Use dmpart for partition tables instead of a private parser

[JohnAmadis]

Part of choco-technologies/dmod-ecosystem#23 / #17.

Needs dmpart from the registry. That is now satisfied: dmblkid v0.1 is released and includes dmpart.

Summary

The MBR/GPT parser existed twice: src/partitions.c here, for partition nodes, and a copy in dmblkid, for telling a partitioned disk from a filesystem. It now lives in one place, dmpart, a small library module in the dmblkid repository (libs/dmpart, released with dmblkid):

int dmpart_scan(const dmpart_medium_t* medium, dmpart_found_t found,
                void* found_ctx, dmpart_info_t* info);

• src/partitions.c/.h removed (−443 lines).
• find_partitions() describes the medium as a dmpart_medium_t (driver read callback, block size, block count) and calls dmpart_scan(). The callbacks are unchanged static functions.
• dmod_link_modules(... dmpart ...): dmdevfs.dmd now requires dmpart@>=0.1<1.0, so every package or image that contains dmdevfs must also contain dmpart.
• Behaviour differences:
  • running out of memory while reading a table is now logged, where before it silently looked like "no table";
  • an exFAT boot sector is no longer taken for an MBR (FAT already wasn't).
• README: the Partition Nodes section points to dmpart. dmblkid uses the same parser, so a medium that gets partition nodes here is always reported by dmblkid as a partition table, never as a filesystem to mount whole.

Tests: install dmdevfs's own dependencies (3ce65ac)

The test modules load dmdevfs by name (Dmod_LoadModuleByName("dmdevfs")) instead of depending on it. ctest therefore only installed the test module's own .dmd. Until now, dmdevfs's dependencies happened to be covered by the tests' own dependencies (libsystemd pulls in dmini), but dmpart is not.

With dmpart missing, dmod_loader correctly refused to load dmdevfs (Cannot load required module 'dmpart'). load_devfs() returned false, and the test steps, which do not check for that, then called the unset DIF pointers (g_fs.fopen is NULL). That NULL call caused the segfault in CI.

Correction: the message of 3ce65ac says dmdevfs was loaded without dmpart and dmpart_scan() jumped to 0. That is wrong. The loader never started dmdevfs; the NULL call is in the test code.

The fix installs dmdevfs.dmd before each test.

Test plan

• In chocotechnologies/dmod:1.0.4, with dmpart 0.1 from the registry (no local packages):
  • ctest: 8/8;
  • test_partitions 12/12, running on dmpart;
  • test_dmdevfs and test_dmdevmon still pass.
• Without 3ce65ac, reproduced the CI failure: Cannot load required module 'dmpart', then a segfault in the test.
• Cortex-M7 build: no new warnings. The only warning (read_dir_name_from_path defined but not used) is already on main.
• dmpart's own tests (test_dmpart, 15 steps, in dmblkid) cover the parser directly on an in-memory medium.

🤖 Generated with Claude Code

https://claude.ai/code/session_01TgkwHQKqf6c4DzypSKzZT7