477. Two readers, one corpus

2026-09-09 - directed, continuing 476

REQ-20260909T1720Z-4e17 - the exception context PPSA25872's handler dereferences - is still open, so the signal wall is waiting on hardware. This pass paid the debt owed to SELFish since 474: the two-reader differential, claimed three passes ago.

The harness was wrong twice before either reader was

First run: all 29 modules disagreeing, not an ELF: begins 54 14 f5 ee. Orbistoun's Container::parse unwraps the signed container on the way in; selfish_elf::Elf::parse expects the ELF itself. The two were reading different files. Second run: still seven, because Wrapper::is_wrapped misses the previous-generation magic and is_either_generation is the one to ask.

Neither is a parsing defect. Both are facts about the two crates' entry points that anyone wiring them together would first read as parsing bugs (D653).

Three classes, once it was measuring the readers

• 22 - SELFish refuses the inner ELF: "the program header table runs past the end of the file". Every sce_module/*.prx. The inner ELF of a signed container is a view whose segments live in the container's own list, so the table legitimately points past the slice. Orbistoun tolerates it. Which is right is unsettled, and orbistoun's tolerance has never been tested against a deliberately truncated file - so this may be a defect here.
• 6 - orbistoun finds a vendor dynamic table where SELFish finds none. The defect shape their request already described, on six real eboots.
• 1 - obscene/eboot.bin, six fields, every address differing by exactly 0x6bc000.

The constant is the finding

A rebasing difference, not a parse difference - D247's own distinction between a standard tag holding a virtual address and a vendor tag holding an offset into the vendor data segment. The differential prints the file's tags beside the disagreement: it carries both standard (1-12, 20, 23) and vendor (0x61000007 … 0x61000049), which is what Table::Current describes. So SELFish's reading may be the right one and orbistoun's vendor_tables = false the wrong one - taken away to investigate.

Surprises

• The differential measured the harness until it didn't. Two rounds of all-29 failures, both mine. Same rule as a guard nobody has made fail: a comparison is not comparing anything until somebody has checked what it is comparing.
• Printing the file's own tags changed what the report is worth. Without them it says "two readers differ"; with them it says which one the bytes support. Four extra lines.
• The most likely defect the differential found is in this repository, not the sibling one.

Next

• vendor_tables = false on a file carrying vendor tags - orbistoun's side of (4).
• Whether orbistoun's tolerance of a program header table past the end of the slice is right, or is a bound it should be checking. Untested either way.
• The exception context at +0xf8, still open.