Synthesis: ELF section header table drops unnamed sections (stripped ELF re-loads with base 0, loses functions)

[r0ny123]

Summary

ElfSynthesizer._assembleFile builds the section header table from
named_sections = [s for s in sections if s["name"]] (ElfSynthesizer.py:~329).

A stripped ELF — where SMDA recovers real VA ranges but every code_sections entry has name == "" — therefore synthesizes to a file with e_shnum == 2 (the null section plus .shstrtab) even though the real ranges are still emitted as PT_LOADs. Re-loading that file gives base_addr = 0 and loses functions.

Reproduction

tests/mirai_x64_xored on master (4.4.1), with the section names cleared to model the stripped case:

report.code_sections = [("", s, e) for _, s, e in report.code_sections]
synth = report.synthesizeBinary(output_format="ELF", with_strings=False)

all-unnamed code_sections -> e_shnum = 2 | names: ['', '.shstrtab']
  PT_LOADs emitted: 4
re-loaded: 145 functions (original 151), base 0x0 (original 0x400000)

Four PT_LOADs describing the real layout, but no section headers describing them; 6 of 151 functions are lost on re-disassembly and the base address is wrong.

(For contrast, the unmodified fixture has 9 of 10 sections named and synthesizes with e_shnum == 10.)

Suggested direction

Fabricating names for unnamed ranges (.smda0, .smda1, …) would preserve the section header table and keep the synthesized file self-describing. That also removes the base_addr = 0 case, which is one of the two paths into the wrong-base problem filed separately.

---

Found during a whole-repo audit of src/smda/** and re-verified against master before filing. Synthesis is documented as experimental, so filing rather than patching.