Fix mixed line endings lost in vendored requirements (#1567) - #1612

LLM Description written by Claude Code:claude-opus-5-5

Fixes #1567

Summary

A vendored rewrite of a requirements.txt (or a -r include) that
mixes CRLF and LF lines no longer re-terminates lines it never touched.
vendor --revert and remove now restore the file byte for byte.

Root cause

The three vendored requirements writers in vendor/pypi_requirements.rs
(plan_requirements, plan_rewire for the superseding re-vendor, and
revert_requirements) split the file with str::lines(), which drops
every \r. They then re-joined it with a single terminator, the file's
majority ending. So every minority-style line changed on vendor and again
on revert, while both reported success.

Fix

• The writers now edit physical lines that keep their own terminator
  (Line { text, eol }), the way the hosted rewriter
  (patch/redirect/requirements.rs) already does. A vendor line ends
  the way the pin it replaces ended, and every other line is written back
  unchanged.
• A replaced span (a hash-pinned \-continuation) records each
  earlier line's ending when it differs from the span's last line. Then
  even a continuation that mixes endings internally reverts exactly. A
  uniform span records plain texts as before, so ledgers for ordinary
  files don't change, and ledgers written before this PR still revert.
• The ending-preserving splitter moved to utils::line_endings::split_lines,
  shared with the vendored Cargo.toml writer (vendor/cargo_manifest.rs),
  which had its own copy.

Tests (per issue)

• Vendored requirements.txt rewrite re-joins every line of a mixed CRLF/LF file with one terminator, so untouched lines change and vendor --revert / remove can't restore the file #1567: mixed_line_endings_survive_vendor_and_revert covers the LF-majority
  and CRLF-majority repros from the issue, plus a CRLF pin before an
  unterminated last line. mixed_line_endings_around_a_continuation_span,
  a_continuation_span_with_mixed_endings_reverts_byte_identically,
  mixed_line_endings_survive_rewire_and_revert and
  mixed_line_endings_survive_a_transitive_append cover the other
  writers, and a_uniform_span_records_plain_texts covers the
  unchanged ledger.
• Red → green: on main's writer, the first three new tests failed with
  exactly the issue's symptom, for example
  left: "idna==3.7\n./…whl # socket-patch vendor: six==1.16.0\nrequests==2.31.0\n# tail\n"
  against the expected "idna==3.7\r\n…\nrequests==2.31.0\r\n# tail\n".
  They pass with this change.

Verification run locally

• cargo fmt --all -- --check: pass
• cargo clippy --workspace --all-features -- -D warnings: pass
• cargo test -p socket-patch-core --all-features: 6540 passed. 4 failed,
  and the same 4 fail on main: they are read-only-directory tests that
  can't fail a write when the sandbox runs as root (wire_failure_rolls_back_already_written_files,
  wire_write_failure_maps_error_and_leaves_lock_untouched,
  relax_loop_must_not_traverse_symlinked_root,
  an_unremovable_hidden_lock_keeps_every_store_entry). CI runs as non-root.
• cargo test -p socket-patch-cli --all-features --lib: 921 passed
• e2e_vendor_pypi_build -- --include-ignored: 55 passed.
  in_process_vendor_pypi_takeover, mode_migration_pypi,
  scan_vendor_requirements_unwired and scan_requirements_lock_only
  all pass. e2e_vendor_cargo_build -- --include-ignored: 40 passed
  (shared splitter).
• I didn't run the full cargo test --workspace locally: linking every
  CLI test binary filled the sandbox disk. CI covers it.
• No wrapper changes (npm/, pypi/, gem/ don't touch this path).

/code-review high findings

Fixed: the mixed-ending continuation span, the duplicated splitter,
missing tests for the span and append shapes, and simplified restore
logic. Not fixed:

• Reverting a transitive (Added) line in a root file that had no
  final newline leaves the separator wire added. This predates this PR
  and is unrelated to mixed endings. The existing test
  transitive_append_to_root_without_trailing_newline documents it. A
  fix needs a new record shape for Added, which other readers
  (lock_inventory/recover.rs) key on, so it belongs in its own change.
• Borrowing line text instead of cloning: requirements files are small,
  and the extra copy isn't measurable next to the I/O.

🤖 Generated with Claude Code

Generated by Claude Code