perf(pm): add resolver provider foundation

[elrrrrrrr]

Summary

• Add the ManifestProvider boundary and mock provider support.
• Move version manifest parsing to Vec buffers and add speculative full-manifest core extraction.
• Add registry HTTP client fan-out and initial PM wiring.

Review Focus

• Provider trait shape and ownership boundaries.
• No scheduler/mainloop behavior change yet.

[gemini-code-assist]

Code Review

This pull request introduces a ManifestProvider trait to abstract manifest fetching and parsing, alongside a new HTTP client pooling mechanism to improve fan-out for registry resolution. It also optimizes manifest processing by reducing memory copies and enabling speculative version extraction during full manifest parsing. Review feedback highlighted concerns regarding the switch to unbounded channels in the DiskManifestStore, recommending the use of bounded channels and non-blocking try_send operations to prevent potential memory exhaustion and performance stalls during disk I/O.

[gemini-code-assist]

The change from a bounded channel to an unbounded one removes the safety mechanism that prevented memory growth when disk I/O stalled. Since the manifest store is opportunistic, it is safer to use a bounded SyncSender and drop writes when the queue is full to ensure memory usage remains bounded, especially given that the previous implementation explicitly handled this case.

Suggested change

	tx: Sender<ManifestWriteJob>,
	handle: JoinHandle<()>,
	}
	impl ManifestWriter {
	fn spawn() -> Self {
	let (tx, rx) = mpsc::sync_channel(MANIFEST_WRITE_QUEUE_CAPACITY);
	let (tx, rx) = mpsc::channel();
	tx: mpsc::SyncSender<ManifestWriteJob>,
	handle: JoinHandle<()>,
	}
	impl ManifestWriter {
	fn spawn() -> Self {
	let (tx, rx) = mpsc::sync_channel(1024);

[gemini-code-assist]

To maintain the non-blocking, "fire-and-forget" nature of the manifest store, use try_send instead of a blocking send. This ensures the resolver is never stalled by disk latency, which is critical for performance and consistent with the opportunistic nature of the cache.

        match self.tx.try_send(job) {
            Ok(()) => {}
            Err(mpsc::TrySendError::Full(_)) => {
                tracing::debug!("Manifest store writer queue full; dropping cache write");
            }
            Err(mpsc::TrySendError::Disconnected(_)) => {
                tracing::debug!("Manifest store writer stopped before accepting write");
            }
        }

[github-actions]

📊 pm-bench-phases · fc0b3a4 · 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	9.46s	0.10s	10.71s	10.75s	731M	324.6K
utoo-next	8.29s	0.23s	10.63s	12.49s	1007M	125.3K
utoo-npm	8.40s	0.22s	10.93s	12.69s	951M	122.3K
utoo	8.78s	0.17s	11.32s	12.98s	1.05G	142.1K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	14.8K	19.1K	1.22G	6M	1.93G	1.81G	1M
utoo-next	115.1K	80.8K	1.19G	5M	1.77G	1.77G	2M
utoo-npm	120.2K	84.4K	1.19G	5M	1.77G	1.77G	2M
utoo	132.9K	75.4K	1.19G	6M	1.77G	1.76G	2M

p1_resolve

PM	wall	±σ	user	sys	RSS	pgMinor
bun	1.92s	0.04s	4.11s	1.07s	509M	159.4K
utoo-next	2.86s	0.03s	5.43s	1.73s	622M	82.8K
utoo-npm	3.12s	0.04s	5.67s	2.13s	617M	76.1K
utoo	3.20s	0.03s	5.95s	1.95s	629M	97.5K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	8.1K	4.8K	205M	3M	111M	-	1M
utoo-next	47.6K	71.7K	203M	2M	7M	3M	2M
utoo-npm	71.5K	92.0K	203M	2M	7M	3M	2M
utoo	54.0K	74.4K	205M	3M	7M	3M	2M

p3_cold_install

PM	wall	±σ	user	sys	RSS	pgMinor
bun	7.04s	0.16s	6.48s	10.24s	653M	212.4K
utoo-next	6.40s	0.93s	5.17s	11.24s	448M	66.2K
utoo-npm	7.21s	1.29s	5.21s	11.37s	462M	61.3K
utoo	6.60s	1.07s	5.24s	11.19s	550M	67.1K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	5.3K	7.3K	1.02G	4M	1.82G	1.82G	1M
utoo-next	104.0K	49.8K	1018M	3M	1.76G	1.76G	2M
utoo-npm	112.1K	54.6K	1018M	3M	1.76G	1.76G	2M
utoo	96.3K	52.4K	1018M	2M	1.76G	1.76G	2M

p4_warm_link

PM	wall	±σ	user	sys	RSS	pgMinor
bun	3.31s	0.01s	0.22s	2.40s	135M	33.0K
utoo-next	2.30s	0.13s	0.48s	3.87s	80M	18.7K
utoo-npm	2.19s	0.03s	0.49s	3.83s	79M	18.5K
utoo	2.24s	0.09s	0.48s	3.87s	80M	18.6K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	345	30	5M	21K	1.98G	1.81G	1M
utoo-next	41.5K	18.9K	41K	11K	1.76G	1.76G	2M
utoo-npm	40.4K	19.1K	41K	24K	1.76G	1.76G	2M
utoo	42.1K	19.7K	41K	7K	1.77G	1.76G	2M

npmmirror.com: no output captured.