Lean Kernel Arena / perf/proj-applied-struct

Test "perf/proj-applied-struct"

Expected: 👍 accept · Size: 110.6 KB · Lines: 2.1 k · lean4export: 3.1.0 · Lean: 4.36.0 · Timeout: 3.83 min · 📄 Declaration · 🔗 Source

The declarations to check compare

(gC slowA c).fn x      and      (gC' slowA c).fn x

(and two variants), where gC' s c unfolds to gC s c, and gC s c is if s = c then ⟨…, 0⟩ else ⟨…, 1⟩ for a structure whose field fn is a function. slowA is a closed Nat that evaluates in 2^19 unit steps, and c and x are variables.

Unfolding gC' makes the sides identical, for Θ(1). Reducing the structure gC slowA c to a constructor instead decides slowA = c: it evaluates slowA, for Θ(2^19), and then is stuck on c anyway. Unlike in proj-cheap-struct, the projections are not compared as such: each is the head of an application. Official reaches whnf_core with full reduction of projected structures before it compares the applications by their arguments.

Official (v4.35.0) needs about 19 s and 1.4 GB here, nearly all of it evaluating slowA, slowB and slowC. Distilled from AxiomMath/CFSG's CertificateChecks.KernelSporSch.RuE64.invol_eq, where the same shape takes it over 24 GB and more than 1.5 hours.

Checker Result ⏱️ 🧠
nano-lean 👍 1.31 ms (÷10704) 40.5 MB (÷36)
gruszka 👍 2.76 ms (÷5077) 50.0 MB (÷29)
sokonanoda 👍 1.99 ms (÷7046) 47.6 MB (÷30)
rpylean 👍 2.01 ms (÷6969) 12.1 MB (÷120)
pestka 👍 4.04 ms (÷3470) 17.8 MB (÷82)
kefir 👍 5.12 ms (÷2739) 6.4 MB (÷226)
palladium 💥 1.84 s 775.6 MB
con-ron 👍 10.3 s (-26%) 2.7 GB (+88%)
kefir-wasm 👍 214 ms (÷66) 105.7 MB (÷14)
con-agua 👍 12.0 s (-14%) 3.4 GB (×2.4)
nanobruijn 👍 10.5 s (-25%) 3.2 GB (×2.2)
nanoda 👍 12.8 s (-9%) 2.6 GB (+84%)
con-leche 👍 14.6 s (+4%) 3.2 GB (×2.2)
pestka-wasm 👍 133 ms (÷105) 120.9 MB (÷12)
tenet-jvm 👍 14.8 s (+6%) 3.7 GB (×2.6)
official-nightly 👍 11.2 s (-20%) 1.4 GB (0%)
eink0rn 👍 5.77 ms (÷2430) 19.5 MB (÷75)
official 👍 14.0 s 1.4 GB
ind-models 👍 14.0 s (0%) 1.4 GB (+1%)
tenet 👍 60.9 s (×4.3) 5.6 GB (×3.9)
tenet-hs 👍 23.4 s (+67%) 829.7 MB (-43%)
kiota 👍 6.07 ms (÷2310) 11.1 MB (÷131)
ocaml4lean 👍 4.77 min (×20) 5.1 GB (×3.6)
logos-kernel 🚫 295 ms 318.1 MB
evmlean 🚫 67.8 s 434.7 MB
mini 🚫 40.4 ms 65.3 MB
overfull ⌛ — —
lean4cobol 🚫 9.07 s 393.4 MB
lean4lean ✋ 800 ms 235.6 MB
canonical-min 🚫 393 ms 1.0 GB
nanoclo 👍 2.83 ms (÷4954) 63.6 MB (÷23)
vow-lean-kernel 👍 681 ms (÷21) 18.1 MB (÷80)
lazylean 👍 3.93 s (÷3.6) 326.2 MB (÷4.5)
metalean-slow 🚫 2.44 s 68.7 MB
metalean-fast 👍 16.2 s (+16%) 3.7 GB (×2.6)
nanoclo-fortran 👍 6.95 ms (÷2018) 5.5 MB (÷264)
mathgraph 👍 1.91 ms (÷7332) 69.3 MB (÷21)