Digital Cabin - programing https://cabin.digital/ 2025-06-17T23:00:00Z golog Comfy cabin in a rough digital ocean dweller https://cabin.digital/tags/programing.xml:::https://cabin.digital/log/small_elfs.html Small ELFs 2025-06-17T23:00:00Z dweller In a previous log I mentioned that I was trying to make small ELFs. Well that also got me thinking how big is an average ELF. So I spent too much time on writing some sh(1) and awk(1) scripts, and this is the result:

lang size rundeps ↓total type arch linkage debug? strip? buildT runT/100
nasm-elf32-s 134 0 134 ELF32l Intel 80386 static no yes 6.05ms 134.12µs
nasm-elf32 141 0 141 ELF32l Intel 80386 static no yes 7.99ms 165.81µs
nasm-elf-s 165 0 165 ELF64l x86-64 static no yes 14.65ms 131.77µs
nasm-elf 173 0 173 ELF64l x86-64 static no yes 7.66ms 150.67µs
jmpld 56 308 364 bin jmpld - - - 19.09ms 144.06µs
bld 54 493 547 bin bld - - - 9.28ms 233.59µs
nasm 8.30KB 0 8.30KB ELF64l x86-64 static no yes 14.62ms 139.83µs
c-gcc-nlibc 8.72KB 0 8.72KB ELF64l x86-64 static no yes 50.08ms 180.19µs
c-musl-st 13.45KB 0 13.45KB ELF64l x86-64 static no yes 41.59ms 190.95µs
jmpld-c 56 17.45KB 17.50KB bin jmpld - - - 83.34ms 164.79µs
c-gcc-st 648.45KB 0 648.45KB ELF64l x86-64 static no yes 109.54ms 304.16µs
c-clang-st 656.55KB 0 656.55KB ELF64l x86-64 static no yes 141.96ms 301.99µs
oksh-musl-st 42 1010.13KB 1010.17KB script oksh - - - 4.07ms 571.29µs
go-s 1.36MB 0 1.36MB ELF64l x86-64 static no yes 206.59ms 1.08ms
c++gcc-st 2.10MB 0 2.10MB ELF64l x86-64 static no yes 707.85ms 560.04µs
go 2.10MB 0 2.10MB ELF64l x86-64 static yes no 257.03ms 1.07ms
c++clang-st 2.14MB 0 2.14MB ELF64l x86-64 static no yes 722.34ms 563.63µs
rust-nstd-s 14.05KB 13.45MB 13.46MB ELF64l-PIE x86-64 dynamic no yes 163.81ms 598.39µs
c-gcc-s-lto 14.14KB 13.45MB 13.47MB ELF64l-PIE x86-64 dynamic no yes 104.36ms 479.64µs
c-gcc-s 14.14KB 13.45MB 13.47MB ELF64l-PIE x86-64 dynamic no yes 66.75ms 472.23µs
c-clang-s 14.16KB 13.45MB 13.47MB ELF64l-PIE x86-64 dynamic no yes 95.86ms 458.92µs
rust-nstd 19.70KB 13.45MB 13.47MB ELF64l-PIE x86-64 dynamic yes no 82.25ms 457.37µs
c-gcc 19.88KB 13.45MB 13.47MB ELF64l-PIE x86-64 dynamic yes no 63.22ms 621.80µs
c-clang 19.89KB 13.45MB 13.47MB ELF64l-PIE x86-64 dynamic yes no 90.55ms 554.57µs
c-gcc-g 48.71KB 13.45MB 13.50MB ELF64l-PIE x86-64 dynamic yes no 77.53ms 511.37µs
sh-unix 23 13.57MB 13.57MB script dash - - - 3.11ms 729.54µs
sh 34 13.57MB 13.57MB script dash - - - 2.71ms 721.37µs
rust-s 312.67KB 13.63MB 13.93MB ELF64l-PIE x86-64 dynamic no yes 2.97s 658.44µs
rust-s-nlto 360.77KB 13.63MB 13.98MB ELF64l-PIE x86-64 dynamic no yes 160.25ms 753.09µs
rust 436.70KB 13.63MB 14.05MB ELF64l-PIE x86-64 dynamic yes no 139.50ms 797.12µs
c++gcc-s 14.15KB 20.21MB 20.22MB ELF64l-PIE x86-64 dynamic no yes 549.65ms 1.22ms
c++clang-s 14.19KB 20.21MB 20.22MB ELF64l-PIE x86-64 dynamic no yes 612.06ms 1.15ms
c++clang 20.21KB 20.21MB 20.23MB ELF64l-PIE x86-64 dynamic yes no 558.83ms 1.14ms
c++gcc 20.40KB 20.21MB 20.23MB ELF64l-PIE x86-64 dynamic yes no 529.38ms 1.22ms
c++clang-g 35.40KB 20.21MB 20.24MB ELF64l-PIE x86-64 dynamic yes no 629.34ms 1.41ms
c++gcc-g 148.93KB 20.21MB 20.35MB ELF64l-PIE x86-64 dynamic yes no 622.06ms 1.17ms
tail 32 20.55MB 20.55MB script tail - - - 3.19ms 724.24µs
bash 40 21.89MB 21.89MB script bash - - - 2.16ms 2.01ms
zsh 39 25.42MB 25.42MB script zsh - - - 3.19ms 2.45ms
perl 42 27.96MB 27.96MB script perl - - - 2.65ms 1.71ms
ruby 40 29.07MB 29.07MB script ruby - - - 5.30ms 57.78ms
py3 45 29.17MB 29.17MB script python3.13 - - - 4.94ms 14.39ms
nodejs 49 177.31MB 177.31MB script node - - - 7.88ms 26.03ms
c-musl-s - - - FAIL - - - - - -
c-musl - - - FAIL - - - - - -
rust-nstd-st - - - FAIL - - - - - -
rust-st - - - FAIL - - - - - -
rust-tiny - - - FAIL - - - - - -

Explanation of naming:

  • -g = debug;
  • -s = small (stripped + something like -Os if available and I knew of it)
  • -st = static build
  • -noX = without X, like -nolibc = without any libc

Some things failed because:

  1. I only had static musl built, so dynamic failed;
  2. I didn’t know how to coax some rust builds since I don’t use it.

A big caveat regarding the “runtime”. The “runtime” benchmarks are more of a benchmark of shell parsing, execve(2) and fork(2)ing and runtime start up of the program, since the program doesn’t really do anything. Still it does show how slow the startup is, I guess it matters for shell scripts/tools.

You can check out the source for this all at my git repo.

IDK if it has any use except to, maybe, show that static binaries are not as bloated as people often think. (Because any decent lib will be split into many .o and only used code will be linked. + LTO) Which is just another reason to use them over the death of us all that is dynamic linking.

]]>