|
|
Elvin
Admirable
Omnipresent Hero
Endless Revival
|
posted December 08, 2025 10:07 AM |
|
Edited by Elvin at 10:11, 08 Dec 2025.
|
Wow, it lives! Hey insa.
____________
For H5 and Olden Era, join the Heroes Tactical Duels discord server!
H5 Duel Map hosted on Moddb
|
|
dredknight
Honorable
Legendary Hero
disrupting the moding industry
|
|
posted December 08, 2025 11:18 AM |
|
|
|
FireDragon008
Tavern Dweller
|
|
posted April 28, 2026 12:48 PM |
|
|
Elvin said:
Wow, it lives! Hey insa.
Hi Elvin, Insa!I'm glad you're here.
Unfortunately the toheroes.com site is down. 
Where are the great toh5 players and map editors?
Jinxer, Jorko80, Diegis, Fiur, StriderHL, Kispagat, Buxyrs etc...Do you know anything about them?
____________
|
|
RackShaSe
Tavern Dweller
AI Darkwizard
|
|
posted May 01, 2026 06:14 PM |
|
|
Hi there! Did somebody try to RE H5_Game.exe with Claude Code & Ghidra? I've got such stats from AI
Got real numbers now: 51,872 functions — bigger than my earlier estimate. (And the vtable extraction broke — found only 1; my
symbol query was wrong; I'll fix that next.) But this gives me real data for the estimate.
---
Time & token estimate — full RE of H5_Game.exe
Scale baseline (now confirmed)
- 51,872 functions in .text (2× my earlier guess — the engine is huge)
- 2,729 classes (RTTI)
- Estimated ~250,000–500,000 lines of equivalent C++ source (rough rule: 5–10 LoC per decompiled function)
- For comparison: HoMM3 reverse (much smaller) — community has worked on it for ~15 years and still isn't "complete"
Time estimates
Per-function effort (no PDB, no symbols):
- Skilled human reverser: 5–20 functions/hour for naming + light commenting
- AI-assisted: 20–60 functions/hour for structural pass; harder functions (combat AI, RMG) drop to ~5–10/hr
- "Done well enough to commit to a research repo": ~30–80% of functions can be auto-named via heuristics (string xrefs, vtable
membership, call patterns); the remaining 20–70% need manual judgment
Phases & calendar time (1 lead engineer + AI):
Realistic bottom line:
- Modding-ready research repo (Phases 1–4 + targeted Phase 5 for whatever mod feature you want): ~3–6 months solo, or ~6–10 weeks
with 2–3 people.
- Full buildable source (Phase 6): multi-year project, requires a community.
Token estimates (Claude API)
Costs per phase (assuming Sonnet-tier with prompt caching; output ~30% of total):
For the mod-community-useful target (Phases 1–4 + 1–2 deep-dives): ~50–80M tokens total, roughly $300–500 in API costs, spread
across many sessions over several months.
The expensive part isn't the structural extraction (already done — cheap). The expensive parts are:
1. Decompiled function review — me reading FUN_xxxxxxxx, naming it, generating cleaned-up code. Each function is 500–5000 tokens
depending on size.
2. Iteration & validation — proving a hypothesis about a format requires multiple read/test cycles.
3. Cross-referencing & documentation — writing the prose that explains each subsystem to the next person.
Even if it's not possible in short terms to RE whole .exe, mb it will be possible to fix smth like common bugs, like Deep Freeze not working with lucky caster?
|
|
dredknight
Honorable
Legendary Hero
disrupting the moding industry
|
|
posted May 05, 2026 08:05 AM |
|
|
Hello RackShaSe,
This is awesome information! I have quite a deep knowledge about the exe. The game EXE is about 80000 functions big, but this includes stub functions that are never called.
Almost all the game logic is a known thing, Deep freeze bug is fixable but on the other hand it is good to stay as is for its unique effect and make the spell stand out, otherwise it is just another damage dealing spell.
Btw can you DM me on discord ? I have an idea that can be feasable in much less than 1-2 years time.
Cheers
____________
Join our official discord channel
|
| |
|
|
This Popular Thread is pages long: 
