D222 - What a bundled library must carry, found one refusal at a time

Four, each named exactly by the loader, in the order they appeared:

• Header placement. verify_ehdr: offset 0x0 end 0x190 - the first segment started underneath the program header table. That is selfish D078 and a third linker script.
• EI_ABIVERSION. ### ERROR: ABIVERSION mismatch ... val 2. A bundled library and the eboot that loads it must agree, and sce-module was taking the defaults while eboot-min was not.
• DT_SCE_ORIGINAL_FILENAME. preprocess_dt_entries:9600: C: orig fn 0 mod info 1 - one module-info tag, zero filename tags. selfish D079.
• The procedure-linkage tables. preprocess_dt_entries:9632: C: ah 1 pg 1 jr 0 pr 1 prs 0 rl 1. The stub referenced sceKernelWrite by address, which is a GLOB_DAT relocation and not a JUMP_SLOT, so DT_SCE_JMPREL and DT_SCE_PLTRELSZ were absent. This loader requires the full set whether or not a module has anything to relocate, which D218 had already found for the eboot and which generalises. The stub now has a call behind a condition the system will never satisfy - argc == ~0UL - purely so the tables exist.

Every one of these was a single build and a single launch, because the loader named the field. That is worth saying plainly after D075, where an error naming a structure sent the work in the wrong direction for a long stretch: this loader is usually precise, and the two red herrings so far (Failed to load SCE_DYNLIBDATA, Unsupported ELF e_type on a .prx) were both cases where the message came from a layer above the one that failed.