ci(integrated-benchmark): isolate the pnpr acceleration regression

Temporary: benchmark pnpr at revisions cdc897f838 (pre), 85d9e7708a
(pnpm/pnpm#12579), and 88add0909a (pnpm/pnpm#12577), each paired with its
direct-install baseline, on one runner to pinpoint which commit erased pnpr's
speed advantage over direct installs.
This commit is contained in:
Zoltan Kochan
2026-06-23 01:30:30 +02:00
parent 7e1be2f8d2
commit 94df65b2bf

View File

@@ -79,7 +79,7 @@ jobs:
# server stays warm — i.e. pnpr's cold-client-vs-warm-server shape,
# the realistic CI case. Running both tools in one hyperfine
# invocation also yields the pnpr-vs-direct ratio per scenario.
BENCHMARK_TARGETS: ${{ github.ref_name == 'main' && 'pacquet@HEAD pnpr@HEAD' || 'pacquet@HEAD pacquet@main pnpr@HEAD pnpr@main' }}
BENCHMARK_TARGETS: 'pacquet@cdc897f838 pnpr@cdc897f838 pacquet@85d9e7708a pnpr@85d9e7708a pacquet@88add0909a pnpr@88add0909a'
# Network emulation so both install strategies cross the link they
# would in production instead of free loopback (~GB/s, ~0 RTT), which
# would hide every download cost.