perf(pm): skip missing clean deps roots

[elrrrrrrr]

Summary

Small AB experiment for p3/p4 install overhead.

clean_deps currently always enters stale-entry cleanup and unused-package globbing for the root node_modules, even when the directory does not exist. In phase-isolated p3/p4, the prepare step removes node_modules before each run, so this path should be a no-op.

This change skips each node_modules root when it is missing before running stale cleanup and package globbing. Existing workspace node_modules directories are still discovered and cleaned when present.

Hypothesis

This should remove a small amount of cold/warm install filesystem probing and glob setup in p3/p4. Expected impact is small; keep only if same-run p3/p4 wall or ctx moves positively.

Validation

Passed:

• cargo fmt
• cargo test -p utoo-pm service::clean::tests::test_clean_deps_skips_missing_node_modules
• cargo clippy -p utoo-pm --all-targets -- -D warnings --no-deps

Full workspace clippy is not repeated here; it is currently blocked in this worktree by the known pack-core/next.js API mismatch observed on #2969 validation.

Benchmark Plan

Label-trigger GHA bench, compare p3/p4 against same-run utoo-next, utoo-npm, and bun.

[gemini-code-assist]

Code Review

This pull request updates the clean_deps function to skip missing node_modules directories and adds a corresponding test case to verify this behavior. The review feedback highlights a cross-platform compatibility issue regarding hardcoded path separators in existing logic and suggests further performance optimizations by reordering the existence checks and deferring the construction of the valid_packages set.

[gemini-code-assist]

Note that the cleaning logic following this check (specifically remove_unused_packages and its dependency path_to_pkg_name) contains hardcoded forward slashes (e.g., node_modules/ at line 336) and lacks path separator normalization when checking valid_packages. On Windows, where the path separator is , these checks will fail, effectively making the dependency cleaning a no-op. While this issue predates the current PR, it significantly impacts the correctness of the feature being optimized.

[gemini-code-assist]

This existence check is redundant for workspace node_modules directories, as they are already verified to exist at line 156 before being added to nm_dirs. For the root node_modules directory, it would be more efficient to perform this check at line 153 when it is first added to the vector. Additionally, the construction of the valid_packages set (lines 146-149) could be deferred until after verifying that at least one node_modules directory exists, which would further improve performance in the 'missing node_modules' scenario.

[github-actions]

📊 pm-bench-phases · 78453b3 · linux (ubuntu-latest)

Workflow run — ant-design

PMs: utoo (this branch) · utoo-npm (latest published) · bun (latest)

npmjs.org

p0_full_cold

PM	wall	±σ	user	sys	RSS	pgMinor
bun	8.99s	0.23s	10.46s	10.16s	758M	344.2K
utoo-next	8.10s	0.10s	10.51s	12.26s	967M	124.6K
utoo-npm	8.00s	0.15s	10.99s	12.55s	993M	121.0K
utoo	8.45s	1.33s	11.61s	12.58s	930M	149.0K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	14.1K	17.7K	1.20G	6M	1.88G	1.76G	1M
utoo-next	117.8K	80.6K	1.17G	5M	1.73G	1.72G	2M
utoo-npm	132.7K	90.3K	1.17G	5M	1.73G	1.72G	2M
utoo	124.5K	90.0K	1.17G	6M	1.72G	1.72G	2M

p1_resolve

PM	wall	±σ	user	sys	RSS	pgMinor
bun	1.89s	0.01s	4.09s	1.07s	536M	168.0K
utoo-next	2.96s	0.02s	5.49s	1.99s	615M	79.0K
utoo-npm	3.00s	0.03s	5.56s	2.00s	626M	77.9K
utoo	2.41s	0.03s	6.07s	1.65s	647M	121.3K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	7.8K	4.8K	203M	3M	108M	-	1M
utoo-next	69.8K	85.8K	201M	2M	7M	3M	2M
utoo-npm	70.4K	89.7K	201M	2M	7M	3M	2M
utoo	15.2K	18.8K	204M	3M	7M	3M	2M

p3_cold_install

PM	wall	±σ	user	sys	RSS	pgMinor
bun	7.31s	0.28s	6.37s	9.99s	607M	212.5K
utoo-next	6.85s	1.71s	5.07s	11.16s	499M	61.1K
utoo-npm	7.09s	1.78s	5.19s	11.09s	507M	60.2K
utoo	5.69s	0.17s	5.13s	11.39s	485M	59.8K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	4.5K	6.7K	1.00G	3M	1.77G	1.77G	1M
utoo-next	109.3K	48.8K	1000M	3M	1.72G	1.72G	3M
utoo-npm	107.3K	48.6K	1000M	3M	1.72G	1.72G	3M
utoo	120.3K	77.0K	1000M	3M	1.72G	1.72G	3M

p4_warm_link

PM	wall	±σ	user	sys	RSS	pgMinor
bun	3.36s	0.06s	0.19s	2.42s	134M	31.6K
utoo-next	2.33s	0.05s	0.54s	3.89s	80M	18.4K
utoo-npm	2.31s	0.08s	0.53s	3.88s	80M	18.3K
utoo	2.33s	0.11s	0.56s	4.00s	63M	13.9K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	254	23	27K	25K	1.88G	1.75G	1M
utoo-next	43.7K	19.1K	2K	6K	1.72G	1.72G	3M
utoo-npm	44.1K	20.9K	2K	19K	1.72G	1.72G	3M
utoo	54.8K	25.4K	2K	26K	1.73G	1.72G	3M

npmmirror.com: no output captured.

[elrrrrrrr]

GHA run 1 read

Run: https://github.com/utooland/utoo/actions/runs/26017207217

phase	utoo wall	utoo ctx	same-run utoo-next	same-run utoo-npm	same-run bun	read
p3_cold_install	5.69s ±0.17	120.3K / 77.0K	6.85s, 109.3K / 48.8K	7.09s, 107.3K / 48.6K	7.31s, 4.5K / 6.7K	wall looks positive, but ctx regresses materially
p4_warm_link	2.33s ±0.11	54.8K / 25.4K	2.33s, 43.7K / 19.1K	2.31s, 44.1K / 20.9K	3.36s, 254 / 23	no wall win and ctx worse

Conclusion: reject as a perf fold. The missing-node_modules clean skip does not reduce the p4 no-op path, and p3 wall alone is not enough when vCtx/iCtx are worse than same-run next/npm. The p3 wall improvement is likely run noise rather than the clean step being a real lever.

[elrrrrrrr]

Closing this PM performance experiment after the investigation phase. The benchmark data and conclusions are preserved in the PR body/comments; we will split the validated pieces into smaller reviewable PRs for the formal ship path.