Phase 24. TargetIRubyKernel

Field	Value
MEP	MEP-56 §Phases
Status	LANDED
Started	2026-05-29 12:01 (GMT+7)
Landed	2026-05-29 12:01 (GMT+7)
Tracking issue	—
Tracking PR	#22510
Commit	72522b6311

Gate

TestPhase24TargetIRubyKernel in transpiler3/ruby/build/phase24_test.go: builds hello_nb.mochi (print("hi from notebook")\nprint(2 + 2)) via Driver.Build with TargetIRubyKernel, reads the resulting hello_nb.ipynb, asserts json.Unmarshal succeeds (the file is valid JSON), asserts nbformat == 4, asserts metadata.kernelspec.name == "ruby", kernelspec.language == "ruby", and language_info.name == "ruby", asserts there is exactly one cell with cell_type == "code", asserts the joined cell source contains both hi from notebook and frozen_string_literal, and asserts every source line except the last ends in a \n (the nbformat 4 source-array convention).

Build target / audit decisions

buildIRubyNotebook in transpiler3/ruby/build/build.go (lines 296 to 335) marshals a map[string]any with the nbformat 4 schema: top-level nbformat: 4, nbformat_minor: 5, a metadata object pinning the kernel to name: "ruby" / display_name: "Ruby (IRuby)" / language: "ruby" with language_info.name: "ruby" and file_extension: ".rb", and a single code cell with execution_count: nil, empty metadata, empty outputs, and source populated from splitKeepNewline(sf.RubySource()).

splitKeepNewline (lines 498 to 507) splits the source on \n but keeps the trailing newline on each non-final line, matching the Jupyter convention where source is an array of strings and each except the last ends in "\n". The convention exists so notebook merge tools can diff line by line. Concatenating the array with the empty string reconstructs the original file byte for byte.

json.MarshalIndent(nb, "", " ") (single-space indent) keeps the file compact while staying human-readable; nbformat does not require pretty printing, but most notebook tooling assumes it. The kernel is pinned to ruby (not iruby) because the IRuby project registers itself under the bare ruby kernel name via iruby register, matching what jupyter kernelspec list reports.

The build emits a single code cell, not a notebook split into per-statement cells. The motivation: every Mochi source has exactly one main module and one entry point, so a multi-cell split would have to decide arbitrarily where to break. A single cell preserves whole-file semantics and lets the user manually split it after opening the notebook if they want a presentation layout.

Files changed

File	Purpose
transpiler3/ruby/build/build.go	buildIRubyNotebook writes <name>.ipynb with nbformat 4 + IRuby kernelspec (lines 296 to 335); splitKeepNewline helper (lines 498 to 507); Driver.Build dispatches TargetIRubyKernel (line 209)
transpiler3/ruby/build/phase24_test.go	TestPhase24TargetIRubyKernel JSON unmarshal + nbformat / kernelspec / cell-shape assertions

Test set

• TestPhase24TargetIRubyKernel (single test, no subtests): asserts the file is valid JSON, nbformat == 4, kernelspec / language_info pin to ruby, one code cell, source contains the print literal and frozen_string_literal, and the line-ending convention holds.

Closeout notes

Phase 24 validates the notebook by JSON-decoding into a typed struct rather than by substring match. This catches malformed JSON (trailing commas, unquoted keys) that a substring check would miss, and the typed struct surfaces schema drift at unmarshal time. The trailing-newline assertion at the end of the test is what proved most useful in practice: an earlier draft used strings.Split (no trailing \n on the split parts) and the notebook still validated but JSON.parse round-trips showed concatenation gaps. Switching to strings.SplitAfter and trimming a possibly-empty final element produces the exact array shape Jupyter expects.