Refactor flake-status to process repos in parallel #72

Merged

vnprc merged 2 commits from agent/parallel-flake-status into master

2026-06-24 22:51:08 +01:00

vnprc-agent commented

2026-06-24 22:14:56 +01:00

Contributor

Closes #61

Refactor flake-status to process repos in parallel instead of serially.

Changes

Add _parallel_for bounded concurrency loop using wait -n, controlled by FLAKE_STATUS_JOBS env var (default 4)
Split per-repo data collection into _collect_named and _collect_all functions that write results to indexed temp files
Replace in-process _LS_REMOTE_CACHE associative array with _resolve_all_upstreams which deduplicates URLs via sort -u and resolves each unique URL in parallel
Render from temp files in workspace index order so output stays deterministic
No new dependencies beyond Bash, Git, and jq

Validation

All 21 existing tests/flake-status.sh tests pass
Output order is deterministic (temp files indexed by repo position, read back in workspace order)
flake-status, flake-status <input>, and both --upstream modes produce identical output to the serial version
FLAKE_STATUS_JOBS overrides the default parallelism bound
Duplicate upstream URLs are resolved exactly once via deduplication before the parallel resolve phase

Closes #61 Refactor `flake-status` to process repos in parallel instead of serially. ## Changes - Add `_parallel_for` bounded concurrency loop using `wait -n`, controlled by `FLAKE_STATUS_JOBS` env var (default 4) - Split per-repo data collection into `_collect_named` and `_collect_all` functions that write results to indexed temp files - Replace in-process `_LS_REMOTE_CACHE` associative array with `_resolve_all_upstreams` which deduplicates URLs via `sort -u` and resolves each unique URL in parallel - Render from temp files in workspace index order so output stays deterministic - No new dependencies beyond Bash, Git, and jq ## Validation - All 21 existing `tests/flake-status.sh` tests pass - Output order is deterministic (temp files indexed by repo position, read back in workspace order) - `flake-status`, `flake-status <input>`, and both `--upstream` modes produce identical output to the serial version - `FLAKE_STATUS_JOBS` overrides the default parallelism bound - Duplicate upstream URLs are resolved exactly once via deduplication before the parallel resolve phase

vnprc-agent added 1 commit

2026-06-24 22:14:56 +01:00

Refactor flake-status to process repos in parallel 6e7a681222

vnprc force-pushed agent/parallel-flake-status from 6e7a681222 to 55c863602b

2026-06-24 22:17:49 +01:00

Compare

vnprc-agent commented

2026-06-24 22:22:27 +01:00

Author

Contributor

Review findings (subagent)

Numbered by severity. Fixes incoming for the legitimate ones.

Fix (real bugs)

Regex injection in grep (line 444) — grep -m1 "^${name}"$'\t' injects $name raw into a regex. If an input name contained . or * it could match wrong entries. Should use awk with string equality instead.
No validation of FLAKE_STATUS_JOBS — Non-integer values cause an arithmetic error and confusing exit. Add a guard.
Fragile glob when urls/ is empty (line 186) — cat "$_PTMP"/urls/* with no matching files relies on error suppression to work. Use explicit nullglob guard for clarity.

Not fixing (reviewer errors)

Trap single-quote evaluation — Reviewer claims single quotes cause $_PTMP to be evaluated at registration time. This is backwards: single quotes defer evaluation to trap execution time. The code is correct as written.
wait -n portability — Reviewer says bash 5.1+; wait -n was actually added in bash 4.3 (2014). NixOS ships bash 5.x. Not a concern for this codebase.
wait (no args) exit status — Reviewer worries this triggers set -e. Per the bash manual, wait without arguments returns 0. Not an issue.

## Review findings (subagent) Numbered by severity. Fixes incoming for the legitimate ones. ### Fix (real bugs) 1. **Regex injection in grep (line 444)** — `grep -m1 "^${name}"$'\t'` injects `$name` raw into a regex. If an input name contained `.` or `*` it could match wrong entries. Should use `awk` with string equality instead. 2. **No validation of `FLAKE_STATUS_JOBS`** — Non-integer values cause an arithmetic error and confusing exit. Add a guard. 3. **Fragile glob when `urls/` is empty (line 186)** — `cat "$_PTMP"/urls/*` with no matching files relies on error suppression to work. Use explicit nullglob guard for clarity. ### Not fixing (reviewer errors) 4. **Trap single-quote evaluation** — Reviewer claims single quotes cause `$_PTMP` to be evaluated at *registration* time. This is backwards: single quotes *defer* evaluation to trap execution time. The code is correct as written. 5. **`wait -n` portability** — Reviewer says bash 5.1+; `wait -n` was actually added in bash 4.3 (2014). NixOS ships bash 5.x. Not a concern for this codebase. 6. **`wait` (no args) exit status** — Reviewer worries this triggers `set -e`. Per the bash manual, `wait` without arguments returns 0. Not an issue.

vnprc-agent added 1 commit

2026-06-24 22:24:06 +01:00

Fix review findings: grep injection, FLAKE_STATUS_JOBS validation, glob safety 6be045d8b1