Repository navigation
Refresh the port inventory: what 0.0.11 → 0.0.27 added to the port #249
Description
Question
What does main at 0.0.27 actually contain that the feasibility research (measured at 2026-08-07, ~2.6K LOC / 413 tests) did not cover — and what does that do to the port's shape?
Deliverable: a refreshed inventory, as a successor doc to the feasibility research, covering:
- Source inventory by module (
dl.py,worktree/*, workspace clones + LFS, tool provisioning, benchmarking, completions, anything else), each classified as ports ~1:1, needs design, or stays language-agnostic (like the bash completion did). - Test-suite split at HEAD: which tests are mock-free and target the CLI/filesystem boundary (acceptance-harness candidates, the successors of the "61"), vs mock-based behavior spec. Count both; name the harness candidates by file.
- Crate-mapping deltas: anything the new subsystems need that the 2026-08-07 crate survey didn't cover (git worktree ops, LFS probing, PTY handling in tests, zellij/pixi provisioning) — does the "every crate exists" claim still hold?
- Subprocess surface: is it still uniformly list-form with no shell interpolation, or did anything creep in that breaks the
std::process::Command1:1 story?
Facts only; the decisions belong to the tickets this blocks. Work against origin/main (note: local checkouts may be stale).
Activity
blooop commentedon Aug 18, 2026
Resolved. Full inventory: docs/research/rust-rewrite-inventory-0.0.27.md (branch
research/rust-rewrite-inventory, measured 2026-08-18 at54d8280= v0.0.27).Headline: 0.0.11 → 0.0.27 is 298 commits, +33,915/−2,049 lines. Shipped package grew 4,016 → 11,175 Python LOC (26 files, 10 new modules: tools, timing, workspace_state, disk_usage, tty_session, devpod_ssh, devpod_provider, xdg, worktree/locks, worktree/git_errors;
dl.pyalone 1,592 → 4,909). Tests grew 657 → 1,514 functions / 30,362 lines.1. Source inventory by module — almost everything new classifies ports ~1:1 (workspace_state, disk_usage, devpod_ssh, devpod_provider, locks, git_errors, xdg, and the grown worktree managers). Needs design:
dl.py's launch-flow sequencing (still a hand-rolled dispatcher),tools.py's provisioning composition (its embedded POSIX scripts carry over verbatim),timing.py's global registry shape, and whether to portworktree/migration.pyat all. Stays language-agnostic / out of port scope:dl.bash(unchanged story) andscripts/benchmarking (~630 LOC of dev tooling drivingdlover its CLI) plus its 59 tests.2. Test split at HEAD — 312 mock-free vs 1,202 mock/patch-based. Of the mock-free, 96 exercise the CLI/filesystem/process boundary (the successors of #53's "61"): e2e 24, integration 32 (incl. real-LFS and clone-race suites), flock 19, multi-process drivers 5, PTY harness 16 — files named in the doc. The other 216 are in-process pure-function tests (port as ordinary unit tests).
test_worktree_storage.py, counted in the old 61, now patches seams and moved to the mock-based column.3. Crate deltas — four additions cover everything the new subsystems need: fd-lock 4.0.4 (flock), shlex 2.0.1 (remote-payload quoting), portable-pty 0.9.0 / expectrl 0.9.0 (PTY e2e), assert_cmd 2.2.2 (CLI harness). LFS probing and zellij/pixi provisioning need no crate (file-content sniffing + external
git lfs; generated shell strings). "Every crate exists and is maintained" still holds.4. Subprocess surface — still uniformly list-form: zero
shell=True, zeroos.system, zero f-string argv across ~33 spawn sites; thestd::process::Command1:1 story holds. One change in character:tools.py/tty_session.pynow deliberately compose remote shell payloads asshlex.quoted strings (bash -lc <script>viadevpod ssh --command,cd <dir> && cmdinside ssh argv) — interpreted in the container, not by a host shell, and reproduced in Rust by theshlexcrate. The quoting layer is now load-bearing surface.