Ancient Borland C++ 2.0 compiler source code fully recovered byte-for-byte from its binary

Developers have reverse-engineered the entire source code of Borland C++ 2.0, a compiler released back in 1991, directly from its compiled executable binaries using Claude Code tooling and Exhume. The resulting codebase consists of nearly 300 source files split cleanly between C and hand-written assembly modules that compile into a BCC.EXE file matching the original SHA-256 hash down to the very last byte.

Far from being a sterile clean-room rewrite, the recovered files expose the delightfully cursed realities of early nineties software engineering. The parser relies on manual recursive descent instead of auto-generators like yacc, and the compiler casually keeps dormant self-referential debugging stubs alongside mysterious dead code pools. One particularly fragile string function reads the memory byte right before an empty string, while precompiled header macros accidentally preserve ancient heap residue because the compiler pads definitions to even sizes without clearing the trailing space.

Building the project requires digging up original distribution archives from retro software repositories, feeding them into a DOSBox-X emulator, and running automated verification gates that compile every single module and check relocations. Rather than trying to modernize the ancient codebase out of existence, the preservation effort strictly honors the original quirks, preserving duplicate functions and awkward macro logic right down to the compiler's signature error-handling bugs.

Watching ancient proprietary software get dragged into the open-source light via automated de-compilation feels less like traditional software maintenance and more like an archaeological excavation of digital amber. Future historians will likely look at this exact moment as the point where software development stopped forgetting its own past and started brute-forcing its way back to the beginning.

Source: GitHub