[aceRage]

Cause

slic3rutils_tests --order rand crashed with SIGSEGV in hms_lookup_tests.cpp (seeds 1/2/3, at lines 326/326/203). HMS.cpp read the app config as wxApp::GetInstance() ? wxGetApp().app_config : nullptr. dark_mode_background_tests.cpp (Windows-only) starts a plain wxApp, and wxGetApp() static-casts it to GUI_App, so app_config came from past the end of a wxApp. Any HMS lookup that ran after that test (describe_error → hms_language_code) dereferenced garbage. In default order the HMS cases run first, so they passed.

Fix

• HMS.cpp: new gui_app_config_or_null() (dynamic_cast<GUI_App*>) replaces the three guarded reads (hms_language_code, query_print_error_url_action, init_hms_info).
• hms_lookup_tests.cpp: FixtureTables now restores data_dir() on teardown. It used to leave it pointing at a deleted temp dir.
• New case "a plain wxApp is not taken for the slicer's app" installs a plain wxApp and runs the lookups. Without the fix it crashes deterministically.

The unguarded wxGetApp() calls on the cloud-fetch paths are unchanged. Tests don't reach them (local_only), and in the app they only run under GUI_App.

Verification

• edgebuild 363 (Mac, --order rand): all tests passed. The Mac doesn't run the dark-mode test, so this run doesn't cover the crash.
• edgebuild 364–367 (Windows, --order rand --rng-seed 1/2/3/42): results to follow in a comment.

[aceRage]

Windows verification (edgebuild, slic3rutils_tests --order rand --rng-seed N): all four pass with 1319 cases. Jobs: 364 seed 1 (local), 365 seed 2 (vm), 366 seed 3 (vm), 367 seed 42 (vm). Each log shows the dark-mode test (the one that starts the plain wxApp) running before at least one HMS case. Seeds 1/2/3 are the ones that crashed on main.