What downloadable homebrew can and cannot do for us
Scanned GitHub for open-source homebrew to use as test material, ahead of building the corpus tooling. Two findings, both measured, both narrowing the plan rather than widening it (D163).
Homebrew pkgs are encrypted. Two of them, different
authors, different packers, both with bit 2 of the PFS superblock mode
set - the container declaring itself encrypted rather than me inferring
it from entropy at 7.95. No plaintext eboot.bin,
sce_sys or param.sfo anywhere inside. So a pkg
cannot yield a runnable executable here, ever: that needs a key, and
keys are outside this repository permanently.
What a pkg can give is its outer container, which is
entirely plaintext - content id, param.sfo, icons. Enough
for a real library row, which is more than the two pkgs sitting in the
local library get today. Support deferred, design recorded so it need
not be re-derived.
Payload ELFs exercise almost nothing.
ps5-payload-dev publishes loose .elf payloads
for both generations, GPL-3.0, stable tags, and even the same program
built for both targets - which looked like a free differential on the
API surface. They parse cleanly as bare ET_DYN and then
report no PT_DYNAMIC segment: payloads resolve everything
through the loader's own function table at run time, so there is no
import list at all. They would load and be invisible to every layer
above the parser.
So phase 0d is not "download homebrew to run". The value is in building a guest with the open toolchain, which produces a real import table - and that is precisely what obSCEne already is. The manifest matters for pinning what we build and what obSCEne was built from, not for harvesting other people's binaries.
Emulator projects came up prominently in the search - shadPS4, GPCS4, KytyPS5, obliteration
- and none of their source was opened. Principle 1.