fix: don't misclassify deletion-only hunks as already applied

[wolfv]

The bug

Since #20, a hunk containing only context and deletion lines, with ≥1 leading context line and 0 trailing context lines, is silently dropped by apply_bytes_reporting — exit code 0, no diagnostic, the patch just doesn't happen:

@@ -1,6 +1,1 @@
 l1
-l2
-l3
-l4
-l5
-l6

→ file unchanged.

Root cause: is_diff_applied_with_config declares a hunk already applied when its post-image (context + inserted lines) occurs contiguously in the file. For a deletion-only hunk whose deletions sit at the hunk's edge, the post-image is just its context lines — which the unpatched file also contains contiguously. So the hunk is classified AlreadyApplied and skipped. Any trailing context, or any insertion, makes the post-image unambiguous, which is why only this shape was affected. The mirror shape (deletion at the start of the hunk with only trailing context) had the same latent bug.

Real-world impact: this is how php-feedstock's 0001-win-iconv-compat.patch broke under rattler-build ≥ 0.70 (which bumped flickzeug to 0.5.3). The hunk emptying ext/iconv/php_iconv.def (one EXPORTS context line followed by deletions to EOF) vanished while the patch's other 5 hunks applied, leaving stale exports in php8ts.dll.def and an LNK2001 on _libiconv_version.

The fix

For a hunk with no insertions, a post-image match alone is no longer sufficient evidence: additionally require that its pre-image (context plus deleted lines) is absent from the file — i.e. the deleted lines are actually gone. Hunks with insertions keep the existing behavior, since their post-image contains lines the unpatched file doesn't have. find_post_image_position is refactored into a generic find_lines_position used for both searches.

Testing

• 3 new regression tests: the php-feedstock shape (deletion to EOF), the same shape mid-file, and the mirror case (no leading context) — each asserting Applied on the pre-image and AlreadyApplied on the post-image, so already-applied detection still works for these hunks.
• All existing tests and doctests pass (96 + 11), including test_is_diff_applied_pure_deletion and test_deletion_patch_with_stale_context_is_not_already_applied from fix: detect already-applied diffs via post-image matching #20.
• Verified end-to-end: built rattler-build 0.72.0 against this branch and ran a 7-case reproducer covering all hunk shapes — all pass, including the three previously-failing deletion shapes.

🤖 Generated with Claude Code