[bensaadmucret]

What and why

A game that links the EOS client into main while its EOS module NRO is refused as an online SDK (Online SDKs in NROs are never loaded) crashed on a null EOS_HPlatform: EOS_Platform_Create was a generic stub that returned 0 and never wrote the output handle, and each EOS_Platform_Get*Interface stub returned null too, which the game dereferenced and then called through (sig=11 at boot, then pc=0). Hades survives the refused module because its EOS calls live inside it; MARVEL Cosmic Invasion (FNA, SDL_GPU, title 010059D020C26000) imports EOS_Platform_Create directly.

EOS_Platform_Create now writes the options pointer itself — a mapped, non-null guest address the game treats as an opaque EOS_HPlatform — and a stub whose name holds EOS_Platform_Get and Interface answers the platform handle it was called with, so the offline path continues. The handle is never a real EOS object and the host never serves Epic services; the game ticks it and plays offline.

Adds the finding to docs/reference/services.md (EOS_Platform_Create called on the refused module's client in main).

Checked

• Tests: python3 -m unittest dm.tools.tests.test_docs (6 tests, links and symptom index)
• Games run, with title ID and version: MARVEL Cosmic Invasion 010059D020C26000 v0 (build D6A37DEAFF3F5AEA, provisional profile): before the fix it aborted on the null platform dereference at boot; after, EOS initialization completes and the game reaches its frame loop — intro video at 60 fps with audio, ~870 draws/frame (with the sibling Vulkan PRs, see below)
• Devices and OS: Apple M6, macOS 27.0
• regress.py impact: the change reaches hle:* in theory, but only stubs named EOS_Platform_Get*Interface behave differently — every other unimplemented import returns 0 exactly as before, so no profiled game's path changes. regress.py check --games was not run: no local game build exercises this path, and the only caller pattern affected is the one this fix is for.
• This is one of four PRs from the same bring-up; the others (vk_shaders entry point, vkCmdBlitImage, profile_guess) are independent and listed in each other's bodies.

Docs

• The finding is in its game note or its docs/reference/<subsystem>.md entry, and python3 dm/tools/docs_symptoms.py was run (the symptom index)
• A chain tool changed: dmsh_chain.py --header was run and dm/gfx/nvn/dmsh_chain.h is part of the change
• No game files, keys or data extracted from a game in the change or the attachments