SCC: fix aggregate lvalues, PPC socketcalls, concat, and target selection

[plafosse]

Summary

This PR addresses the currently open Binary Ninja issues labeled Component: SCC, with implementation changes, documentation updates, and regression coverage. The changes are split into issue-focused commits, followed by one shared test-suite commit.

Issues fixed

• Fixes scc unions in structs binaryninja-api#563 — Nested unions and aggregate lvalues. SCC now recursively computes addresses through nested structs, unions, pointers, dereferences, and array indexes. This fixes reads, writes, address-taking, and pre/post increment or decrement while evaluating side-effecting bases and indexes only once.

• Fixes Powerpc socketcall is not setup properly binaryninja-api#628 — PowerPC socket calls. Linux/PPC networking wrappers now use the legacy socketcall ABI, placing syscall 102 in r0, the operation in r3, and a pointer to packed arguments in r4. Regression tests validate all 16 wrappers on both ppc and ppcel.

• Fixes SCC Documentation Fixes binaryninja-api#1422 — SCC help and documentation. Unsupported options and the unimplemented __initial_<reg> helper are no longer advertised, while the stale claim that byte blacklisting is unimplemented has been removed. The Python wrapper and docs now use --markov-chain-file correctly and describe the supported architectures and retained compatibility parameters accurately.

• Fixes SCC: jump to end of buffer option emits syscall(exit) binaryninja-api#1423 — Concatenation control flow. --concat now transfers control to the byte after the generated output instead of calling exit, including when global initialization creates branching control flow. Exact-length and aligned padding now participate in linker layout until relocations and __end converge on the real output boundary.

• Fixes SCC: Predefined API on Windows x86_64 target broken binaryninja-api#2181 — Windows x64 predefined APIs. The MessageBox example now uses SCC's runtime-provided declaration instead of introducing a conflicting manual import declaration. Regression coverage verifies both direct use of predefined Windows APIs and compatible redeclarations on the x64 Windows target.

• Closes invalid handling address math in long mode binaryninja-api#2391 — Full-width x64 address calculations. The current backend already emits correct 64-bit frame setup and address calculations, so no implementation change was required. Raw-binary and ELF PIE regressions now require the appropriate REX.W encodings and verify RIP-relative global targets, preventing the reported truncation from returning.

• Fixes scc defaults to 32-bit shellcode without warning, causing unexpected behavior binaryninja-api#6577 — Implicit target architecture. Normal compilation now warns when it defaults to 32-bit x86, while --exec selects and reports the native x86/x64 target or fails clearly on unsupported hosts. Invalid explicit architectures fail immediately, and late --arch, -m32, or -m64 options retain precedence regardless of argument order.

Tests

• Added CTest suites for CLI behavior, x64 regressions, and PowerPC socket calls.
• Coverage includes concat control flow and padding, help and Python-wrapper arguments, architecture diagnostics, nested aggregate lvalues, Windows imports, raw and ELF PIE x64 addressing, and all 16 socket wrappers on ppc and ppcel.
• ctest --test-dir <build-dir> --output-on-failure — 3/3 tests passed.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}