
file.source/file.target/file.header using the existing _rewrite_path and regex
logic, and return the new Patch instance; also update convert_type to follow the
same non-mutating pattern so add_prefix and convert_type have the same
copy-on-write semantics as filter/reverse/combine.
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (3)
dfetch/vcs/patch.py (3)
100-121:for_new_filereturnsstrwhile every other factory method returnsPatch.Every static constructor (
from_file,from_bytes,from_string,empty) returns aPatch.for_new_filereturns a rawstr, forcing callers to doPatch.from_string(Patch.for_new_file(...))(as insvnsuperproject.py) whilegit.pyconsumes the string directly. Returning aPatchdirectly would unify the API and remove the intermediate wrapping call.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@dfetch/vcs/patch.py` around lines 100 - 121, The factory method for_new_file currently returns a raw string while other constructors return a Patch; change for_new_file to return a Patch by building the same diff string currently returned and wrapping it with Patch.from_string(...) before returning. Locate Patch.for_new_file and use the same logic that constructs diff (using _unified_diff_new_file(path), PatchType checks, _git_mode(path), and _git_blob_sha1(path)) but return Patch.from_string(the_final_string) instead of the raw string so callers can receive a Patch directly.
179-185:dump_headerdoes not referenceself— should be@staticmethod.- def dump_header(self, patch_info: PatchInfo, patch_type: PatchType) -> str: + `@staticmethod` + def dump_header(patch_info: PatchInfo, patch_type: PatchType) -> str:🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@dfetch/vcs/patch.py` around lines 179 - 185, The method dump_header is defined as an instance method but does not use self; mark it as a static method by adding the `@staticmethod` decorator and remove the unused self parameter so the signature becomes dump_header(patch_info: PatchInfo, patch_type: PatchType) -> str; keep the existing logic that calls patch_info.to_git_header() and patch_info.to_svn_header() and update any internal calls if they currently invoke it as an instance method to call it as a static method when necessary.
187-193:reverse()mutatesself;combine()returns a new object — pick one convention.
reverse,filter,add_prefix, andconvert_typeall mutateselfin-place and returnself.combineconstructs and returns a brand-newPatch. This asymmetry is a latent footgun:a = patch b = a.filter(ignore) # b is a; both are now mutated c = a.combine([other]) # c is a NEW Patch; a is unchangedAll current call sites reassign the result, so there are no bugs today, but the inconsistency will surprise future callers. Consider making all transform methods either all return-new (functional style) or all mutate-and-return-self (fluent style).
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@dfetch/vcs/patch.py` around lines 187 - 193, The patch transformation methods are inconsistent: combine() returns a new Patch while reverse(), filter(), add_prefix(), and convert_type() currently mutate self; standardize on a non-mutating functional style by having reverse(), filter(), add_prefix(), and convert_type() return a new Patch instead of mutating self. For reverse(), stop assigning to self._patchset — create a new Patch from the transformed bytes (use Patch.from_bytes or the existing from_bytes factory and the dump()/reverse_patch call) and return that new instance; apply the same pattern in filter(), add_prefix(), and convert_type() so they construct and return new Patch objects rather than modifying self, preserving combine()’s behavior.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@dfetch/vcs/patch.py`:
- Around line 251-271: convert_type currently updates self._patchset.type even
when required is not GIT or SVN, leaving per-file file.header and file.type
inconsistent (e.g., when required == PatchType.PLAIN); update convert_type to
either reject/raise on unsupported required types or treat PLAIN as a no-op: if
required.value is neither patch_ng.GIT nor patch_ng.SVN, do not change
self._patchset.type and return self (or raise a ValueError), otherwise proceed
with the existing per-file header/type updates; reference function convert_type,
enum PatchType (PLAIN/GIT/SVN), the patchset self._patchset and per-file
attributes file.header and file.type so the change keeps dump() and
is_git/is_svn consistent.
---
Nitpick comments:
In `@dfetch/vcs/patch.py`:
- Around line 100-121: The factory method for_new_file currently returns a raw
string while other constructors return a Patch; change for_new_file to return a
Patch by building the same diff string currently returned and wrapping it with
Patch.from_string(...) before returning. Locate Patch.for_new_file and use the
same logic that constructs diff (using _unified_diff_new_file(path), PatchType
checks, _git_mode(path), and _git_blob_sha1(path)) but return
Patch.from_string(the_final_string) instead of the raw string so callers can
receive a Patch directly.
- Around line 179-185: The method dump_header is defined as an instance method
but does not use self; mark it as a static method by adding the `@staticmethod`
decorator and remove the unused self parameter so the signature becomes
dump_header(patch_info: PatchInfo, patch_type: PatchType) -> str; keep the
existing logic that calls patch_info.to_git_header() and
patch_info.to_svn_header() and update any internal calls if they currently
invoke it as an instance method to call it as a static method when necessary.
- Around line 187-193: The patch transformation methods are inconsistent:
combine() returns a new Patch while reverse(), filter(), add_prefix(), and
convert_type() currently mutate self; standardize on a non-mutating functional
style by having reverse(), filter(), add_prefix(), and convert_type() return a
new Patch instead of mutating self. For reverse(), stop assigning to
self._patchset — create a new Patch from the transformed bytes (use
Patch.from_bytes or the existing from_bytes factory and the dump()/reverse_patch
call) and return that new instance; apply the same pattern in filter(),
add_prefix(), and convert_type() so they construct and return new Patch objects
rather than modifying self, preserving combine()’s behavior.
There was a problem hiding this comment.
Actionable comments posted: 2
🧹 Nitpick comments (5)
dfetch/commands/format_patch.py (1)
129-131: Specifyencoding="utf-8"inwrite_text().Without an explicit encoding, the default is platform-dependent (e.g.,
cp1252on Windows), which may corrupt or error on patches containing non-ASCII content.♻️ Proposed fix
- output_patch_file.write_text( - patch.dump_header(patch_info) + patch.dump() - ) + output_patch_file.write_text( + patch.dump_header(patch_info) + patch.dump(), encoding="utf-8" + )🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@dfetch/commands/format_patch.py` around lines 129 - 131, The call to output_patch_file.write_text(...) in format_patch.py writes patch content without specifying encoding, which is platform-dependent; update the write to pass encoding="utf-8" so the output of patch.dump_header(patch_info) + patch.dump() is written using UTF-8 encoding (locate the write_text call near dump_header/dump usage and add the encoding argument).dfetch/vcs/patch.py (3)
228-231: Regexes recompiled on everyadd_prefixcall — promote to module-level constants.
diff_gitandsvn_indexare invariant; compiling them inside the method on each call is wasteful.♻️ Proposed fix
+_RE_DIFF_GIT = re.compile( + r"^diff --git (?:(?P<a>a/))?(?P<old>.+) (?:(?P<b>b/))?(?P<new>.+?)[\r\n]*$" +) +_RE_SVN_INDEX = re.compile(rb"^Index: (.+)$") + class PatchType(Enum): ... # Inside add_prefix, replace the two re.compile calls: - diff_git = re.compile( - r"^diff --git (?:(?P<a>a/))?(?P<old>.+) (?:(?P<b>b/))?(?P<new>.+?)[\r\n]*$" - ) - svn_index = re.compile(rb"^Index: (.+)$") + diff_git = _RE_DIFF_GIT + svn_index = _RE_SVN_INDEX🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@dfetch/vcs/patch.py` around lines 228 - 231, Move the invariant regex compilations out of the add_prefix function and promote them to module-level constants so they are compiled once; specifically, take the current local variables diff_git and svn_index, create module-level constants (e.g., DIFF_GIT_RE and SVN_INDEX_RE) initialized with re.compile(...) and re.compile(rb"...") respectively, and update the add_prefix implementation to reference these constants instead of recompiling on each call.
293-313:combine()silently mutates theotherpatches as a side effect.The
convert_type()call on Line 307 modifies the input patches inotherin-place (theirfile.header,file.type, and_patchset.type). A caller that holds a reference to a patch it passed tocombine()would see its type unexpectedly changed. Additionally,combined_patchset.items +=uses shallow copies, so later item-level mutations on the returned combinedPatchpropagate back to the original patches.♻️ Proposed fix — convert a copy, deep-copy items
def combine(self, other: Patch | Sequence[Patch]) -> Patch: """Combine two patches into a new patch.""" if isinstance(other, Patch): other = [other] + import copy combined_patchset = patch_ng.PatchSet() combined_patchset.type = self._patchset.type - combined_patchset.items += self._patchset.items + combined_patchset.items += copy.copy(self._patchset.items) for patch in other: if ( patch._patchset.type # pylint: disable=protected-access != self._patchset.type ): - patch.convert_type(PatchType(combined_patchset.type)) - - combined_patchset.items += ( - patch._patchset.items # pylint: disable=protected-access - ) + patch = copy.deepcopy(patch) + patch.convert_type(PatchType(combined_patchset.type)) + combined_patchset.items += copy.copy( + patch._patchset.items # pylint: disable=protected-access + ) return Patch(combined_patchset)🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@dfetch/vcs/patch.py` around lines 293 - 313, combine currently mutates input Patch objects and shares item objects; fix by operating on copies: create a deep copy of self._patchset (or at least deep-copy its items) into combined_patchset rather than using combined_patchset.items += self._patchset.items, and for each patch in other make a deep copy of patch._patchset (so convert_type is invoked on the copy, not the original), call convert_type on that copy (using PatchType(combined_patchset.type)), then extend combined_patchset.items with the copied items; finally return Patch(combined_patchset). Use the unique symbols combine, Patch, PatchType, convert_type, _patchset, and combined_patchset.items when locating and changing the code.
41-55:@dataclassexposes_patchsetas a public constructor param and generates a meaningless__eq__.Two concerns:
- Because
_patchset(intended private) is a plain dataclass field, the generated__init__exposes it as a positional parameter, makingPatch(some_patchset)part of the public API rather than an internal implementation detail.patch_ng.PatchSetalmost certainly doesn't implement__eq__, so the auto-generated__eq__degrades to identity comparison — twoPatchinstances wrapping identical content will never compare equal, which can surprise callers.Consider suppressing the generated equality logic and restricting the direct constructor:
♻️ Suggested approach
-@dataclass -class Patch: +@dataclass(eq=False) +class Patch:Or, if equality by content is meaningful, implement it explicitly. Alternatively, move to a plain class with an explicit
__init__and@classmethodfactories — given all construction goes through static factories anyway, the@dataclass__init__gives little benefit.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@dfetch/vcs/patch.py` around lines 41 - 55, Patch currently exposes the internal _patchset as a public dataclass constructor arg and gets a poor default __eq__; change the class to stop exposing _patchset and to provide meaningful equality: mark the dataclass with eq=False and make _patchset a non-init/non-compare field (e.g., field(init=False, compare=False, repr=False)), keep/implement factory methods (e.g., your existing from_string/from_file constructors) to set self._patchset, and add an explicit __eq__(self, other) that compares normalized patch content (for example by comparing a canonical text/serialized form of self._patchset and other._patchset) so equality reflects patch content rather than identity.tests/test_patch.py (1)
306-310: Preferpytest.fail()overassert Falseto preserve traceback context.
assert Falseswallows the original exception's traceback, making it harder to diagnose failures on flaky runs.♻️ Suggested change
try: Patch.from_file(patch_file).apply(root=str(tmp_path)) except Exception as e: - assert False, f"Reverse patch failed: {e}" + pytest.fail(f"Reverse patch failed: {e}")🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@tests/test_patch.py` around lines 306 - 310, Replace the "assert False" idiom in the except block with pytest.fail to preserve traceback context: inside the try/except around Patch.from_file(...).apply(root=str(tmp_path)) catch, call pytest.fail(f"Reverse patch failed: {e}") (and add "import pytest" at the top if missing) instead of assert False so failures surface with proper test reporting and context.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@dfetch/vcs/patch.py`:
- Around line 157-162: The error message in Patch.apply is unhelpful when
self.path is empty (inline patches created by from_string/from_bytes); update
the RuntimeError raised in apply to include a fallback identifier (e.g. "<inline
patch>" or use a property that returns self.path or a descriptive fallback) so
the message becomes something like 'Applying patch "<identifier>" failed';
modify the apply method to compute an identifier from self.path (or a new helper
on the Patch class) and use that identifier in the RuntimeError to make failures
for inline patches clear.
- Around line 200-210: The error message in reverse() is unreachable because
from_bytes raises on empty input; change reverse() (method reverse) to check the
result of _reverse_patch(self.dump()) in the reversed_text variable before
calling from_bytes: if reversed_text is empty, raise RuntimeError("Failed to
reverse patch; patch parsing returned empty."); otherwise proceed to call
self.from_bytes(reversed_text.encode("utf-8"))._patchset and keep the existing
is_empty() guard to catch any other parsing issues; reference functions:
reverse(), _reverse_patch, from_bytes(), is_empty(), dump(), and attribute
_patchset.
---
Nitpick comments:
In `@dfetch/commands/format_patch.py`:
- Around line 129-131: The call to output_patch_file.write_text(...) in
format_patch.py writes patch content without specifying encoding, which is
platform-dependent; update the write to pass encoding="utf-8" so the output of
patch.dump_header(patch_info) + patch.dump() is written using UTF-8 encoding
(locate the write_text call near dump_header/dump usage and add the encoding
argument).
In `@dfetch/vcs/patch.py`:
- Around line 228-231: Move the invariant regex compilations out of the
add_prefix function and promote them to module-level constants so they are
compiled once; specifically, take the current local variables diff_git and
svn_index, create module-level constants (e.g., DIFF_GIT_RE and SVN_INDEX_RE)
initialized with re.compile(...) and re.compile(rb"...") respectively, and
update the add_prefix implementation to reference these constants instead of
recompiling on each call.
- Around line 293-313: combine currently mutates input Patch objects and shares
item objects; fix by operating on copies: create a deep copy of self._patchset
(or at least deep-copy its items) into combined_patchset rather than using
combined_patchset.items += self._patchset.items, and for each patch in other
make a deep copy of patch._patchset (so convert_type is invoked on the copy, not
the original), call convert_type on that copy (using
PatchType(combined_patchset.type)), then extend combined_patchset.items with the
copied items; finally return Patch(combined_patchset). Use the unique symbols
combine, Patch, PatchType, convert_type, _patchset, and combined_patchset.items
when locating and changing the code.
- Around line 41-55: Patch currently exposes the internal _patchset as a public
dataclass constructor arg and gets a poor default __eq__; change the class to
stop exposing _patchset and to provide meaningful equality: mark the dataclass
with eq=False and make _patchset a non-init/non-compare field (e.g.,
field(init=False, compare=False, repr=False)), keep/implement factory methods
(e.g., your existing from_string/from_file constructors) to set self._patchset,
and add an explicit __eq__(self, other) that compares normalized patch content
(for example by comparing a canonical text/serialized form of self._patchset and
other._patchset) so equality reflects patch content rather than identity.
In `@tests/test_patch.py`:
- Around line 306-310: Replace the "assert False" idiom in the except block with
pytest.fail to preserve traceback context: inside the try/except around
Patch.from_file(...).apply(root=str(tmp_path)) catch, call pytest.fail(f"Reverse
patch failed: {e}") (and add "import pytest" at the top if missing) instead of
assert False so failures surface with proper test reporting and context.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
cleanup patch code and make a more coherent oo-interface
Summary by CodeRabbit
Refactor
Tests
Chores