Skip to content

ROADMAP.md

Latest commit

 

History

History
124 lines (107 loc) · 7.01 KB

ROADMAP.md

File metadata and controls

124 lines (107 loc) · 7.01 KB
Raw

Roadmap

Working agreement: every step is sized to be completed (and verified) in one sitting. An implementation only gets an expected-output.txt captured from an actual observed run — never written from reading the source. Anything that can't be run locally gets verified on the CI runner before merge.

Milestone 1 — CI backfill: every existing language tested on every PR

The harness: run-tests.sh loops over languages/*/*/test.sh, runs each, and diffs stdout against that folder's expected-output.txt. The GitHub Actions workflow (.github/workflows/test.yml) runs it on every PR and push to main. Folders without a test.sh are skipped until backfilled.

  • Step 1 — harness + locally verifiable languages. Workflow file, run-tests.sh, and test.sh/expected-output.txt for Python, JavaScript, TypeScript (Bun), Perl, Racket, and C. All six verified locally.
  • Step 2 — C++ and Go. C++ needs GCC 14 (<print> header) — not on the dev machine, so capture its output from the CI runner log on the PR, then commit it as the expected file. Go is preinstalled on runners.
  • Step 3 — Ruby, Lua, Haskell. test.sh added for all three (ruby lambda-core.rb, lua5.4 lambda-core.lua, runghc lambda-core.hs), the apt installs (ruby lua5.4 ghc) added to the workflow, and each folder's expected-output.txt captured from a real local run. All three PASS locally.
  • Step 4 — OCaml, Elixir, Clojure, F#. test.sh for all four (ocaml lambda_core.ml, elixir lambda-core.exs, dotnet fsi lambda-core.fsx, and cd lambda-core && clojure -M:test for Clojure's deps.edn test runner). Workflow installs ocaml+elixir via apt and the Clojure CLI via its official installer; F# uses the runner's preinstalled dotnet. OCaml + Elixir captured locally; F# and Clojure captured from the CI log (PR #31). Clojure uses clojure, not the interactive clj wrapper (which printed an rlwrap notice instead of running). All four green on CI.
  • Step 5 — Java and Kotlin. Java: PR #29 (the java branch) fixes the numeral typing so PRED works; add test.sh + expected output to that branch, let CI validate it, then merge #29. Kotlin: runner has a JDK; install kotlinc in the workflow.
  • Step 6 — exotic languages: ArkScript, FatScript, bruijn, Language 84. Each needs its own toolchain acquisition (GitHub releases, cargo/stack installs, or building from source). If one is genuinely unobtainable in CI, document that in its folder instead of leaving it silently untested.
  • Step 7 — flip the default and update the front door. Once all folders have tests: make run-tests.sh fail on folders missing a test.sh, update README contribution instructions to require test.sh + expected-output.txt in new-language PRs, and enable branch protection so PRs need a green check to merge.

Milestone 2 — fill out the missing major languages

Most additions land one language at a time with an implementation, README, test.sh, and observed expected output. The five-language batch below is one explicitly requested single-pass phase.

Phase 1 — 2025 popularity expansion

Selection method: filter the Stack Overflow 2025 "have used" language ranking against the repository's existing implementations, excluding HTML/CSS because it is markup. The five highest absent entries are SQL (58.6%), Bash/Shell (48.7%), PowerShell (23.2%), PHP (18.9%), and Rust (14.8%). GitHub's 2025 Octoverse independently places PHP and Shell in its top ten, while the August 2026 TIOBE index ranks SQL eighth, Rust tenth, and PHP thirteenth.

  • Step 1 — verify the selection and starting state. None of the five language folders existed before this phase; choosing them creates new implementations rather than fixing existing ones.
  • Step 2 — implement all five cores. Add Church booleans, Church numerals, native output adapters, and self-checking examples for Bash, PHP, PowerShell, Rust, and SQL. SQL may use an explicitly documented relational emulation because SQLite has no first-class functions.
  • Step 3 — add verification artifacts. Every folder gets README.md, test.sh, and expected-output.txt captured from an observed run.
  • Step 4 — integrate and verify. Verify the five toolchains in the current GitHub runner inventory (all are preinstalled), list the new implementations in the README, run focused checks, and run the repository harness. All five focused baseline checks pass; the full local harness reports 11 passed, 0 failed, 10 skipped for unavailable local toolchains, and the pre-existing Java baseline still pending.

Phase 2 — next five languages by 2025 usage

Selection method: continue down the Stack Overflow 2025 "have used" language ranking after removing repository implementations and C#, whose implementation already exists in active PR #37. The next five absent languages are Assembly (7.1%), Dart (5.9%), Swift (5.4%), R (4.9%), and Groovy (4.8%).

  • Step 1 — verify the selection and starting state. None of languages/a/assembly, languages/d/dart, languages/s/swift, languages/r/r, or languages/g/groovy exists on main. This phase creates five new implementations; it does not modify an existing language implementation.
  • Step 2 — implement all five cores. Add Church booleans, Church numerals, native output observers, and self-checking examples for Assembly, Dart, Swift, R, and Groovy. Assembly uses an explicitly documented x86-64 System V closure ABI because GNU assembler has no first-class closure type.
  • Step 3 — add verification artifacts. Every folder gets README.md, test.sh, and expected-output.txt captured from an observed run.
  • Step 4 — integrate and verify. Use the runner's preinstalled GCC, Swift, and Gradle toolchains; install only the R and Dart runtimes; list the new implementations in the README; run every focused check and the repository harness. GitHub CI reports 25 passed, 0 failed, 0 skipped, and the pre-existing Java baseline still pending (PR #39).

Later major-language phases

  • C#
  • Scala
  • Zig
  • Gleam (reopens the slot from closed PR #28)

Keep a "wanted" list in the README for languages left open to contributors (Prolog, Erlang, APL, Idris, …).

Milestone 3 — housekeeping

  • Remove the committed binary languages/c/c/app and add a root .gitignore for build artifacts.
  • TypeScript implementation only exercises Church numerals at runtime — booleans exist at the type level only. Decide whether that satisfies the spec or needs a runtime supplement.
  • Comment on merged PR #6 noting the Function<Term, Term> fix, closing the loop on the thread there.
  • Java README still describes an Oracle JDK manual install — point it at apt/Temurin and the test.sh instead.