feat(config): load imports from global and explicit config paths

[rpendleton]

Previously, the import directive was only honored during recursive project config discovery. Now the global config and explicitly provided --config paths also resolve their imports, and config-files surfaces global imports too.

[greptile-apps]

Greptile Summary

This PR extends the import directive to be honoured in three previously unsupported contexts: the global config (~/.config/fnox/config.toml), explicitly-provided --config paths, and config-files output. The refactor consolidates import loading into a new load_with_imports helper, replaces the two in-lined import loops in load_recursive with per-file calls to that helper, and adds a normalize_path utility (with unit tests) for consistent path deduplication in config-files.

Confidence Score: 5/5

Safe to merge; no correctness bugs found in the changed paths.

All P0/P1 findings from prior review rounds (non-recursive nested imports) have been acknowledged. The new code is logically correct for the advertised single-level import behaviour, the deduplication in config-files is consistent, and both integration and unit tests cover the new paths.

No files require special attention.

Important Files Changed

Filename	Overview
src/config.rs	Introduces load_with_imports helper; wires it into load_global, load_smart (explicit paths), and every file loaded inside load_recursive. The import-loading loops that were previously inlined in the root and non-root branches of load_recursive are removed cleanly. Logic and merge priority are preserved for the common single-file case.
src/commands/config_files.rs	Extracts import-printing into print_imports_from_partial and read_partial_config helpers; adds a normalize_path function with unit tests; wires global-config imports into run(). The refactor reduces duplication and is well-tested.
test/config_recursion.bats	Adds an integration test confirming that an explicit --config path resolves its import declarations correctly.
test/global_config.bats	Adds two integration tests: one confirming that global config imports are loaded by the engine, and one confirming that config-files surfaces the imported path.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[load_smart] -->|explicit path| B[load_with_imports]
    A -->|default filename| C[load_with_recursion]
    C --> D[load_recursive]
    D -->|per file in dir| B
    D -->|root = true| E[load_global]
    D -->|at FS root| E
    E --> B
    B --> F[load file]
    F --> G{import list?}
    G -->|yes| H[load_import for each]
    H --> I[merge: import_config < file_config]
    G -->|no| J[return config]
    I --> J

Loading

_{Reviews (3): Last reviewed commit: "fix(config): collapse /.. to / in normal..." | Re-trigger Greptile}

[gemini-code-assist]

Code Review

This pull request improves configuration loading by ensuring imports are processed for global and explicit config files. It refactors import printing in the config-files command and adds a normalize_path utility for path deduplication, supported by new integration tests. Feedback suggests making import loading recursive to handle nested imports and improving path normalization for root directory edge cases.

[gemini-code-assist]

The current implementation of load_with_imports only resolves one level of imports. If an imported file itself contains an import directive, those nested imports will not be loaded because load_import calls Self::load (which doesn't process imports) rather than Self::load_with_imports. While this matches the previous behavior in load_recursive, it limits the utility of imports in global and explicit config paths. Consider making this recursive, though you would need to handle potential circular imports (e.g., by passing a HashSet of visited paths).

[gemini-code-assist]

The normalize_path function provides lexical normalization which is useful for deduplicating paths in the HashSet. However, it doesn't handle the case where .. is applied to a root directory (e.g., /.. remains /.. instead of collapsing to /). While unlikely to affect config file discovery in practice, a more robust implementation or using a crate like path-clean (if dependencies are allowed) might be safer for general path handling.

[rpendleton]

Fixed in 7cb803b, though I wondered if using something like path-clean would be preferable. I didn't want to add a dependency without further discussion though.

[rpendleton]

One thing worth calling out on this line: since we're using load_import directly, imports aren't resolved transitively. If the imported file declares its own import, those won't be loaded. So, while this PR widens where import works (global config and explicit --config paths), it doesn't make it work everywhere.

I considered making it transitive, but that pulls in edge cases like cycle detection. If we do want to support that, a cleaner approach might be a shared config walker that does a partial parse and resolves imports recursively in one place, and then both config.rs and config_files.rs could rely on that.

For now, I figured I'd open this as a draft to discuss the direction before committing to anything.