tools/loading_tip_patcher/RE_notes.md

RE notes — EU4 1.37.5 loading tip / screen cycle mechanism

Static-analysis findings on eu4.exe 1.37.5 (build 2024-10-03, SHA-256 9ad3efe1af169f40ee577f9dae5debbc87af6fb8b5450fb345ebf110dc4d771a). The goal of this file is to give the next maintainer enough breadcrumbs to skip the dead-ends and jump straight to the cycle function once Ghidra analysis finishes.

Target binary

C:\Program Files (x86)\Steam\steamapps\common\Europa Universalis IV\eu4.exe

Property	Value
Size	38,462,504 bytes (36.7 MiB)
SHA-256	9ad3efe1af169f40ee577f9dae5debbc87af6fb8b5450fb345ebf110dc4d771a
Version	release_1.37.5 (rev 835bfdf8ca24c291a1b3f1b5bc72d47e7df1ae18)
PE timestamp	2024-10-03 17:58:17
Format	PE32+ x86-64, 7 sections, no UPX
Linker	MSVC 14.29 (VS 2019 16.9–16.10)
ImageBase	0x140000000
PDB	none shipped (path string eu4\eu4\outputdir_windows\eu4.pdb only)
.text VMA / file	0x140001000 / 0x00000400 size 0x1B64D78
.rdata VMA / file	0x141B66000 / 0x01B65200 size 0x44999A
.data VMA / file	0x141FB0000 / 0x01FAEC00 size 0x384400

Loading-screen string table

A single string block in .rdata (file offset 0x1D62B4B – 0x1D62B9C) contains every name the loading-screen code looks up by string:

File offset	VMA	Bytes / purpose
0x1D62B4B	0x141D62B4B	"load_screen2\0" — sprite name bg_load_screen2 (status bg)
0x1D62B6C	0x141D62B6C	"tip\0" — the loading-tip windowType
0x1D62B70	0x141D62B70	"load_screen\0" — the root load_screen windowType
0x1D62B80	0x141D62B80	"LOADING_TIP_\0" — localization key namespace (no xref; built dynamically via format + integer)
0x1D62B88	0x141D62B88	"status\0" — the loading-status windowType

None of loading_tip, LOADING_TIP, loading_screen, LoadingScreen, getLocalization, getLocalisation, resolveKey, Timer, tick, interval, next_tip, setTextFromKey, set_loading_tip appear as strings anywhere in the binary. The localization keys are built with format("LOADING_TIP_{}", i) at runtime, which is why grep "LOADING_TIP_" finds only the namespace literal and zero references.

Loading-screen construction (sub_141174F50)

Entry: 0x141174F50 (image-base + offset, file offset = VMA − 0x140000000 = 0x1D74F50).

Called from two places, both inside the same TU:

• 0x1411749E9 — first call site
• 0x141174EE2 — second call site

The function:

1. Allocates a 32-byte string buffer (call 0x1400950B0 is operator new).
2. Copies a 16-byte structure from .rdata into it, then a 2-byte and 1-byte tail.
3. Calls vtable[0x120](rdi, &local) (string-keyed registration or addChildByName).
4. After allocation, sets [rbx+0x100] = 1 (or 0) based on [rip+0x12b6d79] + 0x28b — almost certainly an is_loading_synchronous / is_replay flag.
5. Calls lea rdx, [rip+0xbedadd] → "load_screen" and call [rax+0x98] (a findChildByName virtual).
6. The branch at 0x14117509C:

   cmp [rbp+0x28], 1
   je  0x1411750CA            ; ironman/replay branch
   cmp [rip+0x11caea6], 0     ; probably online check
   jne 0x1411750CA
   mov rdx, [rbx]
   mov rcx, rbx
   call [rdx+0x88]            ; vtable[0x88]
   

   The vtable[0x88] and vtable[0x90] are the two setVisible / setActive overloads, depending on [rbp+0x28] == 1.

This is the construction / activation path, not the cycling path.

Status-text updater (sub_141174840)

Entry: 0x141174840.

1. call 0x141172B20 — likely the LoadingScreen::onEnter initializer.
2. lea rdx, [rip+0xbee32b] → "status"; call [rax+0xB8] = findChildByName("status").
3. The found child at [rax+0xC8] is the text instantTextBox inside status.
4. Then the function appends a fixed sequence of characters to the textbox:

   0x141174893:  mov dl, 0x5C   '\'
   0x14117489D:  mov dl, 0x2F   '/'
   0x1411748A7:  mov dl, 0x3A   ':'
   0x1411748B1:  mov dl, 0x2A   '*'
   0x1411748BB:  mov dl, 0x3F   '?'
   0x1411748C5:  mov dl, 0x3C   '<'
   0x1411748CF:  mov dl, 0x3E   '>'
   0x1411748D9:  mov dl, 0x7C   '|'
   

   These are the classic ASCII spinner glyphs. The function re-renders them in sequence. This is the status progress indicator, not the tip cycler.
5. Then lea rdx, [rip+0xbee2a1] → 0x141D62B94 (some other name, the bytes are immediately after "status\0" in the string table) and call [rax+0x68]. The function then tail-calls jmp [rax+0x8] to a setter.

The status updater is driven by an outer loop — search for callers of 0x141174840 to find the per-frame tick.

Tip cycle function (NOT YET LOCATED)

What we know it must do:

1. Find the "tip" windowType child of the load_screen GUI (findChildByName("tip") or via the child at offset 0xC8).
2. Increment an internal Timer member every frame.
3. When elapsed >= interval, pick a random i in [0, 67), build the key "LOADING_TIP_" + i, and call setText(localizedKey) on the tip textbox.
4. Reset the timer to 0.

The "tip\0" C-string exists at VMA 0x141D62B6C but has zero xrefs in the disassembly as a literal. The function therefore either:

• Uses a hardcoded offset / index instead of a string lookup (e.g. getNthChild(1) of the load_screen root), or
• Builds the "tip" name dynamically via format or string concat.

Next step (with Ghidra decompiler): decompile the call graph of sub_141174F50 (the load_screen setup) and look for the per-frame update path. The cycling function is somewhere in that graph.

Timer constants table

At VMA 0x141DC1750 (file offset 0x1DC0950) is a 16-entry table of double constants used by the engine for UI timers:

VMA	Value	Likely use
0x141DC1750	4.0	fast spinner tick
0x141DC1758	5.0
0x141DC1760	8.0000019	(FP rounding artifact, paired with next)
0x141DC1768	8.0	prime suspect for tip cycle interval
0x141DC1770	9.6	(mid-game tick?)
0x141DC1778	10.0	slow spinner / pause
0x141DC1780	(garbage)	(denormal / uninitialised)
0x141DC1788	15.0	tooltip dwell
0x141DC1790	20.0
0x141DC1798	30.0
0x141DC17A0	40.0
0x141DC17A8	(garbage)
0x141DC17B0	100.0
0x141DC17B8	180.0
0x141DC17C0	200.0

The 8.0 value at 0x141DC1768 is referenced from 11 disassembly sites. Most are not the tip cycler (verified: 0x14081D83A is part of a 0x14081D... per-frame update, 0x14182CC75 is a graphics normalize function). The true cycle call site is one of the remaining xrefs and will be identified by the call graph in the Ghidra decompile.

Patch candidates (most likely first):

• 0x1DC0968 (VMA 0x141DC1768) — the 8.0 literal itself. Patch: overwrite 8 bytes with the user's tip_interval_seconds as a double. 8-byte write.
• Alternatively, if the cycle function takes the value via movsd xmm1, [rip+offset] to the constant, we patch the movsd to load from a writable .data cell the patcher initializes at runtime.
• A third option: NOP the jbe / jge after the elapsed-time comparison in the cycle function to slow the cycle to "never".

What remains to do (Ghidra pass)

1. Open eu4.exe in C:\tools\ghidra_12.1.2_PUBLIC\ghidraRun.bat.
2. Search → For Strings → "load_screen", "tip", "status", "LOADING_TIP_".
3. For each, jump to the xref and decompile the containing function.
4. The tip cycler will be a function that:
   • Reads [rdi+0x30] (vtable), looks up the "tip" child via vtable[0xB8] (or a sibling lookup).
   • Has a Timer member (look for add, sub, cmp with 8.0 and a jb/ja that conditionally calls setText).
5. Record the VMA + file offset of the comparison constant, and identify the byte pattern that uniquely matches across 1.37.5.x builds. The patcher will use this pattern for --rescan after game updates.

Ironman / achievement bypass (Phase 2 task, partially found)

The MD5-based integrity check is at VMA range 0x1410F78B0 – 0x1410F8666, with the hash init at 0x1410F80B8 (mov DWORD PTR [rbp+0xb0], 0x67452301 — the standard MD5 init constant). The function:

1. Sets up the timer / state at 0x1410F8084 (movsd xmm2, [rip+…] → 8.0, passed as the third arg of a virtual call at 0x1410F808E: call QWORD PTR [rax+0x10]).
2. Computes the MD5 hash of the game state at 0x1410F84B0: call 0x1410DD830.
3. Uploads / records the hash via 0x1410F8573: call 0x1416554B0 (with at least 6 string / data arguments, the most likely "write to save header + sync to Paradox" call site).
4. Sets a global flag at 0x1410F85FA (mov DWORD PTR [rip+0x1247877 + 0x747c], 1) — almost certainly is_saving_ironman = true.
5. Continues into the "save complete" path (call to 0x1410F864E: call r8 for vtable[0x10], which is the onSaveComplete callback).

Ironman patch candidates (to verify in next pass)

VMA	Action
0x1410F84B0	5-byte call 0x1410DD830 (MD5 compute). NOP it.
0x1410F8573	5-byte call 0x1416554B0 (hash upload). NOP it.
0x1410F85FA	10-byte mov DWORD PTR [...+0x747c], 1. Patch to mov [...+0x747c], 0 to defeat "this is ironman" state.
0x1410F80B8–0x1410F80C7	16-byte MD5 init constants. Wipe with NOPs.

Any one of these patches alone is enough to disable a single link in the chain; for full ironman-bypass, do all three. The --revert path must restore the original bytes.

Hash-upload destination string

The 6 string arguments to 0x1416554B0 at 0x1410F8500–0x1410F8573 include:

• 0x141D5C438 — likely a Paradox account ID or session token (used at 0x1410F84F3: lea rdx, [rip+0xc63f3e]).
• A save-header key at 0x141D228B8 (used at 0x1410F799F).
• A field at 0x141C4238C (used at 0x1410F86E0 — to be confirmed).

These are good search anchors for the Ghidra xref pass.