ArchPcode: Report no program counter for a language that declares none.

[zardus]

THIS MESSAGE WAS GENERATED BY AN AUTOMATED PROCESS

Dalvik and the DATA sleigh languages declare no program counter and have no register that could serve as one, so ArchPcode pointed ip_offset at 0x80000000, an offset no register occupied. registers["ip"] and register_names[ip_offset] raised KeyError, get_register_by_name("ip") returned None, and translate_register_name returned the decimal offset.

ip_offset is now left unset for those languages. Arch already declares it optional, and of the 187 languages pypcode ships only those 15 change: they lose the invented offset, and no other field of any language moves.

The regression covers both affected families, Dalvik:LE:32:DEX_Nougat and DATA:LE:64:default.

Raised in review of angr/angr#6804, which makes angr tolerate the missing register and carries the end-to-end regression.

Validation: #367 (comment)

[zardus]

THIS MESSAGE WAS GENERATED BY AN AUTOMATED PROCESS

Validation record for head 6196611c8426a922581319403f197215f6ca1216 against baseline da171ca0dd8eec1e16dd9c4f04576b2c958e2aea, with angr/angr#6804 at e0cea1f961fe2ccbef3b3afff844be4c6ab323bb against angr baseline 0c293dc0dea6a97efd583b7fe9e56912f0e8e2c1.

• Regression: python -m pytest tests/test_pcode.py -k program_counter — Dalvik:LE:32:DEX_Nougat and DATA:LE:64:default report ip_offset None and define neither ip nor pc; x86:LE:64:default still maps both names onto rip
• Focused (archinfo): python -m pytest tests — 19 passed
• Focused (angr): python -m pytest -n 3 tests/engines/pcode tests/storage tests/simos tests/state_plugins tests/factory tests/sim tests/test_load_shellcode.py — 384 passed, 4 skipped, 127 subtests passed
• Lint/type: merge-base pylint and pyright comparison over the changed files — no regressions. arch_pcode.py typecheck badness drops from 0.238 to 0.0; the type-checking import resolves the four pre-existing "pypcode is possibly unbound" errors, which a net line removal would otherwise have turned into a failure
• Hooks: pre-commit run --all-files — passed, tree unchanged
• Blast radius: building every one of the 187 languages pypcode 4.0.1.dev0 ships, before and after, leaves exactly 15 differing — 13 Dalvik variants plus DATA:LE:64:default and DATA:BE:64:default — and the only field that moves is ip_offset, 0x80000000 to None. Register lists, sp_offset, bp_offset, ret_offset, lr_offset and initial_sp are identical everywhere
• Stack pointer: the sibling sp_offset fallback is left alone. No language pypcode ships reaches it, so it is unused rather than wrong
• Downstream: with Tolerate an architecture that has no program counter. angr#6804, blank_state, entry_state and CFGFast all work on Dalvik:LE:32:DEX_Nougat; state.addr and state.regs.ip raise, which is the truthful answer for an architecture with no program counter
• Test inputs: check-test-inputs.py — clean

Caveats: the complete local workspace gate was not rerun on this head — a corpus sweep holds the machine — so the local evidence is the focused runs above plus the 187-architecture comparison.

[angr-bot]

Corpus decompilation diffs can be found at angr/dec-snapshots@master...angr/archinfo_367