$ ruby program.rb

Case

3 cases: try With YJIT and Spinel too

About this model. Based on CRuby 3.4 and 4.0 with default settings (Prism parser; YJIT off unless the case uses --yjit) on a 64-bit computer. The basic case's bytecode matches real Ruby 4.0.7 output. The Spinel case follows the pipeline in Spinel's README; its generated C is a simplified sketch. Addresses, sizes, register values and cache results are simplified examples; hardware figures are typical x86-64 values. Facts were checked against the sources listed at the bottom of the page (September 2026).

This page is a learning tool and simplifies many details, so parts of it may be wrong or out of date. Every component on the map is machine code, taking turns on the same CPU, except the machine code YJIT writes while your program runs.

Keyboard: ← and → move between steps.

1/9

Source

program.rb, on disk

Syntax tree

Bytecode

Inside the ruby program

activeunusedhover or tap a box for details

Hardware typical x86-64

CPU: waiting

Messages in this step

· Spotted a mistake? Ping @ervinismu on X or @ervinismu@ruby.social Vibe coded by ervinismu with Opus 5.5

Facts on this page were checked against these sources in September 2026. Where they disagree with the page, trust the sources.

CRuby and YARV bytecode

Ruby 3.4.0 Released ruby-lang.org: Prism as default parser, YJIT 3.4
Ruby 3.3.0 Released ruby-lang.org: Lrama, YJIT register allocation
Ruby 4.0.0 Released ruby-lang.org: ZJIT, Ractor changes
RubyVM::InstructionSequence Ruby 4.0 docs
iseq.c ruby/ruby: syntax tree freed after compiling
Advent of YARV: Part 0 Kevin Newton: instructions, frames, call data
Implement chilled strings ruby/ruby commit: putchilledstring
Frozen String Literals: Past, Present, Future? Jean Boussier
class Integer Ruby 3.4 docs: Integer#times written in Ruby

Prism parser

Prism C API documentation ruby/prism
Prism: Ruby 3.3's new error-tolerant parser Kevin Newton
Prism in 2024 Rails at Scale

GC and memory

Optimizing Ruby's Memory Layout: Variable Width Allocation Shopify Engineering: 40-byte slots, size pools
Garbage Collection in Ruby Peter Zhu: tagged values, GC phases
Finding Memory Leaks in the Ruby Ecosystem Rails at Scale: no free-at-exit, RUBY_FREE_AT_EXIT

YJIT and ZJIT

YJIT: Yet Another Ruby JIT Ruby 4.0 docs: options, call threshold, memory
YJIT 3.4: Even Faster and More Memory-Efficient Rails at Scale
Evaluating YJIT's Performance in a Production Context MPLR 2023 paper
ZJIT Ruby 4.0 docs
The inliner is yielding benefits for ZJIT Rails at Scale: real Ruby 4.x bytecode dumps

Spinel

matz/spinel README pipeline, runtime, limitations, benchmark table used in the speed race (fetched 26 Sep 2026; changes often)
Spinel issue #4968 generated sp_int_add, RangeError on overflow
Spinel issue #4748 times {} emitted as an inline loop

🏁 Speed race: same benchmark, three ways

Times from the benchmark table in Spinel's README on the master branch, fetched 26 September 2026: CRuby 4.0.4 run with --disable-gems, each time the mean of five runs (perf stat -r 5). The race is slowed down so you can watch it.

Interpreterruby

0 ms

YJITruby --yjit

0 ms

Spinel./program

0 ms

Real speedups depend on the program and the machine. The README is updated often, so its current table may differ: see the Spinel README benchmarks.