Alan-Jowett / ebpf-verifier / 22075299923

88%

Build Type

push

github

Committed by web-flow

Commit Message

Replace Number backend: boost's cpp_int → __int128 (#1007)

Backport the i128 representation from the Rust port. Number is an
intermediate representation for eBPF verification where all inputs
and outputs are ≤64-bit; i128 provides headroom for widening
arithmetic without the overhead of arbitrary-precision integers.

This commit has been verified to produce the same invariants as its parent on the test data.

Key changes to src/arith/num_big.hpp:
- Use __int128 on GCC/Clang, boost::multiprecision::int128_t on MSVC
- All arithmetic operators use checked helpers (__builtin_*_overflow
  on GCC/Clang, manual pre-checks on MSVC) to prevent signed overflow UB
- Simplify sign_extend/zero_extend: direct bit arithmetic replaces
  template dispatch over fixed widths
- fill_ones computed in unsigned space to avoid shift UB
- Custom decimal parser accumulates in UInt128 to handle kInt128Min
- Width assertions tightened to eBPF domain bounds (≤64/65)
 Consequential simplifications:
- Weight = Number (was already the case), so convert_NtoW_overflow
  was a passthrough returning false — inlined at all ~18 call sites
  in zone_domain.cpp, removing dead overflow branches
- eval_expression_overflow → eval_expression, returns Weight directly
- Remove dead SafeI64 overload and num_safeint.hpp include
- Custom int128_to_string in printing.cpp (no std::to_string for __int128)
- Add missing transitive includes (<algorithm>, <cassert>) previously
  provided by boost headers

Also: remove unused stats infrastructure, rename_classes.py, and
miscellaneous cleanup (docs ASCII arrows, splitdbm simplifications).

Signed-off-by: Elazar Gershuni <elazarg@gmail.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Run Details

139 of 166 new or added lines in 5 files covered. (83.73%)

52 existing lines in 4 files now uncovered.

9389 of 10777 relevant lines covered (87.12%)

3125310.38 hits per line