Skip to content

ROADMAP.md

Latest commit

 

History

History
55 lines (48 loc) · 2.97 KB

ROADMAP.md

File metadata and controls

55 lines (48 loc) · 2.97 KB
Raw

libAPG Roadmap

v2.0 — Stable public API

  • ABI stabilization and documentation
  • C++ bindings (header-only, bindings/cpp/)
  • Python bindings
  • Correctness tests: installation unit tests, fuzzing (package JSON, dep constraints)
  • pkg-config and CMake find module
  • Drop the gpgme (OpenPGP) signing backend entirely; libsodium (Ed25519) becomes the only one

v2.2.0 — SAT-based dependency solver

Additive only: existing dep_graph_resolve() / dep_graph_resolve_parallel() keep their current behavior and ABI. This ships as a new, separate resolve entry point; it does not touch .apg, struct package, or struct package_metadata.

  • dep_graph today maps one name to exactly one node (node_map, alias_map); it cannot represent multiple candidate versions/providers for the same name. Needs a parallel candidate-set representation usable by the solver without disturbing the existing single-candidate graph used by the current resolvers.
  • Constraint model: turn each package's dependencies (name + version op) and conflicts into SAT clauses over candidate selection variables.
  • Provider/alias handling: a provides/replaces alias can be satisfied by any of several candidates — needs "at least one of" clauses instead of today's single deterministic pick. Falls out of the unified candidate_set model (a provides name and a real package name share the same lookup) — no separate handling needed, verified with a multi-provider dependency clause. replaces is not covered yet.
  • Solver core: pick and implement an actual SAT (or CDCL-lite/PubGrub-style incremental) algorithm; decide on backtracking strategy and a search budget/timeout for pathological inputs. Shipped as a plain DPLL solver (unit propagation + backtracking, first-unassigned-variable decision order, decision-count budget) in src/graph/sat_solve.c. No clause learning (not CDCL) — a reasonable first cut, not the final version. Note: sat_model_build() alone doesn't force any package to be installed — the caller must add a unit clause for each root package it actually wants, same as the [app forced] case used to verify this.
  • Conflict reporting: on UNSAT, produce a human-readable explanation (which constraints clashed), not just an error code.
  • New public API surface (new header declarations, new exported symbols) — additive, APG_API, no changes to existing signatures.
  • Threading: decide whether/how this interacts with the existing pthread-parallel resolve path, or whether the solver stays single-threaded initially.
  • Tests: unit tests for satisfiable/unsatisfiable cases, fuzzing over randomly generated constraint sets, and a perf benchmark against the current resolver on typical (non-pathological) dependency sets.

Maybe in the future

  • Atomic installation: all-or-nothing semantics