Review fixes for the Ogre 14.6 port

[Upabjojr]

Follow-up to tomluchowski#15, based on ff63881. This targets ogre-v14.6-v3 directly so it stacks onto that PR rather than duplicating it — the review it addresses is here.

Three commits, roughly in order of how much they matter.

Correctness

• ODSocketServer::mActualPort was never initialised. Not in the constructor's init list, only assigned inside createServer() — so ODServer::getNetworkPort(), which now calls getActualPort() unconditionally, read an indeterminate value and could return a garbage port instead of falling through to the configured one.
• ODFrameListener::renderQueueStarted() dereferenced a pointer that starts null. CameraManager::mActiveCamera is nullptr until setActiveCamera() runs; the new condition called ->getName() on it regardless.
• ReflMetal.material was deleted while three meshes still reference it — Roundshield.mesh, Sabre.mesh, Wyvern.mesh — and shaders/ReflMetal.{vert,frag} and materials/textures/EnvmapMetal.png were left orphaned. Restored the script; removing it properly means retargeting those meshes too, which is a separate decision.
• Seven converted normal-map materials had their braces reflowed so normal_map } closed the rtshader_system block on the same line and pass lighting lost its own closing line. Balanced, so it parses, but the nesting no longer reads correctly. Restored the structure the other ~50 materials use, and made the mapping space explicit as tangent_space — matching the implicit default of the pre-conversion lighting_stage normal_map <tex> line.

Debug leftovers

• dumpWindowTree() walked the whole widget tree and wrote a line per widget to std::cerr — not through LogManager — every time the settings window opened.
• MenuModeMain::activate() logged root window geometry on every entry to the main menu.

Mechanical cleanup

• Moved the new <iostream>/<fstream> includes below each file's own header, and dropped the two that duplicated an include already a few lines further down (ODPacket.cpp, ConfigManager.cpp are now byte-identical to before the PR).
• Restored indentation on the five statements that ended up in column 0 beside the new buildTangentVectorsIfNeeded() calls; took the MeshPtr by const reference instead of copying the shared pointer, and added a null guard since several call sites feed getByName() straight in.
• Dropped the redundant .get() calls and the three #if OGRE_VERSION < 0x10A00 / #else blocks whose branches had become identical — matching how MovableTextOverlay.cpp handled the same situation in Ogre v14.6 v3 tomluchowski/OpenDungeonsPlus#15.
• CMake: set(CMAKE_CXX_STANDARD 14) was competing with the OD_CXX11_FLAGS probe still prepending -std=c++11 to CMAKE_CXX_FLAGS. Removed the probe and added CMAKE_CXX_STANDARD_REQUIRED ON. CMAKE_CXX_EXTENSIONS defaults to ON, giving -std=gnu++14, which covers the MinGW case the old probe worked around.
• resources.cfg.in: folded the duplicated [Graphics] section into one and dropped the six hardcoded @CMAKE_INSTALL_PREFIX@/share/OGRE/Media/RTShaderLib* paths, now redundant with @RTSHADER_DIR@.
• Trailing newlines on the two new .glsl files.

Not touched

Left alone deliberately, and raised as questions on tomluchowski#15 instead: the ephemeral-port fallback's effect on LAN hosting, the x11_mouse_hide flip, the SettingsWindow scaling approach, the vendored RTShaderLib GLSL, and the 194 re-serialised meshes.

Verification

Not built or run. This machine has Ogre 1.12 and tomluchowski#15 needs 14.6 plus tomluchowski/cegui#1, so none of this is compile-verified — please build it before merging. The changes are deliberately narrow and local for that reason.

🤖 Generated with Claude Code