[Mikola Lysenko (mikolalysenko)]

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

Architecture refactor sweep of core utils/http.rs, utils/line_endings.rs (clean, nothing to do), utils/mod.rs, utils/hatch_workspace.rs and utils/hatch.rs. It follows the audit notes in coverage/files.tsv on arch-audit/ledger (living document: discussion #1830, Part 4 structure and Part 5 tests).

Changes

utils/http.rs

• Tests fixed: the cap tests matched "too large" in a String. They now assert the ReadCappedError variant and the exact message.
• Tests added: the two branches no test reached:
  • streamed_body_over_cap_is_refused_mid_stream: a chunked body with no Content-Length, so only the streamed count catches it.
  • body_cut_short_of_its_content_length_is_truncated: covers Truncated.
• Red→green: disabling the mid-stream check fails the first test. Mapping a stream error to CapExceeded fails the second.
• Test server: a raw loopback server replaces wiremock, which always declares a length.
• Comments: the history note on read_capped and the (#1107) banner are removed.
• Rename: production_builder_builds becomes production_builder_accepts_the_shared_tls_config. Its doc had claimed a platform-store check that it never asserted.
• Not done: the String read_capped wrapper stays. Deleting it touches vendor/maven_repo.rs (Fix Maven 4 implicit subprojects missed outside vendored (#1809) #1886, Build: share one test env guard in core (#1889) #1891) or api/client.rs (One serve GET and one vendor retry loop in api/client.rs (#1869) #1870, Build: share one test env guard in core (#1889) #1891).

utils/hatch.rs

• Named documents: PYPROJECT and HATCH_TOML replace the HATCH_FILES[0]/[1] indexing, including 3 sites in vendor/pypi_hatch.rs.
• One merge rule: merged_key encodes "a hatch.toml top-level key replaces [tool.hatch]'s key". It replaces the hand copy in locks_environments and the environments forwarder, which is deleted. tool_hatch replaces 3 open-coded tool.hatch lookups.
• Round::new: builds the targets and index that plan and rewrite_all each built by hand (2 × ~20 lines).
• Target::root_relative: replaces 3 {root:uri} prefix literals.
• Parsing: require_pip_installer reuses parse_documents.
• external_metadata: replaces collecting every hatch.toml key into a Vec just to test for metadata.
• Comments: 13 bare issue references removed (Vendored Hatch wiring makes every Hatch install fail when the checkout path contains %XX, # or ; (e.g. a Jenkins "feature%2Fx" workspace), because Hatch's {root:uri} only escapes spaces #547, Hatch 1.17+ projects with a hatch-generated pylock.toml get only the lock rewritten, and Hatch regenerates it from pyproject, so hosted and vendored patches are silently dropped #479, Hosted Hatch scan exits 0 "redirected" when [tool.uv.sources] pins the patched package, but Hatch's uv installer obeys the source: url/path sources install the unpatched wheel and git sources break every install #673, Hatch never picks up a superseding patch: re-scan refuses its own earlier wiring ("existing direct source must be reverted"), so hosted exits 0 still pinned to the old patch uuid #650, Hosted requirements.txt rewrite skips PEP 440-equivalent pins like six==1.16 for an installed 1.16.0, so scan exits 0 and pip installs the unpatched release (regression from v4.0.0) #475, Perf regression: hatch/hosted wall +21% (9c43dfc9..17793173) #1471).

utils/hatch_workspace.rs

• Inlined: expand, a single caller, into members_of, with its doc merged.
• Flattened: a constant take_while(|_| !at_checkout) becomes an if.
• Named documents: used in place of the index constants.
• Comments: review and issue history (Vendored scan in a Hatch workspace member writes {root:uri} that the workspace root env never expands, so every hatch run at the root fails to install #505, Bugbot, review) dropped from 5 test docs.

utils/mod.rs

• The hatch modules are sorted into place, and hatch_workspace is pub(crate): its only users are in core.

/code-review high

I fixed 3 findings:

• The HatchPermission → HatchEdit merge is reverted. A plan's permission holds whole file texts, while a HatchEdit holds one line.
• The environments forwarder is deleted.
• The members_of doc is rewrapped.

I left 5 unfixed:

• Visibility narrowing on hatch_workspace: the audit asked for it, and prior sweeps narrowed core items the same way.
• require_pip_installer_documents ignores the hatch.toml envs override: this is pre-existing, and fixing it changes refusals, so it wants an issue.
• walk's key-stripping merge: it mutates a copy for the rewrite, a different operation from a lookup.
• respond_raw duplicating other modules' private test servers: sharing them is test_support work across files held by open PRs.
• #547/#505 references in vendor/pypi_hatch.rs: that file isn't in this sweep.

Totals

git diff --stat: 5 files, +231 / −208.

• Production: about −19.
• Tests: about +42, from the two new branch tests.

Behavior

None. The output bytes, error strings and refusal order are unchanged.

Verification

• CI: all 82 checks green on ece45b6. Bugbot found no issues.
• cargo clippy --workspace --all-features -- -D warnings is clean.
• cargo test -p socket-patch-core --lib: 6997 pass and 4 fail. All 4 rely on chmod and fail under the root sandbox, in files this PR doesn't touch:
  • copy_tree::relax_loop_must_not_traverse_symlinked_root
  • vlt_heal::an_unremovable_hidden_lock_keeps_every_store_entry
  • pypi_poetry::wire_write_failure_maps_error_and_leaves_lock_untouched
  • pypi_requirements::wire_failure_rolls_back_already_written_files
• Mutation check: making merged_key ignore hatch.toml for lock-envs fails hatch_lock_detection_follows_the_config_merge.

Risk

Low. The changes are local to core, and no cfg(windows)/cfg(unix) code changed.

https://claude.ai/code/session_011f9aeNjuH4Ykvk5p4WEcU3

---

Note

Low Risk
Structural refactors and test harness changes in core utils; PR asserts byte-identical outputs and unchanged refusal order.

Overview
Refactors Hatch TOML handling without changing rewrite or refusal behavior. hatch.rs introduces PYPROJECT / HATCH_TOML, centralizes pyproject [tool.hatch] access via tool_hatch, and replaces ad hoc hatch.toml-vs-pyproject merging with merged_key (replacing the old environments helper). Round::new, Target::root_relative, external_metadata, and shared parse_documents cut duplicated setup; pypi_hatch.rs follows the new path constants.

hatch_workspace.rs uses the same merge helper and document names, inlines member-spec {root}/{home} expansion into members_of, and simplifies checkout-root discovery in roots. mod.rs reorders exports and makes hatch_workspace pub(crate).

http.rs test-only changes: wiremock is replaced with a raw loopback server so cap logic can be exercised on chunked bodies and truncated Content-Length responses; assertions target ReadCappedError variants. Stale issue/history comments are stripped across these files.

^{Reviewed by Cursor Bugbot for commit e598769. Configure here.}

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

BugBot review

---

Generated by Claude Code

[cursor]

Stale Bugbot comment from a previous run.

[Mikola Lysenko (mikolalysenko)]

Burn-down agent: labeled Ready for review.

• Head: ece45b6b47b928f21085641b1b88a5cb85178e78, mergeable.
• CI: 32/32 ran green (50 skipped by path filters).
• Bugbot: reviewed this head, no new issues; no open review threads.
• Reviewer focus: refactor-only sweep of utils/http.rs, utils/hatch.rs, hatch_workspace.rs — check behavior is unchanged.
• Slack announcement could not be sent this run (send tool unavailable); next run retries.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Final review brief (ece45b6)

What it does. This cleans up utils/hatch.rs, hatch_workspace.rs, http.rs and utils/mod.rs without changing production behavior. Two hand-written copies of the hatch.toml-over-pyproject merge rule become one merged_key helper. Round::new replaces two duplicated constructors, expand is inlined, and hatch_workspace is narrowed to pub(crate). The http tests move from wiremock to a raw loopback server so they can reach the mid-stream cap and Truncated branches. In http.rs, the only production change is a doc comment.

Risk: low. I compared each rewrite with the old code and found them equivalent:

• merged_key vs environments: and_then(get) vs filter(contains_key) plus get. Both skip Item::None.
• locks_environments: hatch.toml still wins on a top-level key.
• external_metadata: same check as before.
• require_pip_installer: it now iterates the documents in BTreeMap order, but it returns one constant refusal on any match, and parse errors keep HATCH_FILES order, so the result is the same.
• members_of: same order of {root}/{home} substitution and the same guards.
• read_capped: byte-for-byte unchanged.

Look here

• merged_key, the single merge rule
• Round::new
• require_pip_installer, now via parse_documents
• members_of and roots
• raw loopback test server

Verified

• Read the full diff and the description. The description matches the code: 13 issue references removed, 3 sites changed in pypi_hatch.rs, and nothing outside core uses hatch_workspace.
• Tests:
  • cargo test -p socket-patch-core --lib utils::: 434 passed
  • --lib vendor::pypi_hatch: 15 passed
  • --lib hatch: 74 passed
• cargo clippy -p socket-patch-core --lib -- -D warnings is clean. The --all-targets lint errors are already on main and are not from this PR.
• CI on ece45b6: 82 check runs, 32 passed and 50 skipped. ci-ok and clippy are green, and Bugbot passed.
• No open review threads. CHANGELOG.md is untouched, and I found no debug leftovers.

Changes I made: none.

Open questions (not blocking)

• hatch_workspace goes from pub to pub(crate) in the published socket-patch-core crate. Nothing in this workspace uses it, and earlier sweeps did the same, but a downstream user of the crate would break.
• exact_cap_boundary_is_allowed now sends a chunked body. That means no test covers the declared Content-Length len == max boundary any more. A Content-Length variant could be added in a follow-up.

Auto-merge is armed, so approving sends this straight to the merge queue.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

[agent] Final review follow-up: e598769 adds back the case where the declared Content-Length equals the cap. exact_cap_boundary_is_allowed now checks a declared content-length: 16 and a chunked 16-byte body, both against a cap of 16. Changing the declared check to len >= max fails it.

I'm leaving hatch_workspace as pub(crate). The audit asked for it, nothing in the workspace uses it, and earlier sweeps narrowed core items the same way.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

BugBot review

---

Generated by Claude Code

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit e598769. Configure here.}

[Mikola Lysenko (mikolalysenko)]

[final reviewer] Disarmed auto-merge: head e598769 adds a test-only commit after the final-review brief on ece45b6, so the brief no longer covers the current head. The push was under 45 minutes ago, so I am leaving it alone this run. The next final-review run will review e598769 (Bugbot was requested on it) and re-arm auto-merge once the brief is current. CI (ci-ok, clippy) is green on e598769.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Final review brief (e598769)

This covers the one commit added after the previous brief on ece45b6. Everything in that brief still applies.

What it does. This is a refactor-only sweep of utils/hatch.rs, hatch_workspace.rs, http.rs and utils/mod.rs, with no intended behavior change. The new commit e598769 only changes a test: exact_cap_boundary_is_allowed now checks a declared content-length: 16 body as well as a chunked 16-byte body, both against a cap of 16. That closes the coverage gap the previous brief raised.

Risk: low. The commit doesn't touch production code.

Look here

• http.rs:333: the two-case boundary test
• http.rs:221: the declared-length check it pins
• Merge rule and walkers, unchanged since the last brief: merged_key and members_of

Verified

• cargo test -p socket-patch-core --lib utils::http on e598769: 10 passed.
• Mutation check: changing line 221 to len >= max makes exact_cap_boundary_is_allowed fail. I reverted the change and didn't push it.
• CI on e598769: ci-ok and clippy are green. Bugbot found no new issues, and there are no review threads. CHANGELOG.md is untouched.

Changes I made: none.

Open question (not blocking): hatch_workspace goes from pub to pub(crate) in the published core crate. Nothing in this workspace uses it.

Auto-merge is armed, so approving sends this straight to the merge queue.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Burn-down agent: Ready for review re-confirmed on the new head.

• Head: e598769eb91891e2120a0cded1a6a6d973367bf0 (test-only commit added after the earlier label on ece45b6), mergeable.
• CI: 50/50 check runs green on this head.
• Bugbot: reviewed e598769, no findings; no open review threads.
• Slack announcement could not be sent this run (send tool unavailable); next run retries.

---

Generated by Claude Code