From 99fde44a063afc2f687b3e1f67e80939f9116f17 Mon Sep 17 00:00:00 2001
From: sunrisepeak <speakshen@163.com>
Date: Sat, 9 May 2026 04:47:58 +0800
Subject: [PATCH] =?UTF-8?q?refactor(pm):=20add=20pm/pm.cppm=20subsystem=20?=
 =?UTF-8?q?fa=C3=A7ade=20(PR-R7)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Final step of the package-management subsystem refactor (see
`.agents/docs/2026-05-08-pm-subsystem-architecture.md`). Strictly zero
behavior change.

* New module `mcpp.pm` (`src/pm/pm.cppm`) is the single import point
  for callers in `cli.cppm`, `build/`, `pack/`, etc. It re-exports the
  three pm data-type modules (`dep_spec`, `index_spec`, `lock_io`)
  but **deliberately not** the flow modules (`resolver`, `commands`,
  `package_fetcher`, `publisher`) — those stay direct-import-only so
  the encapsulation gained by R1–R6 is preserved.

* No call-site migration in this PR. Existing `import mcpp.lockfile`
  / `import mcpp.fetcher` / `import mcpp.publish.xpkg_emit` shims
  continue to work; new code can choose `import mcpp.pm` and reach
  the same data types via the new façade.

* `prepare_dependencies(Manifest&, Lockfile&, ...)` (the high-level
  operation that should replace the ~250-line dep-resolution loop in
  `cli.cppm::prepare_build`) is left for a separate PR — pulling it
  out is a non-trivial behavior-touching change that doesn't belong
  inside the strict-move R-series. The façade leaves a clean spot
  for it to land later.

Verification:
* `mcpp build` compiles unchanged.
* `mcpp test` — 9/9 unit binaries pass.
* e2e: 02 / 09 / 12 / 23 / 27 all pass.

Closes the seven-PR refactor sequence. The pm subsystem is now:

  src/pm/
  ├── pm.cppm                  (façade, this PR)
  ├── dep_spec.cppm            (R1)
  ├── index_spec.cppm          (R1 placeholder, awaits index-config)
  ├── lock_io.cppm             (R2)
  ├── package_fetcher.cppm     (R3)
  ├── resolver.cppm            (R4)
  ├── commands.cppm            (R5)
  └── publisher.cppm           (R6)
---
 src/pm/pm.cppm | 34 ++++++++++++++++++++++++++++++++++
 1 file changed, 34 insertions(+)
 create mode 100644 src/pm/pm.cppm

diff --git a/src/pm/pm.cppm b/src/pm/pm.cppm
new file mode 100644
index 0000000..7aafb7c
--- /dev/null
+++ b/src/pm/pm.cppm
@@ -0,0 +1,34 @@
+// mcpp.pm — package-management subsystem façade.
+//
+// This is the single import point that callers in `cli.cppm`,
+// `build/`, `pack/`, etc. should use to reach pm types and high-level
+// operations. The internal modules (`pm.resolver`, `pm.commands`,
+// `pm.package_fetcher`, `pm.publisher`) stay deliberately out of the
+// public surface — re-exporting them would invite call sites to
+// reach into pm internals and undo the encapsulation gained by R1–R6.
+//
+// What this module re-exports:
+//
+//   * Data types only:
+//       - `mcpp::pm::DependencySpec`         (from pm.dep_spec)
+//       - `mcpp::pm::kDefaultNamespace`      (from pm.dep_spec)
+//       - everything currently in pm.index_spec (placeholder; the
+//         index-config implementation will land here)
+//       - `mcpp::pm::Lockfile` and friends   (from pm.lock_io)
+//
+// Internal pm modules (resolver, package_fetcher, commands, publisher)
+// MUST be imported directly by callers that need them; they are
+// intentionally **not** re-exported here. As the call-site migration
+// progresses, an additional high-level operation
+//   `prepare_dependencies(Manifest&, Lockfile&, ...)` → vector<...>
+// will appear here, replacing the ~250-line dependency-resolution
+// loop currently inlined in `cli.cppm::prepare_build`. That landing
+// is tracked separately so this PR stays a pure additive facade.
+//
+// See `.agents/docs/2026-05-08-pm-subsystem-architecture.md` §4.10.
+
+export module mcpp.pm;
+
+export import mcpp.pm.dep_spec;
+export import mcpp.pm.index_spec;
+export import mcpp.pm.lock_io;