Today was a restart day. A while back, a broad, automated GUID-relinking pass (meant to fix a couple of stray “missing script” errors) went wrong and corrupted the project — likely from its filename-fallback matching mismatching generic-named assets across folders. Rather than fight the corruption, I restored from backup and rebuilt from a fresh AssetRipper export, this time with the lessons from that mistake already in hand.

The good news: since every fix from the first pass was documented, re-deriving them was fast. Here’s what today’s progress covered, roughly in the order it came up.

Compile-time cleanup

A few small but blocking issues on the way to a clean build:

  • Duplicate AssemblyInfo.cs — two copies existed (one under Assembly-CSharp, one under legacy Assembly-UnityScript boilerplate), which C# doesn’t allow. Deleted the redundant one.
  • Decompiler false alarms — ILSpy occasionally reconstructs valid logic in a shape the C# compiler’s definite-assignment checker can’t quite follow, throwing CS0165: Use of unassigned local variable even though the code always assigns the variable before it’s read. Hit this twice — once in a low-level ByteReader.cs line-parsing loop, once in a DialoguerEditor theme-naming utility with a gnarly (array2 = array)[num = i] = ... compound-assignment pattern. Fix in both cases: just give the variable a default value at declaration (= 0, = null). Doesn’t change behavior, just satisfies the compiler.

Runtime: the Android-only landmines, again

Getting Play mode working meant re-clearing the same category of issue as the first pass: code that assumes it’s running on an actual Android device.

  • The permission-check plugin (net.sanukin.OverrideUnityActivity, wired through M00_Init.cs) crashes hard on Windows since there’s no native Android JNI bridge underneath it. Guarded both Setup() and CheckPermissionProcess() (the latter gets called from two different places — the boot sequence and OnApplicationPause — so the guard has to live inside the method itself, not at each call site) with Application.platform == RuntimePlatform.Android checks.
  • The Adjust SDK (mobile analytics, whose servers don’t exist to report to anyway) was the bigger project today. Instead of guarding each of its ~20 public methods one at a time as they surfaced as separate crashes — which is how it went last time — I went through the whole file and guarded every method in one pass. Then, since every guarded method’s entire remaining body is now permanently unreachable dead code, I went a step further and rebuilt the file as a minimal stub: same public API surface, zero actual native-call logic. 550 lines became 61. Nothing else in the codebase needed to change, since every method signature stayed identical — it just does nothing now, correctly, instead of doing nothing after crashing first.

The UIRoot race condition

Hit a List<UIRoot> index-out-of-range error the moment the WorldMap scene loads — UIManager’s World class was reading UIRoot.list[0] in a field initializer, which runs the instant the object gets constructed, with no guarantee UIRoot had registered itself into that static list yet. Converted it to a lazy property instead, deferring the lookup until something actually needs it. By that point in the game’s lifecycle, UIRoot is definitely there.

Shaders, again

The recurring theme of this whole project: AssetRipper doesn’t reliably extract shader assets from a compiled build. Every fresh export needs the same fix — pull NGUI’s Shaders folder from the actual GitHub source and drop it into Assets/Resources/Shaders. Skip this step and you get both broken-looking UI and invisible text, since NGUI’s rendering and text drawing both depend on it. Did it again today; both symptoms cleared in one move, same as last time.

One shader-related NullReferenceException turned out to be a red herring — it only fires from the Editor’s Scene view trying to render NGUI’s [ExecuteInEditMode] panels, not from the actual Game window. Confirmed gameplay itself is unaffected and parked it rather than chasing an editor-only cosmetic quirk.

Where things stand

Play mode is clean again — booting, UI rendering, no crashes — at roughly the same point the first attempt reached before the GUID mishap. The difference this time: every fix is captured, the mistake that caused the corruption is documented and avoided, and the whole restart took a fraction of the time the original debugging did, since there was a real playbook to follow instead of discovering everything from scratch.

Next up: getting back to an actual Windows Build, and — this time — being much more careful about how any future GUID relinking gets done.