Fix Windows shell and path handling

[ferologics]

Summary

• route !command config values through pi's configured shell abstraction instead of the platform default shell
• normalize Windows paths to slash-style for matching, completion, display, and tool output while keeping native paths for filesystem access
• harden package-manager git commands against interactive credential prompts on Windows
• update platform-assumptive tests and add Windows regressions for config-command execution and autocomplete path handling

Fixes #1775.
Also addresses #1769.
Related prior attempt: #1902.

Validation

• ./test.sh
• npm run check

Out of scope / follow-ups

• I did not add a windows-latest CI job in this PR. The goal here was to get the Windows test suite green locally first; CI can be added in a follow-up once this behavior lands cleanly.
• There is still room to consolidate some slash-path normalization helpers across packages, but I left that out to keep this PR focused on fixing Windows shell/path behavior and regressions.

[yhslai]

Hello, there are still some tests failed for me.

I'm at commit 948fde2 and running ./test.sh under Git bash.

⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯ Failed Tests 6 ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯

 FAIL  test/tree-selector.test.ts > TreeSelectorComponent > branch navigation and folding with ctrl+arrow keys > ctrl+right unfolds a folded node, then does segment jump when unfolded
AssertionError: expected 'asst-4a' to be 'user-3a' // Object.is equality

Expected: "user-3a"
Received: "asst-4a"

 ❯ test/tree-selector.test.ts:367:45
    365|
    366|    selector.handleInput(CTRL_LEFT); // asst-4a → user-3a
    367|    expect(list.getSelectedNode()?.entry.id).toBe("user-3a");
       |                                             ^
    368|
    369|    selector.handleInput(CTRL_LEFT); // fold user-3a

Full log is attached:

test.log

[ferologics]

Thanks — I tracked this down.

The remaining failures were not from the Windows shell/path fix itself, but from selector tests (tree-selector / related keybinding-driven tests) depending on the global editor keybindings singleton. If someone has custom local keybindings configured, those raw Ctrl/Alt arrow escape sequences can stop mapping to the expected actions and the tests fail locally.

I fixed that by resetting those tests to DEFAULT_EDITOR_KEYBINDINGS, so they no longer depend on the runner's local hotkey setup.

I also cleaned up the branch history and force-pushed it so this PR no longer includes the unrelated --session path changes.

Revalidated with:

• ./test.sh
• npm run check

[yhslai]

Sorry, my bad.

After running npm install and npm run build the tests all pass.

ℹ tests 442
ℹ suites 92
ℹ pass 442
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 986.3826

I also manually tested the path slash bug in #1769 and this PR fixes it for me too.

[badlogic]

sorray, i did this myself :/

[jamwil]

When did it become so trendy to be so rude.

[yhslai]

@badlogic @ferologics Hello, I've noticed that this PR is closed because the main issue is fixed on main already. However there is a minor thing missing: the custom keybinding causing tests to fail. This PR seemed to fix that but the latest main doesn't. Could you consider incorporating that fix into main too?

[yhslai]

A simple keybindings.json to make one test fail:

{
    "submit": ["shift+enter"],
    "newLine": ["enter"]
}

Relevant log:

 FAIL  test/session-selector-rename.test.ts > session selector rename > enters rename mode on Ctrl+R and submits with Enter
AssertionError: expected "spy" to be called 1 times, but got 0 times
 ❯ test/session-selector-rename.test.ts:98:25
     96|   await flushPromises();
     97|
     98|   expect(renameSession).toHaveBeenCalledTimes(1);
       |                         ^
     99|   expect(renameSession).toHaveBeenCalledWith(sessions[0]!.path, "XOld");
    100|  });

[badlogic]

When did it become so trendy to be so rude.

@jamwil i truly wonder what about "sorry, i implemented this myself" is rude? have you also considered that there may be out-of-channel coms between me and fero (a friend of mine)?

[jamwil]

I hadn’t considered that, no. From the context in this thread alone it seemed like someone put a fair amount of effort into a PR only to have it closed unceremoniously and with a comment that would be hard to interpret as anything but rude, were it not for the fact that you know each other and communicated via other means, I suppose.