[xiaoyueyoqwq]

Summary

Fixes #84 — MultiAddonManager fails to load on CS2 1.41.9.7 / build 2000943 (Failed to lookup the signature for ReplyConnection), and would crash on the first client message even once it loaded.

• Linux ReplyConnection signature extended. On 2000943 the compiler swapped the function's first two movs (55 41 B8 … B9 … 48 89 E5 41 57 41 56 48 8D 95); the shorter existing pattern now matches two functions so lookup fails.
• Hook_SendNetMessage no longer touches the message handle. 14197 turned INetworkMessageInternal handles into plain NetMessageInfo_t records with no vtable, so GetNetMessageInfo() was a virtual call through a name string. CNetMessage's own vtable is unchanged, so pData->GetMessageId() is used directly.
• GetAddonSignonStateMessage works around the SDK's stale declarations. INetworkMessages::FindNetworkMessagePartial moved from slot 14 to slot 10 and the returned record is allocated via its m_pfnAllocate field at +0x8. The helper reaches both through the raw vtable/record until the SDK headers catch up (s2sdk#124 pending upstream).

Analysis largely from @jakejarrett's notes in #84, verified against s2sdk#124.

Test plan

• Linux build compiles (gcc, SteamRT sniper container, s2sdk@cs2 headers)
• Plugin loads on live 14197 server: meta list shows MultiAddonManager (v1.7.3-7fc2d6f) loaded, [MultiAddonManager] Plugin loaded successfully!, Refreshing addons, Logged on to Steam
• Hook_SendNetMessage path exercised continuously (SourceTV/HLTV client receiving messages), no crash
• Addon push path verified end-to-end on live server: host_workshop_map 3789922008 → HostStateRequest ... addons(3789922008) → Mounting addon → SignonState messages sent → real client connected with S2C_CONNECTION [addons:'3789922008'] and validated — no crash
• Windows build untested (no build env here)

Generated with Devin

[xiaoyueyoqwq]

Live verification results on CS2 1.41.9.7 / build 2000943 (Linux dedicated, Metamod 2.0 git1483, SteamRT sniper runtime).

End-to-end addon flow — all green

Switched to workshop map 3789922008 with a real client connected:

profile via host_workshop_map 3789922008
HostStateRequest::Start(HSR_GAME): loop(levelload) id(3) addons(3789922008)
Mounting addon '3789922008'
SV:  addon='3789922008'
Sending S2C_CONNECTION to <client> [addons:'3789922008']
"AI<65280><[U:1:753702285]><>" STEAM USERID validated

• ReplyConnection hook fires on connect and pushes the addon list correctly (new signature resolves)
• GetAddonSignonStateMessage produces valid CNETMsg_SignonState [7] messages via the slot-10 vcall + m_pfnAllocate (signon went through, client validated)
• Hook_SendNetMessage handles all outbound traffic incl. SourceTV/HLTV — ~15 min continuous operation, zero crashes

Signature uniqueness check on the actual binary

Scanned the server's libengine2.so (build 2000943):

55 41 B8 ? ? ? ? B9 ? ? ? ? 48 89 E5                     → 2 matches (0x4f9d80, 0x877d00)
55 41 B8 ? ? ? ? B9 ? ? ? ? 48 89 E5 41 57               → 1 match  (0x4f9d80)
55 41 B8 ? ? ? ? B9 ? ? ? ? 48 89 E5 41 57 41 56 48 8D 95 → 1 match  (0x4f9d80)

Worth noting: the published v1.7.3 release binary was built at 3e04eff, i.e. before 7fc2d6f — it still carries the pre-14197 55 B9 … 41 B8 signature, which matches 0 functions on this build. That's the lookup failure servers are hitting (#84). Current main's signature does resolve uniquely here; this PR extends it anyway since the shared prefix alone already collides at 0x877d00 — one prologue shuffle away from an ambiguous match.

The runtime side is the load-critical half: with only the signature fixed, the plugin loads but crashes on the first client message (GetNetMessageInfo() virtual call on the vtable-less handle — same fault class as roflmuffin/CounterStrikeSharp#1491).

[xiaoyueyoqwq]

Understood on the close — for anyone landing here: rebuilding MAM against s2sdk 81020d1 (alliedmodders/s2sdk#124) together with the 7fc2d6f signature resolves the crash with no source changes, as confirmed in #84. That is the cleaner long-term path since the ABI details live in the SDK headers.

This branch still works as a stopgap — the build above was verified end-to-end on 2000943 (workshop map switch, addon push, real client join, zero crashes) — for anyone who needs a working binary before an official rebuild ships.

One small request for future contributions: a one-line rationale when closing a PR helps contributors understand the direction. Happy to help test the official build when it lands.