CI: make Common Lisp core tests reproducible from a clean checkout #9

New issue

Closed

opened 2026-08-16 05:08:50 +00:00 by lost-rob0t · 0 comments

lost-rob0t commented

2026-08-16 05:08:50 +00:00

(Migrated from github.com)

Problem

Hackmode now has meaningful ASDF regression tests, but the repository does not yet declare/fetch its complete Common Lisp dependency graph from a clean CI checkout. The current monorepo workflow can enforce source/layout contracts and Emacs syntax, but it cannot honestly claim asdf:test-system :hackmode passes in an isolated runner.

Current behavior

source/hackmode-core/hackmode-tests.asd exercises operation registry reopen and asset normalize/dedupe/persist/event behavior.
Hackmode depends on external systems including Tek9 and the canonical starintel Common Lisp schema runtime, plus existing Hackmode dependencies.
the root flake currently provides SBCL/system tools but does not pin the full Lisp source graph.

Desired behavior

A clean GitHub Actions runner can execute the Hackmode ASDF tests without relying on the developer's pre-populated Quicklisp/local-project tree.

Architecture

Prefer one reproducible dependency path shared by local development and CI. Do not vendor StarIntel Server, Quasar, or Tek9 into the Hackmode monorepo merely to make CI work. Pin/source reusable libraries as dependencies (Nix/Quicklisp/Qlot/ASDF source registry as appropriate).

Implementation scope

inventory every direct/transitive non-Quicklisp dependency required by :hackmode
choose the least-surprising reproducible dependency mechanism compatible with the existing Nix/ASDF setup
pin starintel and tek9 as external library dependencies
run asdf:test-system :hackmode in CI
preserve the lightweight monorepo hygiene job

Dependencies

Depends on #4. Can proceed in parallel with #6, but #6 must not be declared fully verified until these runtime tests execute on a clean runner.

Acceptance criteria

clean runner loads :hackmode
clean runner executes asdf:test-system :hackmode
operation registry reopen test passes
asset normalize -> dedupe -> persist -> event test passes
no developer-local Quicklisp state is assumed
reusable external libraries remain separate repositories

Verification

Delete/ignore all user-local ASDF/Quicklisp state and reproduce the exact CI command from a clean checkout.

## Problem Hackmode now has meaningful ASDF regression tests, but the repository does not yet declare/fetch its complete Common Lisp dependency graph from a clean CI checkout. The current monorepo workflow can enforce source/layout contracts and Emacs syntax, but it cannot honestly claim `asdf:test-system :hackmode` passes in an isolated runner. ## Current behavior - `source/hackmode-core/hackmode-tests.asd` exercises operation registry reopen and asset normalize/dedupe/persist/event behavior. - Hackmode depends on external systems including Tek9 and the canonical `starintel` Common Lisp schema runtime, plus existing Hackmode dependencies. - the root flake currently provides SBCL/system tools but does not pin the full Lisp source graph. ## Desired behavior A clean GitHub Actions runner can execute the Hackmode ASDF tests without relying on the developer's pre-populated Quicklisp/local-project tree. ## Architecture Prefer one reproducible dependency path shared by local development and CI. Do not vendor StarIntel Server, Quasar, or Tek9 into the Hackmode monorepo merely to make CI work. Pin/source reusable libraries as dependencies (Nix/Quicklisp/Qlot/ASDF source registry as appropriate). ## Implementation scope - inventory every direct/transitive non-Quicklisp dependency required by `:hackmode` - choose the least-surprising reproducible dependency mechanism compatible with the existing Nix/ASDF setup - pin `starintel` and `tek9` as external library dependencies - run `asdf:test-system :hackmode` in CI - preserve the lightweight monorepo hygiene job ## Dependencies Depends on #4. Can proceed in parallel with #6, but #6 must not be declared fully verified until these runtime tests execute on a clean runner. ## Acceptance criteria - clean runner loads `:hackmode` - clean runner executes `asdf:test-system :hackmode` - operation registry reopen test passes - asset normalize -> dedupe -> persist -> event test passes - no developer-local Quicklisp state is assumed - reusable external libraries remain separate repositories ## Verification Delete/ignore all user-local ASDF/Quicklisp state and reproduce the exact CI command from a clean checkout.