Optimize add dependency resolution

[cijiugechu]

Reuse project state and carry pre-resolved npm metadata into install to avoid duplicate add-time resolver work. Keep the new packument cache scoped to the add path so cold installs stay unchanged.

Benchmark

All benchmarks were run locally on macOS with release binaries.

Warm repeated add kleur

This benchmark measures a warm repeated add kleur path by removing kleur before each iteration.

Baseline binary:
/tmp/zpm-base-bench/target/release/yarn-bin
Patched binary:
/Users/nemurubaka/repos/zpm/target/release/yarn-bin

# Warmup for the baseline binary:
"/tmp/zpm-base-bench/target/release/yarn-bin" remove kleur >/dev/null 2>&1 || true
"/tmp/zpm-base-bench/target/release/yarn-bin" add kleur >/dev/null 2>&1
"/tmp/zpm-base-bench/target/release/yarn-bin" remove kleur >/dev/null 2>&1

# Warmup for the patched binary:
"/Users/nemurubaka/repos/zpm/target/release/yarn-bin" remove kleur >/dev/null 2>&1 || true
"/Users/nemurubaka/repos/zpm/target/release/yarn-bin" add kleur >/dev/null 2>&1
"/Users/nemurubaka/repos/zpm/target/release/yarn-bin" remove kleur >/dev/null 2>&1

# Measured with hyperfine:
hyperfine --warmup 3 --min-runs 20 \
  --export-json /tmp/zpm-add-compare-base-kleur.json \
  --prepare '"/tmp/zpm-base-bench/target/release/yarn-bin" remove kleur >/dev/null 2>&1' \
  '"/tmp/zpm-base-bench/target/release/yarn-bin" add kleur'

hyperfine --warmup 3 --min-runs 20 \
  --export-json /tmp/zpm-add-compare-current-kleur.json \
  --prepare '"/Users/nemurubaka/repos/zpm/target/release/yarn-bin" remove kleur >/dev/null 2>&1' \
  '"/Users/nemurubaka/repos/zpm/target/release/yarn-bin" add kleur'

Results:

• Before patch: 684.8ms ± 226.0ms
• After patch: 10.2ms ± 0.6ms

[github-actions]

⏱️ Benchmark Results

gatsby install-full-cold

Metric	Base	Head	Difference
Mean	2.806s	2.821s	+0.54% ⚠️
Median	2.803s	2.815s	+0.41% ⚠️
Min	2.761s	2.758s
Max	2.853s	2.993s
Std Dev	0.024s	0.041s

📊 Raw benchmark data (gatsby install-full-cold)

Base times: 2.803s, 2.835s, 2.803s, 2.761s, 2.813s, 2.823s, 2.779s, 2.782s, 2.775s, 2.827s, 2.839s, 2.847s, 2.800s, 2.802s, 2.787s, 2.811s, 2.789s, 2.790s, 2.845s, 2.853s, 2.828s, 2.803s, 2.818s, 2.818s, 2.815s, 2.807s, 2.782s, 2.774s, 2.801s, 2.774s

Head times: 2.993s, 2.758s, 2.873s, 2.838s, 2.838s, 2.835s, 2.839s, 2.801s, 2.825s, 2.831s, 2.825s, 2.799s, 2.814s, 2.815s, 2.827s, 2.788s, 2.799s, 2.810s, 2.805s, 2.806s, 2.787s, 2.841s, 2.815s, 2.807s, 2.820s, 2.805s, 2.807s, 2.789s, 2.872s, 2.774s

---

gatsby install-cache-and-lock (warm, with lockfile)

Metric	Base	Head	Difference
Mean	0.409s	0.416s	+1.81% ⚠️
Median	0.407s	0.416s	+2.31% ⚠️
Min	0.398s	0.408s
Max	0.425s	0.426s
Std Dev	0.007s	0.004s

📊 Raw benchmark data (gatsby install-cache-and-lock (warm, with lockfile))

Base times: 0.405s, 0.420s, 0.422s, 0.415s, 0.406s, 0.402s, 0.410s, 0.405s, 0.417s, 0.414s, 0.403s, 0.403s, 0.413s, 0.407s, 0.404s, 0.406s, 0.412s, 0.413s, 0.406s, 0.414s, 0.413s, 0.425s, 0.402s, 0.398s, 0.406s, 0.405s, 0.409s, 0.412s, 0.402s, 0.399s

Head times: 0.416s, 0.411s, 0.409s, 0.408s, 0.416s, 0.418s, 0.421s, 0.425s, 0.419s, 0.420s, 0.411s, 0.415s, 0.417s, 0.414s, 0.410s, 0.419s, 0.410s, 0.416s, 0.416s, 0.420s, 0.416s, 0.413s, 0.419s, 0.419s, 0.418s, 0.426s, 0.421s, 0.416s, 0.419s, 0.414s

[arcanis]

I think before this optimization I'd be interested we first try to implement the metadata cache originally implemented for Berry in yarnpkg/berry#5491 (also a nit, we don't use the term "packument", we use "metadata").