[Mikola Lysenko (mikolalysenko)]

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

Fixes #1833

Summary

A Cargo workspace whose members list uses a character class, such as
members = ["crates/[ab]*"], expanded to no members. Vendored Cargo,
the upstream Cargo redirect and hosted Cargo (disk and in-memory) all read
members through cargo_workspace::member_manifests*, so they never saw
those crates.

Root cause

utils/cargo_workspace.rs matched each glob segment with its own
wildcard_match, which read only * and ? and treated [ as a
literal. Cargo expands members with the glob crate, which also reads
[abc], [a-z] and [!x].

Change

• expand_from matches segments through
  workspace_globs::segment_glob_matches (the npm-family matcher), and
  wildcard_match is deleted.
• The glob crate negates only on !, so [^x] is the class of ^
  and x. The shared matcher also negates on ^, so the cargo walk
  puts a NUL member (which no directory name can hold) ahead of a ^
  that opens a class.
• New is_member_pattern: a segment is a glob when it has *, ? or a
  [..] class that closes. The member walk and the lockless Cargo
  redirect refusal (cargo_lockless_other_dependencies) both use it, so
  the refusal names crates/[ab] as a members pattern, not a missing
  literal member. An unclosed [ stays on the literal path as before.

Tests (red → green)

Issue	Test	main	this PR
#1833	cargo_workspace::tests::member_globs_read_character_classes ([ab]*, [a-b]?*, [!a]*, [^x]*)	FAIL (only unrelated libs/* hits)	pass
#1833	cargo_workspace::tests::a_bracket_class_matches_a_literal_bracket (x[[]?)	FAIL	pass
#1833	cargo_workspace::tests::member_patterns_are_wildcards_or_closed_classes	n/a (new helper)	pass
#1833	redirect::tests lockless refusal names crates/[ab] as a pattern	FAIL (named as literal member)	pass

Commands run locally:

• cargo fmt --all -- --check: clean
• cargo clippy --workspace --all-features -- -D warnings: clean
• cargo test -p socket-patch-core --all-features: 6996 passed, 4 failed. The 4 (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) depend on chmod 0555 blocking writes, which it doesn't for root (this sandbox runs as uid 0). They don't touch Cargo code.
• cargo test -p socket-patch-cli --all-features --test e2e_cargo --test e2e_redirect_cargo_shapes --test e2e_vendor_cargo_build --test e2e_redirect_cargo_build -- --include-ignored: 25 + 35 + 40 + 23 passed
• The full cargo test --workspace didn't fit in this sandbox's disk allowance (it ran out of space linking every CLI test binary), so CI's test job covers the rest.

No wrapper changes: npm/, pypi/ and gem/ don't read Cargo workspaces.

/code-review high findings not fixed here

These all predate this PR. Each rule already applied to * and ?;
the PR only extends it to classes:

• exclude is globbed. Cargo reads exclude as literal path
  prefixes, but member_manifests_with already globbed it with *.
  Making it prefix-based changes * excludes too, so that belongs in its
  own issue.
• Dot directories. A wildcard segment never matches a dot-directory,
  but cargo's glob() uses require_literal_leading_dot: false. This
  rule is older than this PR, and * already followed it.
• Backtracking cost. segment_glob_matches backtracks with no memo,
  and so did the deleted wildcard_match. This PR doesn't change the
  cost.
• Unclosed [. Cargo refuses an unclosed [ (could not parse pattern). Here it stays a literal path, as it was on main, and no new
  test locks that in.

Picked over the p1 refactors #1844 and #1839 because both would rewrite
files that open PRs #1865 and #1855 are changing. #1862 (p2) builds on
go_build_list.rs, which open PR #1843 rewrites.

https://claude.ai/code/session_011f9aeSAqJp6HmcFwxb9HL5

---

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 at 54c38dad5 (head).

• CI: 32/32 run checks green (50 skipped by path filters); mergeable, no conflicts; CHANGELOG.md untouched.
• Bugbot: reviewed 54c38da, no findings; no open review threads.
• For the reviewer: Cargo workspace members globs now honour [..] character classes (Cargo workspace member globs ignore [...] character classes, so members = ["crates/[ab]*"] expands to no members #1833); the PR description lists four pre-existing glob quirks deliberately left out of scope.
• Slack: not announced; Slack posting isn't available in this run, so the next run retries.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Final review brief: 54c38da

What it does. Cargo workspace members globs with a character class (crates/[ab]*, [!x]*) expanded to nothing, because cargo_workspace had its own matcher that understood only * and ?. The member walk now uses the shared workspace_globs::segment_glob_matches matcher, with a cargo-specific tweak so [^x] stays literal ^/x, as in the glob crate. The lockless redirect refusal now also reports class entries as patterns.

Risk: low. The change is limited to member-glob expansion (vendored, redirect and hosted Cargo all share it). Patterns using only */? match exactly as before, and an unclosed [ stays literal.

Look here

• cargo_workspace.rs:383-405:`` is_member_pattern / `segment_is_pattern` and the `[^` → `[\0^` rewrite
• cargo_workspace.rs:421 and :445-451:``` the walk switched to the shared matcher
• cargo_workspace.rs:633-687:```` new tests ([ab]*, `[a-b]?`, `[!a]`, `[^x]*`, `[[]`)
• redirect/mod.rs:2722: the refusal message uses the new helper

Verified

• Read the full diff (2 files, +98/−26). CHANGELOG.md is untouched. No other member-glob check still uses the old */?-only test (grepped).
• Ran cargo test -p socket-patch-core --lib -- cargo_workspace:: lockless: 23/23 pass. The old matcher read [ as a literal, so member_globs_read_character_classes can't pass without the fix.
• CI: 82 checks on the head (32 success, 50 skipped, 0 failing). ci-ok and clippy are green. Bugbot: no issues on 54c38da. No open review threads.

Changes I made: none.

Open questions (non-blocking): segment_is_pattern strips a leading ^ as if it negated, so the degenerate member x[^] is still read as a literal directory name. The glob crate would read it as the class {^}. Nobody is likely to write that.

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

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

[agent] Fixed the review brief's open question in d177158: segment_is_pattern now treats only ! as negation, as cargo's glob crate does, so x[^] is read as the class {^} instead of a literal directory. Added to member_patterns_are_wildcards_or_closed_classes.

---

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 d177158. Configure here.}

[Mikola Lysenko (mikolalysenko)]

Final review brief: d177158 (follow-up to the 54c38da brief above)

What it does. Cargo workspace members globs with a character class (crates/[ab]*, [!x]*) expanded to nothing, because cargo_workspace had its own matcher that understood only * and ?. The member walk now uses the shared workspace_globs matcher, and [^x] stays a literal ^/x class, as in cargo's glob crate. Since the last brief, one commit (d177158) also reads the degenerate x[^] as the class {^}, which settles the open question from the earlier brief.

Risk: low. Only member-glob expansion changes. Patterns that use only */? match as before, and an unclosed [ stays literal. The new commit changes one strip_prefix call.

Look here

• cargo_workspace.rs:390-397:`` segment_is_pattern now strips only `!` as a negation, matching the `glob` crate
• cargo_workspace.rs:383-405:``` is_member_pattern and the `[^` → `[\0^` rewrite
• cargo_workspace.rs:680-694:```` the pattern table, which now includes x[^]
• redirect/mod.rs:2722: the lockless refusal uses the new helper

Verified

• Read the 54c38da..d177158 delta (one commit, +10/−2). The earlier brief covers the rest of the diff. Traced x[^], x[^a], [^]] and x[!] through segment_is_pattern and the [\0^ rewrite: only x[^] changes classification (literal → class {^}), and that matches the glob crate. The new x[^] table entry fails on the old code. CHANGELOG.md is untouched.
• Ran cargo test -p socket-patch-core --lib -- cargo_workspace:: lockless on d177158: 23/23 pass.
• CI on the head: ci-ok and clippy are green, with no failing checks. Bugbot: success on d177158. No open review threads.

Changes I made: none.

Open questions: none.

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

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

[burn-down agent] Ready for review confirmed at d177158 (the head changed after the earlier label comment at 54c38da).

• CI: every check on d177158 is green (32 success, 50 skipped); mergeable.
• Bugbot: reviewed d177158, no findings. No unresolved review threads.
• Slack: not announced. The Slack send tool isn't available in this run.

---

Generated by Claude Code

[Wenxin Jiang (Wenxin-Jiang)]

[agent] Reviewed d17715810ae593a1f79b43db53d876d6f5c1e8f2. Checked Cargo member/exclude character classes, literal brackets and shared segment matching. No major correctness issues found. Validation: source and regression-test review plus the available exact-head CI checks; no local test pass is claimed.